@trigger.dev/sdk 0.0.0-prerelease-20260909071038 → 0.0.0-prerelease-streamfix-20260909094302

package/dist/commonjs/v3/ai.js

@@ -1,7 +1,10 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.chat = exports.__buildManagedStreamTextOptionsForTests = exports.upsertIncomingMessage = exports.PENDING_MESSAGE_INJECTED_TYPE = exports.ai = exports.snapshotTranscriptStorage = exports.reduceTranscriptChanges = exports.memoryTranscriptStorage = exports.defaultStorage = exports.__writeChatSnapshotProductionPathForTests = exports.__setWriteChatSnapshotImplForTests = exports.__setReadChatSnapshotImplForTests = exports.__readChatSnapshotProductionPathForTests = void 0;
-exports.__setTranscriptStorageForTests = __setTranscriptStorageForTests;
+exports.chat = exports.upsertIncomingMessage = exports.PENDING_MESSAGE_INJECTED_TYPE = exports.ai = void 0;
+exports.__setReadChatSnapshotImplForTests = __setReadChatSnapshotImplForTests;
+exports.__setWriteChatSnapshotImplForTests = __setWriteChatSnapshotImplForTests;
+exports.__readChatSnapshotProductionPathForTests = __readChatSnapshotProductionPathForTests;
+exports.__writeChatSnapshotProductionPathForTests = __writeChatSnapshotProductionPathForTests;
 exports.__mergeByIdReplaceWinsForTests = __mergeByIdReplaceWinsForTests;
 exports.__setReplaySessionOutTailImplForTests = __setReplaySessionOutTailImplForTests;
 exports.__replaySessionOutTailProductionPathForTests = __replaySessionOutTailProductionPathForTests;
@@ -16,16 +19,6 @@ const v3_1 = require("@trigger.dev/core/v3");
 // ESM-only `ai@7` (see ../imports/ai-runtime.ts).
 const api_1 = require("@opentelemetry/api");
 const ai_runtime_js_1 = require("../imports/ai-runtime.js");
-const transcriptStorage_js_1 = require("./transcriptStorage.js");
-let transcriptStorageOverride;
-/**
- * Test-only override for the storage `chat.agent` persists through, so a
- * test can capture the exact changesets the runtime produces.
- * @internal
- */
-function __setTranscriptStorageForTests(storage) {
-    transcriptStorageOverride = storage;
-}
 const ai_shared_js_1 = require("./ai-shared.js");
 Object.defineProperty(exports, "PENDING_MESSAGE_INJECTED_TYPE", { enumerable: true, get: function () { return ai_shared_js_1.PENDING_MESSAGE_INJECTED_TYPE; } });
 Object.defineProperty(exports, "upsertIncomingMessage", { enumerable: true, get: function () { return ai_shared_js_1.upsertIncomingMessage; } });
@@ -42,8 +35,6 @@ const metadata_js_1 = require("./metadata.js");
 // pulled in transitively here never reach a client chunk.
 const agentSkillsRuntime_js_1 = require("./agentSkillsRuntime.js");
 const aiAutoTelemetry_js_1 = require("./aiAutoTelemetry.js");
-const externalDeploymentId_js_1 = require("./externalDeploymentId.js");
-const chatVersionSkew_js_1 = require("./chatVersionSkew.js");
 const sessions_js_1 = require("./sessions.js");
 const shared_js_1 = require("./shared.js");
 const streams_js_1 = require("./streams.js");
@@ -130,16 +121,182 @@ async function findLatestSessionInCursor(chatId) {
     }
     return latestCursor;
 }
-var chatSnapshotIo_js_1 = require("./chatSnapshotIo.js");
-Object.defineProperty(exports, "__readChatSnapshotProductionPathForTests", { enumerable: true, get: function () { return chatSnapshotIo_js_1.__readChatSnapshotProductionPathForTests; } });
-Object.defineProperty(exports, "__setReadChatSnapshotImplForTests", { enumerable: true, get: function () { return chatSnapshotIo_js_1.__setReadChatSnapshotImplForTests; } });
-Object.defineProperty(exports, "__setWriteChatSnapshotImplForTests", { enumerable: true, get: function () { return chatSnapshotIo_js_1.__setWriteChatSnapshotImplForTests; } });
-Object.defineProperty(exports, "__writeChatSnapshotProductionPathForTests", { enumerable: true, get: function () { return chatSnapshotIo_js_1.__writeChatSnapshotProductionPathForTests; } });
-var transcriptStorage_js_2 = require("./transcriptStorage.js");
-Object.defineProperty(exports, "defaultStorage", { enumerable: true, get: function () { return transcriptStorage_js_2.defaultStorage; } });
-Object.defineProperty(exports, "memoryTranscriptStorage", { enumerable: true, get: function () { return transcriptStorage_js_2.memoryTranscriptStorage; } });
-Object.defineProperty(exports, "reduceTranscriptChanges", { enumerable: true, get: function () { return transcriptStorage_js_2.reduceTranscriptChanges; } });
-Object.defineProperty(exports, "snapshotTranscriptStorage", { enumerable: true, get: function () { return transcriptStorage_js_2.snapshotTranscriptStorage; } });
+let readChatSnapshotImpl;
+function __setReadChatSnapshotImplForTests(impl) {
+    readChatSnapshotImpl = impl;
+}
+let writeChatSnapshotImpl;
+function __setWriteChatSnapshotImplForTests(impl) {
+    writeChatSnapshotImpl = impl;
+}
+/**
+ * Read the persisted snapshot for a session. Returns `undefined` on:
+ *   - missing object (404 from the presigned GET — fresh session, never
+ *     persisted)
+ *   - presign failure (network/auth issue)
+ *   - malformed JSON
+ *   - version mismatch (forward-compat — older runtimes ignore newer blobs)
+ *
+ * Always swallows errors via `logger.warn`. The agent boot loop must stay
+ * available even if S3 hiccups; the worst case is replaying more of
+ * `session.out` than strictly necessary.
+ * @internal
+ */
+async function readChatSnapshot(sessionId) {
+    if (readChatSnapshotImpl) {
+        return (await readChatSnapshotImpl(sessionId)) ?? undefined;
+    }
+    const apiClient = v3_1.apiClientManager.clientOrThrow();
+    let presignedUrl;
+    try {
+        const resp = await apiClient.getChatSnapshotUrl(sessionId);
+        presignedUrl = resp.presignedUrl;
+    }
+    catch (error) {
+        v3_1.logger.warn("chat.agent: snapshot presign (read) failed; continuing without snapshot", {
+            error: error instanceof Error ? error.message : String(error),
+            sessionId,
+        });
+        return undefined;
+    }
+    let response;
+    try {
+        response = await fetch(presignedUrl, { method: "GET" });
+    }
+    catch (error) {
+        v3_1.logger.warn("chat.agent: snapshot fetch failed; continuing without snapshot", {
+            error: error instanceof Error ? error.message : String(error),
+            sessionId,
+        });
+        return undefined;
+    }
+    if (response.status === 404) {
+        // First-ever boot for this session — no snapshot yet. Caller falls
+        // through to replay-only.
+        return undefined;
+    }
+    if (!response.ok) {
+        v3_1.logger.warn("chat.agent: snapshot fetch returned non-OK; continuing without snapshot", {
+            status: response.status,
+            sessionId,
+        });
+        return undefined;
+    }
+    let parsed;
+    try {
+        parsed = await response.json();
+    }
+    catch (error) {
+        v3_1.logger.warn("chat.agent: snapshot JSON parse failed; continuing without snapshot", {
+            error: error instanceof Error ? error.message : String(error),
+            sessionId,
+        });
+        return undefined;
+    }
+    if (!parsed || typeof parsed !== "object")
+        return undefined;
+    const candidate = parsed;
+    if (candidate.version !== 1 || !Array.isArray(candidate.messages)) {
+        v3_1.logger.warn("chat.agent: snapshot version/shape mismatch; ignoring", {
+            version: candidate.version,
+            sessionId,
+        });
+        return undefined;
+    }
+    return candidate;
+}
+/**
+ * Persist the snapshot for a session. Awaited by callers immediately after
+ * `onTurnComplete` — the agent may suspend right after this point, and
+ * fire-and-forget promises don't reliably complete on suspend.
+ *
+ * Errors are swallowed via `logger.warn`. A failed write means the next
+ * boot replays slightly more of `session.out` (back to the previous
+ * snapshot's cursor) instead of failing — the conversation stays
+ * coherent, only the boot path does marginally more work.
+ * @internal
+ */
+async function writeChatSnapshot(sessionId, snapshot) {
+    if (writeChatSnapshotImpl) {
+        await writeChatSnapshotImpl(sessionId, snapshot);
+        return;
+    }
+    const apiClient = v3_1.apiClientManager.clientOrThrow();
+    let presignedUrl;
+    try {
+        const resp = await apiClient.createChatSnapshotUploadUrl(sessionId);
+        presignedUrl = resp.presignedUrl;
+    }
+    catch (error) {
+        v3_1.logger.warn("chat.agent: snapshot presign (write) failed; next run will replay further", {
+            error: error instanceof Error ? error.message : String(error),
+            sessionId,
+        });
+        return;
+    }
+    let response;
+    try {
+        response = await fetch(presignedUrl, {
+            method: "PUT",
+            headers: { "content-type": "application/json" },
+            body: JSON.stringify(snapshot),
+        });
+    }
+    catch (error) {
+        v3_1.logger.warn("chat.agent: snapshot upload failed; next run will replay further", {
+            error: error instanceof Error ? error.message : String(error),
+            sessionId,
+        });
+        return;
+    }
+    if (!response.ok) {
+        v3_1.logger.warn("chat.agent: snapshot upload returned non-OK; next run will replay further", {
+            status: response.status,
+            sessionId,
+        });
+    }
+}
+/**
+ * Test-only entry point that bypasses `__setReadChatSnapshotImplForTests`
+ * and reaches the real `apiClient.getPayloadUrl` + `fetch` + JSON-parse path.
+ * Used by `chat-snapshot.test.ts` to verify 404 / 500 / malformed JSON /
+ * version-mismatch / network-error behavior end-to-end. Tests mock global
+ * `fetch` and the api-client config; this wrapper lets them drive the
+ * production code without the override hook short-circuiting.
+ *
+ * Not part of the public API. The `__` prefix and `ForTests` suffix mirror
+ * the override-hook setters above.
+ * @internal
+ */
+async function __readChatSnapshotProductionPathForTests(sessionId) {
+    const saved = readChatSnapshotImpl;
+    readChatSnapshotImpl = undefined;
+    try {
+        return await readChatSnapshot(sessionId);
+    }
+    finally {
+        readChatSnapshotImpl = saved;
+    }
+}
+/**
+ * Test-only entry point that bypasses `__setWriteChatSnapshotImplForTests`
+ * and reaches the real `apiClient.createUploadPayloadUrl` + `fetch` PUT
+ * path. Pairs with `__readChatSnapshotProductionPathForTests` — see that
+ * function's note for the rationale.
+ *
+ * Not part of the public API.
+ * @internal
+ */
+async function __writeChatSnapshotProductionPathForTests(sessionId, snapshot) {
+    const saved = writeChatSnapshotImpl;
+    writeChatSnapshotImpl = undefined;
+    try {
+        await writeChatSnapshot(sessionId, snapshot);
+    }
+    finally {
+        writeChatSnapshotImpl = saved;
+    }
+}
 /**
  * Merge two `UIMessage[]` lists by `id`, with the second list winning on
  * collision. Used at run boot to combine the snapshot's persisted history
@@ -1615,13 +1772,6 @@ async function installChatInputRouter(chatId, options) {
             checkpoint.appliedThrough = Math.max(checkpoint.appliedThrough ?? replayWindowEnd, replayWindowEnd);
         }
     }
-    // A boot that replayed `.in` itself has already answered everything up to
-    // `recoveredThrough`, so the floor has to cover it before the tail opens.
-    if (options?.recoveredThrough !== undefined) {
-        const recovered = options.recoveredThrough;
-        checkpoint.resumeFrom = Math.max(checkpoint.resumeFrom ?? recovered, recovered);
-        checkpoint.appliedThrough = Math.max(checkpoint.appliedThrough ?? checkpoint.resumeFrom, checkpoint.resumeFrom);
-    }
     const router = entry.router;
     router.restore(checkpoint);
     const floor = router.resumeFrom();
@@ -1630,10 +1780,6 @@ async function installChatInputRouter(chatId, options) {
         v3_1.sessionStreams.setLastDispatchedSeqNum(chatId, "in", floor);
     }
     v3_1.sessionStreams.onRecord(chatId, "in", (record) => {
-        // The floor is the tail's `Last-Event-ID`, but a reconnect can still
-        // re-deliver below it and a replayable route would re-queue it.
-        if (floor !== undefined && record.seqNum <= floor)
-            return true;
         router.ingest(record);
         return true;
     });
@@ -1805,35 +1951,6 @@ function spliceHandoverPartial(modelMessages, uiMessages, signal) {
  * @internal
  */
 const chatBackgroundQueueKey = locals_js_1.locals.create("chat.backgroundQueue");
-/**
- * Background injections a step-boundary drain handed to the model this turn,
- * with the transcript message they followed. Reconciled into the model lane
- * and the persisted injections once the turn's response is in.
- */
-const chatPendingBackgroundKey = locals_js_1.locals.create("chat.pendingBackground");
-/**
- * System-role context injected mid-conversation, held for the instructions lane.
- *
- * Kept apart from the message queue because ai@7 rejects a system message inside
- * `messages` for every provider — `standardizePrompt` throws upstream of any
- * provider call, and its own advice is to use the instructions option. Instructions
- * accept `Array<SystemModelMessage>`, so a system-role injection has a correct
- * home: appended as another system block rather than smuggled into the transcript.
- *
- * This is also the only way to inject *trusted* context. A message injected as
- * `user` is untrusted by construction, and a well-aligned model treats it that
- * way — it will say so, and re-derive the answer from tools instead.
- */
-const chatInjectedInstructionsKey = locals_js_1.locals.create("chat.injectedInstructions");
-/**
- * What a turn already consumed from the instructions lane, so a second
- * `toStreamTextOptions()` call in the same turn sees the same blocks.
- *
- * Consumed blocks are moved here rather than left in the pending lane: leaving
- * them there means an injection made during the consumed turn sits behind them,
- * and clearing the lane on the next turn destroys both.
- */
-const chatInstructionsConsumedKey = locals_js_1.locals.create("chat.injectedInstructionsConsumed");
 /**
  * Run-scoped pipe counter. Stored in locals so concurrent runs in the
  * same worker don't share state.
@@ -2344,63 +2461,18 @@ const chatToolsOptionKey = locals_js_1.locals.create("chat.toolsOption");
 const chatResolvedToolsKey = locals_js_1.locals.create("chat.resolvedTools");
 /** @internal Flag set by `chat.requestUpgrade()` to exit the loop after the current turn. */
 const chatUpgradeRequestedKey = locals_js_1.locals.create("chat.upgradeRequested");
-/** @internal Target for the upgrade handoff, set by `chat.requestUpgrade({ externalDeploymentId })`. */
-const chatUpgradeExternalDeploymentIdKey = locals_js_1.locals.create("chat.upgradeExternalDeploymentId");
 /**
  * @internal Flag set by `chat.endRun()` to exit the loop after the current
  * turn completes, without any upgrade semantics. Checked at the same
  * post-turn / pre-wait sites as `chatUpgradeRequestedKey`.
  */
 const chatEndRunRequestedKey = locals_js_1.locals.create("chat.endRunRequested");
-/**
- * @internal Set by `chat.close()`. Holds the close request (and its reason)
- * for the rest of the run: the loop writes the terminal `session-closed`
- * record, closes the session row, and exits at the same post-turn /
- * pre-wait sites as `chatEndRunRequestedKey`.
- */
-const chatCloseRequestedKey = locals_js_1.locals.create("chat.closeRequested");
-/** @internal Matches the server's `CloseSessionRequestBody.reason` cap. */
-const CHAT_CLOSE_REASON_MAX_LENGTH = 256;
-/** @internal Set once the session row is closed, so the close happens once. */
-const chatClosePerformedKey = locals_js_1.locals.create("chat.closePerformed");
-/**
- * @internal Set once the terminal `.out` record is written. Tracked apart from
- * {@link chatClosePerformedKey} so a retried close does not emit a second
- * client-visible event.
- */
-const chatCloseRecordWrittenKey = locals_js_1.locals.create("chat.closeRecordWritten");
 /** @internal */
 const chatAgentCompactionKey = locals_js_1.locals.create("chat.agentCompaction");
 /** @internal */
 const chatPendingMessagesKey = locals_js_1.locals.create("chat.pendingMessages");
 /** @internal */
 const chatSteeringQueueKey = locals_js_1.locals.create("chat.steeringQueue");
-/**
- * This turn's new messages, as `onTurnComplete.newUIMessages` will see them.
- *
- * Held in locals because `drainSteeringQueue` runs outside the turn closure and
- * has to append the messages it injects. Without that, an injected message
- * reaches the model and the browser but no hook, so an app persisting from
- * `onTurnComplete` never learns it existed.
- */
-const chatTurnNewUIMessagesKey = locals_js_1.locals.create("chat.turnNewUIMessages");
-/**
- * Steering messages a drain consumed that the model accumulator has not been
- * given yet.
- *
- * The two accumulators are maintained separately, and the model one is
- * normally advanced by appending each turn's delta. A drained message is
- * appended to the UI one but reaches the model only through the `prepareStep`
- * return value, which is per-step: without this the model lane never learns
- * the message exists and every later turn of the run answers without it,
- * while the browser, the snapshot and `chat.history.*` all still show it.
- *
- * Held as the messages rather than a "rebuild me" flag because the model lane
- * can only be appended to, never reconstructed. Compaction replaces it with a
- * summary and deliberately leaves the UI lane whole, so reconverting the UI
- * lane restores every message the summary replaced.
- */
-const chatPendingSteerKey = locals_js_1.locals.create("chat.pendingSteer");
 /** @internal — IDs of messages that were successfully injected via prepareStep */
 const chatInjectedMessageIdsKey = locals_js_1.locals.create("chat.injectedMessageIds");
 /** @internal — non-transient data parts queued via chat.response or writer.write() for accumulation into the response message */
@@ -2753,32 +2825,20 @@ function chatCompactionStep(options) {
         return result.type === "skipped" ? undefined : result;
     };
 }
-const EMPTY_DRAIN = { injected: [], claimed: [] };
-/**
- * The model messages to record for one claimed message. Without `prepare`
- * each entry's own conversion is used. With it, `prepare` returned one list
- * for the whole batch, so the first claimed message carries all of it and the
- * rest carry none, which keeps the total exactly what the model received.
- */
-function modelFormOf(m, batch, injected) {
-    return batch[0] === m ? injected : [];
-}
+// ---------------------------------------------------------------------------
+// Steering queue drain — shared by toStreamTextOptions, session, accumulator
+// ---------------------------------------------------------------------------
 /**
  * Drain the steering queue as a batch. Calls `shouldInject` once with all
  * pending messages. If it returns true, calls `prepareMessages` once to
  * transform the batch, then clears the queue.
- * Returns the model messages to inject and the UI messages actually claimed.
- *
- * `claimed` is returned rather than only published to locals because each
- * surface files it somewhere different: `chat.agent` has an accumulator in
- * locals, while `chat.createSession` keeps its own. Publishing to locals alone
- * is silently a no-op for any surface that never set the key.
+ * Returns the model messages to inject (empty if none).
  * @internal
  */
 async function drainSteeringQueue(config, messages, steps, queueOverride) {
     const queue = queueOverride ?? locals_js_1.locals.get(chatSteeringQueueKey);
     if (!queue || queue.length === 0)
-        return EMPTY_DRAIN;
+        return [];
     const ctx = locals_js_1.locals.get(chatTurnContextKey);
     const stepNumber = steps.length - 1;
     /**
@@ -2801,7 +2861,7 @@ async function drainSteeringQueue(config, messages, steps, queueOverride) {
     // Call shouldInject once for the whole batch
     const shouldInject = config.shouldInject ? await config.shouldInject(batchEvent) : false;
     if (!shouldInject)
-        return EMPTY_DRAIN;
+        return [];
     const textOfUIMessage = (m) => (m.parts ?? [])
         .filter((p) => p.type === "text")
         .map((p) => p.text)
@@ -2847,7 +2907,7 @@ async function drainSteeringQueue(config, messages, steps, queueOverride) {
                 queue.splice(at, 1);
         }
         if (claimed.length === 0)
-            return EMPTY_DRAIN;
+            return [];
         /**
          * Give the claim back if the transform fails. `prepare` is caller code and
          * can throw; the records have already left the router by this point, so
@@ -2877,37 +2937,6 @@ async function drainSteeringQueue(config, messages, steps, queueOverride) {
             for (const m of claimedUIMessages)
                 injectedIds.add(m.id);
         }
-        // Record them as part of the conversation.
-        //
-        // The model has them and the browser has them; without this the
-        // accumulator does not, so they reach neither `uiMessages` nor
-        // `newUIMessages` on `onTurnComplete` and an app that persists from there
-        // silently loses the instruction the answer was shaped by. Appending here
-        // rather than at turn end keeps them in the order they happened: after the
-        // message that started the turn, before the response that answers it.
-        //
-        // De-duplicated by id because a step boundary can drain more than once per
-        // turn, and because a message that failed to inject falls back to becoming
-        // its own turn, where it is accumulated the normal way.
-        const currentUIMessages = locals_js_1.locals.get(chatCurrentUIMessagesKey);
-        const turnNew = locals_js_1.locals.get(chatTurnNewUIMessagesKey);
-        for (const m of claimedUIMessages) {
-            if (currentUIMessages && !currentUIMessages.some((existing) => existing.id === m.id)) {
-                currentUIMessages.push(m);
-            }
-            if (turnNew && !turnNew.some((existing) => existing.id === m.id)) {
-                turnNew.push(m);
-            }
-        }
-        if (claimedUIMessages.length > 0 && currentUIMessages) {
-            const pendingSteer = locals_js_1.locals.get(chatPendingSteerKey) ?? [];
-            for (const m of claimedUIMessages) {
-                if (!pendingSteer.some((existing) => existing.ui.id === m.id)) {
-                    pendingSteer.push({ ui: m, model: modelFormOf(m, claimedUIMessages, injected) });
-                }
-            }
-            locals_js_1.locals.set(chatPendingSteerKey, pendingSteer);
-        }
         // Write injection confirmation chunk to the stream so the frontend
         // knows which messages were injected and where in the response.
         if (injected.length > 0) {
@@ -2949,7 +2978,7 @@ async function drainSteeringQueue(config, messages, steps, queueOverride) {
                 /* non-fatal */
             }
         }
-        return { injected, claimed: claimedUIMessages };
+        return injected;
     }, {
         attributes: {
             [v3_1.SemanticInternalAttributes.STYLE_ICON]: "tabler-message-forward",
@@ -3175,149 +3204,32 @@ function buildSkillTools(skills) {
     return { loadSkill, readFile, bash };
 }
 /**
- * A `streamText` with the agent's managed options already applied.
- *
- * Handed to `run()` so the managed state cannot be missed by omission. Spreading
- * `chat.toStreamTextOptions()` is still supported and equivalent; this exists
- * because forgetting the spread silently drops the managed prompt, the skill
- * tools, telemetry, and the `prepareStep` that delivers steering, compaction and
- * conversational injection.
- *
- * Caller options win for everything the caller owns (model, messages, signal,
- * stopWhen). The three that would otherwise clobber managed behaviour are
- * merged instead of replaced:
- *
- * - `tools` are passed into the helper, so skill tools survive.
- * - `prepareStep` is composed after the managed one, so a caller's per-step
- *   overrides apply on top of steering and compaction instead of disabling them.
- *
- * `system` may be set at the call site, on `chat.agent({ system })`, or
- * through `chat.prompt.set()`, but only in one of them. Setting it in two
- * places throws: no shape merges two system values on every supported
- * version, and dropping one silently is the failure this seam exists to
- * prevent. Injected instructions append to whichever one is in play.
- */
-/**
- * The agent-level managed options (`registry`, `system`, `cacheControl`,
- * `systemProviderOptions`), published for the run so that
- * `chat.toStreamTextOptions()` applies them too. Without this only the bound
- * `streamText` saw them, and the documented spread form silently ran without
- * the agent's system prompt or model.
- */
-const chatAgentManagedConfigKey = locals_js_1.locals.create("chat.agentManagedConfig");
-/**
- * The caller's `streamText` options merged with the agent's managed ones.
+ * Returns an options object ready to spread into `streamText()`.
+ *
+ * Includes `system`, `experimental_telemetry`, and any config fields
+ * (temperature, maxTokens, etc.) from the stored prompt.
+ *
+ * When a `registry` is provided and the prompt has a `model` string,
+ * the resolved `LanguageModel` is included as `model`.
  *
- * Pure, and separate from the call so it can be asserted directly: everything
- * the caller did not name has to survive the merge, and the way to be sure of
- * that is to look at the merged object rather than at what the model received.
+ * If no prompt has been set, returns `{}` (no-op spread).
  */
-function buildManagedStreamTextOptions(options, config) {
-    const { registry, system: agentSystem, cacheControl, systemProviderOptions, tools: agentTools, } = config;
-    /**
-     * Only the three keys that collide are intercepted. Everything else, telemetry
-     * included, stays in `rest` and reaches `streamText` untouched, with the
-     * caller's value winning because `rest` is spread after `managed`. Pulling a
-     * key out to "handle" it is how a caller's option gets silently dropped.
-     */
-    const { tools, system: callerSystem, prepareStep: callerPrepareStep, ...rest } = options;
-    const managed = toStreamTextOptions({
-        registry,
-        system: callerSystem ?? agentSystem,
-        cacheControl,
-        systemProviderOptions,
-        /**
-         * A call site that names `tools` replaces the agent's set rather than
-         * adding to it, so narrowing the tools for one call still works. Omitting
-         * `tools` falls back to the agent's, which is what an `onAction`
-         * regenerate needs: without it a regenerated answer can call nothing.
-         */
-        tools: (tools ?? agentTools),
-    });
-    const promptSystem = locals_js_1.locals.get(chatPromptKey)?.text;
-    /**
-     * Two managed sources conflict too, not only a caller against a managed one.
-     * `toStreamTextOptions` resolves `prompt?.text || agentSystem`, so the prompt
-     * would win and `chat.agent({ system })` would go nowhere.
-     */
-    if (promptSystem && agentSystem) {
-        throw new Error("chat.agent: `system` is set both on chat.agent({ system }) and by chat.prompt.set(), and only one " +
-            "of them can apply. The prompt would win and the agent's `system` would go nowhere. Keep it in one " +
-            "place, and add per-turn context with chat.inject({ role: 'system' }).");
-    }
-    const managedSystem = promptSystem || agentSystem;
-    if (callerSystem !== undefined && managedSystem) {
-        throw new Error("chat.agent: `system` is already set " +
-            (promptSystem ? "by chat.prompt.set()" : "on chat.agent({ system })") +
-            ", so it cannot also be passed to the `streamText` given to run(). Set it in one place, and add " +
-            "per-turn context with chat.inject({ role: 'system' }) rather than a second system value.");
-    }
-    const managedPrepareStep = managed.prepareStep;
-    if (typeof callerPrepareStep === "function") {
-        managed.prepareStep = async (arg) => {
-            const first = managedPrepareStep ? await managedPrepareStep(arg) : undefined;
-            const second = await callerPrepareStep({ ...arg, ...(first ?? {}) });
-            return { ...(first ?? {}), ...(second ?? {}) };
-        };
-    }
-    return { ...managed, ...rest };
-}
-/** @internal Test hook for {@link buildManagedStreamTextOptions}. */
-exports.__buildManagedStreamTextOptionsForTests = buildManagedStreamTextOptions;
-function createBoundStreamText(registry, agentSystem, agentCacheControl, agentSystemProviderOptions) {
-    const bound = (options = {}) => (0, ai_runtime_js_1.streamText)(buildManagedStreamTextOptions(options, {
-        registry,
-        system: agentSystem,
-        cacheControl: agentCacheControl,
-        systemProviderOptions: agentSystemProviderOptions,
-        /** Read per call, so per-turn tools resolved after binding are included. */
-        tools: locals_js_1.locals.get(chatResolvedToolsKey),
-    }));
-    return bound;
-}
 function toStreamTextOptions(options) {
-    const agentDefaults = locals_js_1.locals.get(chatAgentManagedConfigKey);
-    if (agentDefaults) {
-        options = {
-            registry: agentDefaults.registry,
-            system: agentDefaults.system,
-            cacheControl: agentDefaults.cacheControl,
-            systemProviderOptions: agentDefaults.systemProviderOptions,
-            ...options,
-        };
-    }
     const prompt = locals_js_1.locals.get(chatPromptKey);
     const skills = locals_js_1.locals.get(chatSkillsKey);
     const result = {};
     // Build the combined system prompt: stored prompt + skills preamble.
-    const baseSystem = options?.system;
-    const baseSystemText = typeof baseSystem === "string"
-        ? baseSystem
-        : typeof baseSystem?.content === "string"
-            ? baseSystem.content
-            : "";
-    const promptText = prompt?.text || baseSystemText;
+    const promptText = prompt?.text ?? "";
     const skillsText = skills && skills.length > 0 ? buildSkillsSystemPrompt(skills) : "";
     if (promptText || skillsText) {
         const systemText = [promptText, skillsText].filter(Boolean).join("\n\n");
-        /**
-         * Resolve system-prompt provider options for caching. Precedence, most
-         * specific first and no deep merge: explicit `systemProviderOptions`, the
-         * `cacheControl` sugar, the ones carried on a structured `system` message,
-         * then whatever `chat.prompt.set()` stored.
-         *
-         * A structured `system` counts only when its own text is the one being
-         * sent. When `chat.prompt.set()` supplied the text, its provider options
-         * are the ones that describe it.
-         */
-        const baseSystemProviderOptions = !prompt?.text && baseSystem && typeof baseSystem !== "string"
-            ? baseSystem.providerOptions
-            : undefined;
+        // Resolve system-prompt provider options for caching. Precedence (most
+        // specific wins, no deep merge): explicit `systemProviderOptions` →
+        // `cacheControl` sugar → `providerOptions` stored on `chat.prompt.set()`.
         const systemProviderOptions = options?.systemProviderOptions ??
             (options?.cacheControl
                 ? { anthropic: { cacheControl: options.cacheControl } }
                 : undefined) ??
-            baseSystemProviderOptions ??
             locals_js_1.locals.get(chatPromptProviderOptionsKey);
         // A bare string stays a bare string (the unchanged default). With provider
         // options, emit a structured `SystemModelMessage` so the provider can cache
@@ -3326,88 +3238,6 @@ function toStreamTextOptions(options) {
             ? { role: "system", content: systemText, providerOptions: systemProviderOptions }
             : systemText;
     }
-    /**
-     * Append anything injected as system context, in whichever shape the installed
-     * AI SDK accepts.
-     *
-     * `system` widened over time: on ai@5 it is `string` only, and from ai@6 it is
-     * `string | SystemModelMessage | Array<SystemModelMessage>`. This package's peer
-     * range still spans all three, so emitting an array unconditionally would break
-     * v5 consumers — for whom a system-role injection used to work, since v5 accepted
-     * a system message inside `messages` that v7 rejects.
-     *
-     * So: concatenate into one string when the base is a plain string, which every
-     * version accepts and which loses nothing (separate blocks only matter for
-     * per-block `providerOptions`). Use the array form only when the base is already
-     * a structured message — that path requires v6+ regardless, because it is how
-     * prompt caching marks the system block, and flattening it would silently throw
-     * the cache away.
-     *
-     * Either way the injected text goes last: the base prompt keeps its position for
-     * caching, and the addition reads as a later amendment. A changed prefix does
-     * cost the first call its cache hit, on turns that actually injected.
-     */
-    /**
-     * Consumed once per turn, not once per read, and moved out of the lane rather
-     * than marked read in place.
-     *
-     * Per turn, because a `run()` that builds options twice (a cheap classifier
-     * pass and then the answer) has to see the injection in both, and draining on
-     * read hands it to whichever call ran first. Moved out, because blocks left in
-     * the lane sit in front of anything injected during the same turn, and
-     * clearing the lane on the next turn then destroys both. Outside a turn there
-     * is no turn to scope the stash to, so the lane drains on read there.
-     */
-    const injectedInstructions = locals_js_1.locals.get(chatInjectedInstructionsKey);
-    const currentTurn = locals_js_1.locals.get(chatTurnContextKey)?.turn;
-    const consumedThisTurn = currentTurn === undefined ? undefined : locals_js_1.locals.get(chatInstructionsConsumedKey);
-    let injectedBlocks = [];
-    if (consumedThisTurn && consumedThisTurn.turn === currentTurn) {
-        injectedBlocks = consumedThisTurn.blocks;
-        // Anything injected since the stash was taken joins it, so an instruction
-        // added after an action read the lane still reaches the real turn that
-        // shares the action's turn number, rather than the one after.
-        if (injectedInstructions && injectedInstructions.length > 0) {
-            injectedBlocks.push(...injectedInstructions.splice(0));
-        }
-    }
-    else if (injectedInstructions && injectedInstructions.length > 0) {
-        injectedBlocks = injectedInstructions.splice(0);
-        if (currentTurn !== undefined) {
-            locals_js_1.locals.set(chatInstructionsConsumedKey, { turn: currentTurn, blocks: injectedBlocks });
-        }
-    }
-    if (injectedBlocks.length > 0) {
-        const blocks = injectedBlocks;
-        const injectedText = blocks
-            .map((block) => (typeof block.content === "string" ? block.content : ""))
-            .filter(Boolean)
-            .join("\n\n");
-        const base = result.system;
-        if (base === undefined) {
-            result.system = injectedText;
-        }
-        else if (typeof base === "string") {
-            result.system = [base, injectedText].filter(Boolean).join("\n\n");
-        }
-        else {
-            // Merged into the existing block rather than added as a second one. An array
-            // of system blocks would keep the base block's cache entry, but ai@5 rejects
-            // it outright ("Invalid prompt: system must be a string") while accepting a
-            // single structured block, and this package's peer range still spans v5.
-            // Choosing per version would mean resolving the installed version at runtime,
-            // which is not something to build on: `import.meta.url` is illegal in this
-            // package's CommonJS output, and a bundled task may have no resolvable `ai`
-            // to read. One shape that works everywhere beats a cache hit.
-            const baseBlock = base;
-            result.system = {
-                ...baseBlock,
-                content: [typeof baseBlock.content === "string" ? baseBlock.content : "", injectedText]
-                    .filter(Boolean)
-                    .join("\n\n"),
-            };
-        }
-    }
     // Prompt-related options (only if chat.prompt.set() was called)
     if (prompt) {
         // Resolve model via registry if both are present
@@ -3462,7 +3292,7 @@ function toStreamTextOptions(options) {
             }
             // 2. Pending message injection (steering)
             if (taskPendingMessages) {
-                const { injected } = await drainSteeringQueue(taskPendingMessages, resultMessages ?? messages, steps);
+                const injected = await drainSteeringQueue(taskPendingMessages, resultMessages ?? messages, steps);
                 if (injected.length > 0) {
                     resultMessages = [...(resultMessages ?? messages), ...injected];
                 }
@@ -3472,89 +3302,18 @@ function toStreamTextOptions(options) {
             if (bgQueue && bgQueue.length > 0) {
                 const injected = bgQueue.splice(0); // drain
                 resultMessages = [...(resultMessages ?? messages), ...injected];
-                const pendingBackground = locals_js_1.locals.get(chatPendingBackgroundKey) ?? [];
-                pendingBackground.push({
-                    afterId: locals_js_1.locals.get(chatCurrentUIMessagesKey)?.at(-1)?.id ?? "",
-                    messages: injected,
-                });
-                locals_js_1.locals.set(chatPendingBackgroundKey, pendingBackground);
             }
             return resultMessages ? { messages: resultMessages } : undefined;
         };
     }
     return result;
 }
-const actionTurnBrand = Symbol.for("trigger.dev/chat/actionTurn");
-/**
- * Turn the current action into a turn.
- *
- * Return it from `onAction` after editing history. The action's own work is
- * finished first (the edit is applied and snapshotted), then a turn runs on the
- * result exactly as a message turn does: `onTurnStart`, `run()` with the edited
- * history, `onBeforeTurnComplete`, `onTurnComplete`, and the turn counter
- * advances. That gives the answer everything a turn has, the system prompt,
- * tools, steering, compaction, injected instructions and persistence, with no
- * action-specific handling.
- *
- * @example
- * ```ts
- * onAction: async ({ action }) => {
- *   if (action.type === "regenerate") {
- *     chat.history.slice(0, -1);
- *     return chat.turn();
- *   }
- *   if (action.type === "undo") chat.history.slice(0, -2); // no turn
- * },
- * ```
- */
-function chatTurn() {
-    return { [actionTurnBrand]: true };
-}
-function isActionTurn(value) {
-    return typeof value === "object" && value !== null && value[actionTurnBrand] === true;
-}
-/**
- * Replace, in a model lane, the run of messages one UI message contributed.
- *
- * Used when a UI message is replaced in place (a tool-approval continuation
- * merging onto the trailing assistant, a captured response reusing an existing
- * id, a partial replacing an existing message). Reconverting the whole lane
- * from the UI lane would also replace a compaction summary with the full
- * transcript and drop the model forms `pendingMessages.prepare` produced.
- *
- * The replaced message is the trailing one, so its run is the lane's tail,
- * before any steer forms appended after it this turn (`tailAfter`). If the
- * tail does not match the old message's conversion, nothing is changed and
- * `false` is returned so the caller can fall back to a full reconversion.
- */
-async function replaceModelRun(lane, oldUi, newUi, tailAfter) {
-    const oldRun = await toModelMessages([stripProviderMetadata(oldUi)]);
-    const newRun = await toModelMessages([stripProviderMetadata(newUi)]);
-    const end = lane.length - tailAfter;
-    const start = end - oldRun.length;
-    if (start < 0 || end > lane.length)
-        return false;
-    if (JSON.stringify(lane.slice(start, end)) !== JSON.stringify(oldRun))
-        return false;
-    lane.splice(start, oldRun.length, ...newRun);
-    return true;
-}
 function isUIMessageStreamable(value) {
     return (typeof value === "object" &&
         value !== null &&
         "toUIMessageStream" in value &&
         typeof value.toUIMessageStream === "function");
 }
-const warnedHydrateMessagesDeprecated = new Set();
-function warnHydrateMessagesDeprecatedOnce(agentId) {
-    if (warnedHydrateMessagesDeprecated.has(agentId))
-        return;
-    warnedHydrateMessagesDeprecated.add(agentId);
-    console.warn(`[chat.agent] \`hydrateMessages\` on "${agentId}" is deprecated. Give the agent a transcript ` +
-        "storage instead: `save` receives every change to the conversation and `loadContext` " +
-        "lets the application own the model's context, with crash recovery and durable " +
-        "compaction that `hydrateMessages` never had.");
-}
 let warnedMissingOnAction = false;
 function warnMissingOnActionOnce() {
     if (warnedMissingOnAction)
@@ -3687,22 +3446,10 @@ function chatCustomAgent(options) {
             await installChatInputRouter(payload.chatId, {
                 resuming: Boolean(payload.continuation),
             });
-            // A custom agent's loop is the customer's, so there is no exit site the
-            // SDK controls. Perform a requested close when `run()` returns, whatever
-            // shape the loop had. Idempotent, so the createSession iterator having
-            // already closed on its own exit costs nothing.
-            const withClose = async (result) => {
-                try {
-                    return await result;
-                }
-                finally {
-                    await performChatClose();
-                }
-            };
             // Keep the schema-free path identical to the original custom-agent
             // wrapper, including when userRun starts executing.
             if (!parseClientData) {
-                return withClose(userRun(payload, runOptions));
+                return userRun(payload, runOptions);
             }
             const isHandoverBoot = payload.trigger === "handover-prepare";
             const isMessagelessBoot = payload.trigger === "preload" ||
@@ -3718,7 +3465,7 @@ function chatCustomAgent(options) {
                 writeErrorToStream: !isMessagelessBoot && !isHandoverBoot,
             });
             if (validated.ok) {
-                return withClose(userRun(validated.payload, runOptions));
+                return userRun(validated.payload, runOptions);
             }
             if (isHandoverBoot) {
                 const signal = await waitForHandover({
@@ -3754,7 +3501,7 @@ function chatCustomAgent(options) {
                 sessionId: next.output.sessionId ?? payload.sessionId,
                 idleTimeoutInSeconds: next.output.idleTimeoutInSeconds ?? payload.idleTimeoutInSeconds,
             };
-            return withClose(userRun(recoveredPayload, runOptions));
+            return userRun(recoveredPayload, runOptions);
         },
     });
     // Register clientDataSchema so the CLI converts it to JSONSchema
@@ -3767,19 +3514,7 @@ function chatCustomAgent(options) {
     return task;
 }
 function chatAgent(options) {
-    const { run: userRun, clientDataSchema, onBoot, onRecoveryBoot, onPreload, onChatStart, onValidateMessages, hydrateMessages, storage, actionSchema, onAction, onTurnStart, onBeforeTurnComplete, onCompacted, compaction, pendingMessages: pendingMessagesConfig, prepareMessages, tools: toolsOption, onTurnComplete, maxTurns = 100, turnTimeout = "1h", idleTimeoutInSeconds = 30, chatAccessTokenTTL = "1h", preloadIdleTimeoutInSeconds, preloadTimeout, uiMessageStreamOptions, onChatSuspend, onChatResume, exitAfterPreloadIdle = false, oomMachine, registry: promptRegistry, system: agentSystem, cacheControl: agentCacheControl, systemProviderOptions: agentSystemProviderOptions, versionSkew, ...restOptions } = options;
-    if (hydrateMessages) {
-        if (storage) {
-            throw new Error(`chat.agent: "${options.id}" sets both \`hydrateMessages\` and \`storage\`. ` +
-                "`hydrateMessages` is deprecated and replaced by the storage: `save` receives every " +
-                "change and `loadContext` on the storage owns the model's context. Remove `hydrateMessages`.");
-        }
-        if (typeof (transcriptStorageOverride ?? transcriptStorage_js_1.defaultStorage).loadContext === "function") {
-            throw new Error(`chat.agent: "${options.id}" sets \`hydrateMessages\` and uses a transcript storage with ` +
-                "`loadContext`. Both would own the model's context; keep one.");
-        }
-        warnHydrateMessagesDeprecatedOnce(options.id);
-    }
+    const { run: userRun, clientDataSchema, onBoot, onRecoveryBoot, onPreload, onChatStart, onValidateMessages, hydrateMessages, actionSchema, onAction, onTurnStart, onBeforeTurnComplete, onCompacted, compaction, pendingMessages: pendingMessagesConfig, prepareMessages, tools: toolsOption, onTurnComplete, maxTurns = 100, turnTimeout = "1h", idleTimeoutInSeconds = 30, chatAccessTokenTTL = "1h", preloadIdleTimeoutInSeconds, preloadTimeout, uiMessageStreamOptions, onChatSuspend, onChatResume, exitAfterPreloadIdle = false, oomMachine, ...restOptions } = options;
     const parseClientData = clientDataSchema ? (0, v3_1.getSchemaParseFn)(clientDataSchema) : undefined;
     const parseAction = actionSchema ? (0, v3_1.getSchemaParseFn)(actionSchema) : undefined;
     // chat.agent does not expose generic retry options (see docstring on
@@ -3862,42 +3597,6 @@ function chatAgent(options) {
             // durable snapshot + `session.out` replay (or `hydrateMessages` if
             // registered) — the wire is delta-only now, no longer a seed.
             let accumulatedMessages = [];
-            /**
-             * Give the model accumulator the background injections a step-boundary
-             * drain handed to the model this turn, and record them for persistence.
-             * Returns how many model messages were appended.
-             */
-            const reconcilePendingBackground = () => {
-                const pending = locals_js_1.locals.get(chatPendingBackgroundKey);
-                if (!pending || pending.length === 0)
-                    return 0;
-                locals_js_1.locals.set(chatPendingBackgroundKey, []);
-                let appended = 0;
-                for (const entry of pending) {
-                    accumulatedMessages.push(...entry.messages);
-                    laneInjections.push(entry);
-                    appended += entry.messages.length;
-                }
-                return appended;
-            };
-            /**
-             * Give the model accumulator the steering messages a drain consumed,
-             * in the form the model actually received. Appended, never reconverted
-             * from the UI lane, so a model-only compaction summary survives. Called
-             * on both the success and the error path, before the response or the
-             * partial joins the lane, so the order stays steer-then-answer.
-             */
-            const reconcilePendingSteer = (options) => {
-                const pending = locals_js_1.locals.get(chatPendingSteerKey);
-                if (!pending || pending.length === 0)
-                    return [];
-                locals_js_1.locals.set(chatPendingSteerKey, []);
-                for (const entry of pending) {
-                    accumulatedMessages.push(...entry.model);
-                    options?.turnNew?.push(...entry.model);
-                }
-                return pending;
-            };
             // Accumulated UI messages for persistence. Mirrors the model accumulator
             // but in frontend-friendly UIMessage format (with parts, id, etc.).
             let accumulatedUIMessages = [];
@@ -3915,157 +3614,7 @@ function chatAgent(options) {
             // collectively cost ~600ms on every first-message TTFC. Both reads
             // swallow errors internally; the agent stays available either way.
             const sessionIdForSnapshot = payload.sessionId ?? payload.chatId;
-            const transcriptStorage = storage ??
-                transcriptStorageOverride ??
-                transcriptStorage_js_1.defaultStorage;
-            const storageLoadContext = transcriptStorage.loadContext?.bind(transcriptStorage);
-            /**
-             * Who supplies the model's context each turn: the deprecated
-             * `hydrateMessages` hook, the storage's `loadContext`, or (undefined)
-             * the runtime's own transcript.
-             */
-            const loadContextHook = hydrateMessages
-                ? (event) => hydrateMessages(event)
-                : storageLoadContext
-                    ? (event) => storageLoadContext({ chatId: event.chatId, clientData: event.clientData }, event)
-                    : undefined;
-            let transcriptShadow = (0, transcriptStorage_js_1.createTranscriptShadow)([]);
-            let bootTranscriptState = null;
-            /**
-             * True while the model lane holds a compaction summary, so it cannot be
-             * rebuilt from the transcript and has to be persisted as state. Reset
-             * wherever the lane is reconverted from the UI lane.
-             */
-            let laneCompacted = false;
-            /** Conversational `chat.inject` messages in the lane, anchored to the transcript. */
-            let laneInjections = [];
-            let persistedStateSet = false;
             let bootSnapshot;
-            let bootClientData = payload.metadata;
-            if (parseClientData) {
-                try {
-                    bootClientData = await parseClientData(payload.metadata);
-                }
-                catch {
-                    bootClientData = payload.metadata;
-                }
-            }
-            /**
-             * The `lastOutEventId` the most recent snapshot carried.
-             *
-             * A snapshot written outside a turn — after an action mutates history — has
-             * no turn cursor of its own, and writing `undefined` there would drop the
-             * resume point and make the next boot replay from further back. Retaining it
-             * keeps an action's write cursor-neutral.
-             */
-            let lastSnapshotOutEventId;
-            const storageTrigger = (trigger) => trigger === "regenerate-message"
-                ? "regenerate-message"
-                : trigger === "action" || trigger === "action-turn"
-                    ? "action"
-                    : "submit-message";
-            /**
-             * Hand the runtime's view of the transcript to the storage as a
-             * changeset: the diff against what was last saved, plus the cursors the
-             * next boot resumes from. The shadow only advances when the save
-             * succeeds, so a failed save is folded into the next changeset.
-             */
-            /** The runtime's opaque state as of the last save; carried on every changeset's transcript. */
-            let transcriptState = null;
-            const saveTranscript = async (opts) => {
-                const { changes, shadow } = (0, transcriptStorage_js_1.diffTranscript)(transcriptShadow, opts.messages, {
-                    nonFinalIds: opts.nonFinalIds,
-                });
-                const throughId = opts.messages.at(-1)?.id ?? "";
-                const queued = locals_js_1.locals.get(chatBackgroundQueueKey) ?? [];
-                const runtimeState = laneCompacted || laneInjections.length > 0 || queued.length > 0
-                    ? {
-                        v: 1,
-                        ...(laneCompacted
-                            ? {
-                                compaction: {
-                                    modelMessages: accumulatedMessages,
-                                    throughId,
-                                    fingerprint: (0, transcriptStorage_js_1.prefixFingerprint)(shadow, throughId),
-                                },
-                            }
-                            : {}),
-                        ...(laneInjections.length > 0 ? { injections: laneInjections } : {}),
-                        ...(queued.length > 0 ? { queued: [...queued] } : {}),
-                    }
-                    : null;
-                if (runtimeState !== null || persistedStateSet) {
-                    changes.push({ op: "state", value: runtimeState });
-                }
-                transcriptState = runtimeState;
-                const inCursor = chatInputRouter().resumeFloor();
-                await transcriptStorage.save({
-                    chatId: payload.chatId,
-                    clientData: opts.clientData,
-                    turn: opts.turn,
-                    trigger: opts.trigger,
-                    runId: ctx.run.id,
-                    ctx,
-                }, {
-                    reason: opts.reason,
-                    changes,
-                    transcript: {
-                        entries: opts.messages.map((message) => ({
-                            id: message.id,
-                            final: !shadow.nonFinal.has(message.id),
-                            message,
-                        })),
-                        state: transcriptState,
-                    },
-                    cursors: {
-                        lastOutEventId: opts.lastOutEventId,
-                        lastInEventId: inCursor !== undefined ? String(inCursor) : undefined,
-                    },
-                });
-                transcriptShadow = shadow;
-                persistedStateSet = runtimeState !== null;
-            };
-            /**
-             * Persist the accumulator outside a turn.
-             *
-             * An action is not a turn, so it never reaches the turn-complete path where
-             * the transcript is normally saved, but a `chat.history` mutation changes
-             * the conversation and has to survive the run ending.
-             *
-             * Cursor-neutral: an action has no turn cursor of its own, and writing
-             * `undefined` would drop the resume point the last turn established and make
-             * the next boot replay from further back.
-             */
-            const writeSnapshotOutsideTurn = async (reason, turnContext) => {
-                if (hydrateMessages)
-                    return;
-                try {
-                    await tracer_js_1.tracer.startActiveSpan("snapshot.write", async () => {
-                        await saveTranscript({
-                            reason: "action",
-                            messages: accumulatedUIMessages,
-                            turn: turnContext.turn,
-                            trigger: "action",
-                            clientData: turnContext.clientData,
-                            lastOutEventId: lastSnapshotOutEventId,
-                        });
-                    }, {
-                        attributes: {
-                            [v3_1.SemanticInternalAttributes.STYLE_ICON]: "task-hook-onStart",
-                            [v3_1.SemanticInternalAttributes.COLLAPSED]: true,
-                            "chat.snapshot.reason": reason,
-                            "chat.messages.count": accumulatedUIMessages.length,
-                        },
-                    });
-                }
-                catch (error) {
-                    v3_1.logger.warn("chat.agent: snapshot write outside a turn failed; the change may not survive a continuation", {
-                        error: error instanceof Error ? error.message : String(error),
-                        sessionId: sessionIdForSnapshot,
-                        reason,
-                    });
-                }
-            };
             let replayedSettled = [];
             let replayedPartial;
             let replayedPartialRaw;
@@ -4081,7 +3630,7 @@ function chatAgent(options) {
             // resume-cursor block reuses it instead of re-scanning.
             let bootInCursor;
             let bootInCursorResolved = false;
-            if (couldHavePriorState) {
+            if (!hydrateMessages && couldHavePriorState) {
                 // Single parent span for the whole boot read phase — snapshot
                 // read, session.out replay, session.in replay. Per-phase timing
                 // + result counts are attributes on the span.
@@ -4089,33 +3638,18 @@ function chatAgent(options) {
                     // snapshot read
                     const snapStart = Date.now();
                     try {
-                        const loaded = hydrateMessages
-                            ? undefined
-                            : await transcriptStorage.load({
-                                chatId: payload.chatId,
-                                clientData: bootClientData,
-                            });
-                        if (loaded) {
-                            transcriptShadow = (0, transcriptStorage_js_1.createTranscriptShadow)(loaded.messages, new Set(loaded.nonFinalIds ?? []));
-                            bootTranscriptState = loaded.state;
-                            transcriptState = loaded.state ?? null;
-                            persistedStateSet = loaded.state !== null && loaded.state !== undefined;
-                            bootSnapshot = {
-                                messages: loaded.messages,
-                                lastOutEventId: loaded.cursors?.lastOutEventId,
-                                lastInEventId: loaded.cursors?.lastInEventId,
-                            };
-                        }
+                        bootSnapshot = await readChatSnapshot(sessionIdForSnapshot);
                     }
                     catch (error) {
-                        v3_1.logger.warn("chat.agent: transcript load failed; continuing from the stream tail", {
+                        // `readChatSnapshot` already swallows + warns internally; this catch
+                        // is just belt-and-suspenders against tracer/span errors.
+                        v3_1.logger.warn("chat.agent: snapshot read failed; continuing without snapshot", {
                             error: error instanceof Error ? error.message : String(error),
                             sessionId: sessionIdForSnapshot,
                         });
                     }
                     bootSpan.setAttribute("chat.boot.snapshot.durationMs", Date.now() - snapStart);
-                    bootSpan.setAttribute("chat.boot.snapshot.present", bootSnapshot !== undefined &&
-                        (bootSnapshot.messages.length > 0 || bootSnapshot.lastOutEventId !== undefined));
+                    bootSpan.setAttribute("chat.boot.snapshot.present", !!bootSnapshot);
                     bootSpan.setAttribute("chat.boot.snapshot.messageCount", bootSnapshot?.messages?.length ?? 0);
                     // Seed the trim chain from the snapshot's `lastOutEventId` (the SSE
                     // id of the previous turn's `turn-complete` control record). The
@@ -4123,7 +3657,6 @@ function chatAgent(options) {
                     // Without seeding, the new worker would emit no trim on its first
                     // turn (chain self-bootstraps from turn 2), so this is purely an
                     // optimization to keep continuation runs bounded from the first turn.
-                    lastSnapshotOutEventId = bootSnapshot?.lastOutEventId;
                     if (bootSnapshot?.lastOutEventId !== undefined) {
                         const seeded = Number.parseInt(bootSnapshot.lastOutEventId, 10);
                         if (Number.isFinite(seeded)) {
@@ -4218,17 +3751,13 @@ function chatAgent(options) {
             // Reads the turn boundary and subscribes in one call. `bootInCursor` is
             // only a fallback: the boot block above may already have resolved a
             // cursor from the snapshot, which is used when the boundary itself
-            // carries none. Everything the boot replayed off `.in` is dispatched from
-            // `bootInjectedQueue` below, so it goes into the floor here — folded in
-            // after the subscription opens, the live tail re-delivers it as a turn.
-            const lastRecoveredInSeq = replayedInTail.length > 0 ? replayedInTail[replayedInTail.length - 1].seqNum : undefined;
+            // carries none.
             await installChatInputRouter(payload.chatId, {
                 fallbackResumeFrom: bootInCursorResolved ? bootInCursor : undefined,
-                recoveredThrough: lastRecoveredInSeq,
                 resuming: Boolean(payload.continuation) || ctx.attempt.number > 1,
             });
             // ── Recovery boot + chain reconstruction ────────────────────────
-            {
+            if (!hydrateMessages) {
                 const settledMessages = mergeByIdReplaceWins(bootSnapshot?.messages ?? [], replayedSettled);
                 const inFlightUsers = replayedInTail.map((r) => r.message);
                 const partialAssistant = replayedPartial;
@@ -4332,6 +3861,16 @@ function chatAgent(options) {
                 if (hookBeforeBoot) {
                     await hookBeforeBoot();
                 }
+                // Advance the session.in cursor past every recovered user so
+                // the live subscription doesn't re-deliver them.
+                if (replayedInTail.length > 0) {
+                    const lastRecoveredSeq = replayedInTail[replayedInTail.length - 1].seqNum;
+                    const currentCursor = v3_1.sessionStreams.lastSeqNum(payload.chatId, "in");
+                    if (currentCursor === undefined || lastRecoveredSeq > currentCursor) {
+                        v3_1.sessionStreams.setLastSeqNum(payload.chatId, "in", lastRecoveredSeq);
+                        v3_1.sessionStreams.setLastDispatchedSeqNum(payload.chatId, "in", lastRecoveredSeq);
+                    }
+                }
                 // Synthesize wire payloads for each recoveredTurn. The turn-loop
                 // pops these ahead of `messagesInput.waitWithIdleTimeout` so they
                 // dispatch as normal turns with the existing hook stack.
@@ -4381,8 +3920,7 @@ function chatAgent(options) {
                 // path where wire-borne UIMessage[] still seeds the accumulator,
                 // and it's safe because the route handler isn't subject to the
                 // `/in/append` 512 KiB cap.
-                if (!loadContextHook &&
-                    accumulatedUIMessages.length === 0 &&
+                if (accumulatedUIMessages.length === 0 &&
                     payload.trigger === "handover-prepare" &&
                     Array.isArray(payload.headStartMessages) &&
                     payload.headStartMessages.length > 0) {
@@ -4411,17 +3949,7 @@ function chatAgent(options) {
                         }
                     }
                     try {
-                        const bootRuntimeState = (0, transcriptStorage_js_1.parseTranscriptRuntimeState)(bootTranscriptState);
-                        const restored = await (0, transcriptStorage_js_1.restoreModelLane)(accumulatedUIMessages, bootRuntimeState, (messages) => toModelMessages(messages));
-                        accumulatedMessages = restored.messages;
-                        laneCompacted = restored.compacted;
-                        laneInjections = restored.injections;
-                        if (bootRuntimeState?.queued && bootRuntimeState.queued.length > 0) {
-                            locals_js_1.locals.set(chatBackgroundQueueKey, [
-                                ...(locals_js_1.locals.get(chatBackgroundQueueKey) ?? []),
-                                ...bootRuntimeState.queued,
-                            ]);
-                        }
+                        accumulatedMessages = await toModelMessages(accumulatedUIMessages);
                     }
                     catch (error) {
                         v3_1.logger.warn("chat.agent: toModelMessages failed at boot; starting empty", {
@@ -4435,12 +3963,6 @@ function chatAgent(options) {
                 // before any hook (`onChatStart`, `onTurnStart`, etc.) fires.
                 locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
             }
-            locals_js_1.locals.set(chatAgentManagedConfigKey, {
-                registry: promptRegistry,
-                system: agentSystem,
-                cacheControl: agentCacheControl,
-                systemProviderOptions: agentSystemProviderOptions,
-            });
             // Token usage tracking across turns
             let previousTurnUsage;
             let cumulativeUsage = emptyUsage();
@@ -4798,7 +4320,6 @@ function chatAgent(options) {
                     } // end else (no boot-injected first turn)
                 }
                 for (let turn = 0; turn < maxTurns; turn++) {
-                    let turnClientData = payload.metadata;
                     // Declared here so the finally can detach it — a handler leaked past
                     // its turn duplicates every mid-stream message into the shared buffer.
                     let turnMsgSub;
@@ -4820,7 +4341,6 @@ function chatAgent(options) {
                         // `cleanupAbortedParts` no longer has to be re-applied below.
                         const cleanedIncomingMessages = incomingMessages.map((msg) => msg.role === "assistant" ? cleanupAbortedParts(msg) : msg);
                         const clientData = (parseClientData ? await parseClientData(wireMetadata) : wireMetadata);
-                        turnClientData = clientData;
                         const lastUserMessage = extractLastUserMessageText(cleanedIncomingMessages);
                         // Actions are not turns. They use a different span name
                         // and don't carry a turn.number. Branched on at `isAction`.
@@ -4857,7 +4377,6 @@ function chatAgent(options) {
                             locals_js_1.locals.set(chatDeferKey, new Set());
                             locals_js_1.locals.set(chatCompactionStateKey, undefined);
                             locals_js_1.locals.set(chatSteeringQueueKey, []);
-                            locals_js_1.locals.set(chatPendingBackgroundKey, []);
                             locals_js_1.locals.set(chatResponsePartsKey, []);
                             // NOTE: chatBackgroundQueueKey is NOT reset here — messages injected
                             // by deferred work from the previous turn's onTurnComplete need to
@@ -4939,10 +4458,6 @@ function chatAgent(options) {
                             // Track new messages for this turn (user input + assistant response).
                             const turnNewModelMessages = [];
                             const turnNewUIMessages = [];
-                            locals_js_1.locals.set(chatTurnNewUIMessagesKey, turnNewUIMessages);
-                            // A head-start handover deliberately resumes from an assistant
-                            // message it spliced in, so it isn't a no-op turn.
-                            let splicedHandoverPartial = false;
                             // ── Action handling ──────────────────────────────────────
                             // Actions arrive on the same input stream but with
                             // trigger === "action". They are NOT turns — only
@@ -4953,25 +4468,16 @@ function chatAgent(options) {
                             // an action, return a `StreamTextResult` (auto-piped),
                             // string, or UIMessage from `onAction`. Turn counter
                             // does not advance.
-                            let actionResult = undefined;
-                            /** Set when `onAction` returned `chat.turn()`: the turn block below runs. */
-                            let actionTurn = false;
-                            /**
-                             * Whether this action changed the conversation, by rolling history
-                             * back or by streaming a response. Drives the single snapshot write
-                             * at the end — an action never reaches the turn-complete path that
-                             * normally does it.
-                             */
-                            let actionChangedHistory = false;
+                            let actionStreamResult = undefined;
                             if (isAction) {
                                 // Parse and validate the action payload
                                 const parsedAction = parseAction
                                     ? await parseAction(currentWirePayload.action)
                                     : currentWirePayload.action;
                                 // Hydrate messages from backend if configured
-                                if (loadContextHook) {
+                                if (hydrateMessages) {
                                     const hydrated = await tracer_js_1.tracer.startActiveSpan("hydrateMessages()", async () => {
-                                        return loadContextHook({
+                                        return hydrateMessages({
                                             chatId: currentWirePayload.chatId,
                                             turn,
                                             trigger: "action",
@@ -4991,14 +4497,12 @@ function chatAgent(options) {
                                     });
                                     accumulatedUIMessages = [...hydrated];
                                     accumulatedMessages = await toModelMessages(hydrated);
-                                    laneCompacted = false;
-                                    laneInjections = [];
                                     locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                 }
                                 // Fire onAction — handler may mutate state via
                                 // `chat.history.*` and / or return a model response.
                                 if (onAction) {
-                                    actionResult = await tracer_js_1.tracer.startActiveSpan("onAction()", async () => {
+                                    actionStreamResult = await tracer_js_1.tracer.startActiveSpan("onAction()", async () => {
                                         return await onAction({
                                             action: parsedAction,
                                             chatId: currentWirePayload.chatId,
@@ -5023,10 +4527,7 @@ function chatAgent(options) {
                                         locals_js_1.locals.set(chatOverrideMessagesKey, undefined);
                                         accumulatedUIMessages = [...actionOverride];
                                         accumulatedMessages = await toModelMessages(actionOverride);
-                                        laneCompacted = false;
-                                        laneInjections = [];
                                         locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
-                                        actionChangedHistory = true;
                                     }
                                 }
                                 else {
@@ -5047,7 +4548,7 @@ function chatAgent(options) {
                                 // route handler's first-turn history to the hydrate hook as
                                 // incoming messages instead (gated on the pending handover).
                                 if (turn === 0 &&
-                                    loadContextHook &&
+                                    hydrateMessages &&
                                     cleanedUIMessages.length === 0 &&
                                     (locals_js_1.locals.get(chatHandoverPartialKey)?.length ?? 0) > 0 &&
                                     Array.isArray(payload.headStartMessages) &&
@@ -5077,7 +4578,7 @@ function chatAgent(options) {
                                         },
                                     }));
                                 }
-                                if (loadContextHook) {
+                                if (hydrateMessages) {
                                     // Snapshot the ids the accumulator knew BEFORE this
                                     // turn ran — used below to decide whether an
                                     // incoming wire message is genuinely new or just a
@@ -5095,7 +4596,7 @@ function chatAgent(options) {
                                     // was always true for `submit-message` is now true for every
                                     // trigger.
                                     const hydrated = await tracer_js_1.tracer.startActiveSpan("hydrateMessages()", async () => {
-                                        return loadContextHook({
+                                        return hydrateMessages({
                                             chatId: currentWirePayload.chatId,
                                             turn,
                                             trigger: currentWirePayload.trigger,
@@ -5134,8 +4635,6 @@ function chatAgent(options) {
                                     }
                                     accumulatedUIMessages = merged;
                                     accumulatedMessages = await toModelMessages(merged);
-                                    laneCompacted = false;
-                                    laneInjections = [];
                                     locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                     // Track new messages for onTurnComplete.newUIMessages.
                                     // Only push for genuinely new ids — HITL continuations
@@ -5181,8 +4680,6 @@ function chatAgent(options) {
                                             accumulatedUIMessages.pop();
                                         }
                                         accumulatedMessages = await toModelMessages(accumulatedUIMessages);
-                                        laneCompacted = false;
-                                        laneInjections = [];
                                     }
                                     else if (cleanedUIMessages.length > 0) {
                                         // Submit-message (and the special-cased
@@ -5199,7 +4696,6 @@ function chatAgent(options) {
                                         // where AI SDK regenerates the id (TRI-9137) still
                                         // applies via `rewriteIncomingIdViaToolCallMap`.
                                         let replaced = false;
-                                        const replacedPairs = [];
                                         for (const raw of cleanedUIMessages) {
                                             let incoming = raw;
                                             let idx = accumulatedUIMessages.findIndex((m) => m.id === incoming.id);
@@ -5211,9 +4707,7 @@ function chatAgent(options) {
                                                 }
                                             }
                                             if (idx !== -1) {
-                                                const previous = accumulatedUIMessages[idx];
-                                                accumulatedUIMessages[idx] = mergeIncomingIntoHydrated(previous, incoming);
-                                                replacedPairs.push({ previous, merged: accumulatedUIMessages[idx] });
+                                                accumulatedUIMessages[idx] = mergeIncomingIntoHydrated(accumulatedUIMessages[idx], incoming);
                                                 replaced = true;
                                             }
                                             else {
@@ -5223,19 +4717,9 @@ function chatAgent(options) {
                                             recordToolCallIdsFromMessage(incoming);
                                         }
                                         if (replaced) {
-                                            let inPlace = true;
-                                            for (const { previous, merged } of replacedPairs) {
-                                                if (!(await replaceModelRun(accumulatedMessages, previous, merged, 0))) {
-                                                    inPlace = false;
-                                                    break;
-                                                }
-                                            }
-                                            if (!inPlace) {
-                                                v3_1.logger.warn("chat.agent: replaced message not found at the model lane tail; reconverting the lane");
-                                                accumulatedMessages = await toModelMessages(accumulatedUIMessages);
-                                                laneCompacted = false;
-                                                laneInjections = [];
-                                            }
+                                            // Replacement changes structure — reconvert all model
+                                            // messages instead of appending.
+                                            accumulatedMessages = await toModelMessages(accumulatedUIMessages);
                                         }
                                         else {
                                             const incomingModelMessages = await toModelMessages(cleanedUIMessages);
@@ -5278,33 +4762,10 @@ function chatAgent(options) {
                                             messageId: locals_js_1.locals.get(chatHandoverMessageIdKey),
                                         });
                                         locals_js_1.locals.set(chatHandoverPartialKey, []); // consume once
-                                        splicedHandoverPartial = true;
                                     }
                                 }
                                 locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                             } // end if (trigger !== "action")
-                            // ── No-op turn ──────────────────────────────────────────
-                            //
-                            // A submit that added no new user message and leaves the model
-                            // chain ending on an assistant message has nothing to answer —
-                            // calling the model would prefill its own last reply. Keyed on
-                            // the model tail, so a `tool`-terminated chain (a merged tool
-                            // approval) still runs.
-                            const isNoOpTurn = !isAction &&
-                                !splicedHandoverPartial &&
-                                currentWirePayload.trigger === "submit-message" &&
-                                turnNewUIMessages.length === 0 &&
-                                accumulatedMessages[accumulatedMessages.length - 1]?.role === "assistant";
-                            if (isNoOpTurn) {
-                                msgSub?.off();
-                                v3_1.logger.warn("chat.agent: turn added no new user message; skipping the model", {
-                                    chatId: currentWirePayload.chatId,
-                                    messageId: currentWirePayload.messageId,
-                                });
-                                await writeTurnCompleteChunk(currentWirePayload.chatId);
-                                // Not a turn — don't consume an iteration.
-                                turn--;
-                            }
                             // ── Action result handling ──────────────────────────────
                             // For action turns, skip the turn machinery entirely.
                             // If `onAction` returned a stream / string / UIMessage,
@@ -5314,34 +4775,34 @@ function chatAgent(options) {
                             // The turn counter is decremented so the next iteration
                             // sees the same `turn` value — actions don't count.
                             if (isAction) {
-                                if (isActionTurn(actionResult)) {
-                                    // Persist the edit before the turn starts, so a turn that is
-                                    // cancelled or runs out of memory continues from the edited
-                                    // history rather than from the snapshot the edit replaced.
-                                    // The turn then does its own hooks, completion and snapshot.
-                                    if (actionChangedHistory) {
-                                        await writeSnapshotOutsideTurn("action", { turn, clientData });
+                                msgSub?.off();
+                                if ((locals_js_1.locals.get(chatPipeCountKey) ?? 0) === 0 &&
+                                    isUIMessageStreamable(actionStreamResult)) {
+                                    try {
+                                        const resolvedOptions = resolveUIMessageStreamOptions();
+                                        const uiStream = actionStreamResult.toUIMessageStream({
+                                            ...resolvedOptions,
+                                            generateMessageId: resolvedOptions.generateMessageId ?? ai_runtime_js_1.generateId,
+                                        });
+                                        await pipeChat(uiStream, {
+                                            signal: combinedSignal,
+                                            spanName: "stream response",
+                                        });
                                     }
-                                    actionTurn = true;
-                                }
-                                else if (actionResult !== undefined) {
-                                    throw new Error("chat.agent: onAction returned a value. An action is a state edit; to answer " +
-                                        "after the edit, return chat.turn() and a turn runs on the edited history. " +
-                                        "Returning a StreamTextResult, string or UIMessage is no longer supported.");
-                                }
-                                else {
-                                    msgSub?.off();
-                                    if (actionChangedHistory) {
-                                        await writeSnapshotOutsideTurn("action", { turn, clientData });
+                                    catch (error) {
+                                        if (error instanceof Error &&
+                                            error.name === "AbortError" &&
+                                            runSignal.aborted) {
+                                            return "exit";
+                                        }
+                                        throw error;
                                     }
-                                    await writeTurnCompleteChunk(currentWirePayload.chatId);
-                                    // Don't consume a turn iteration — actions aren't turns.
-                                    turn--;
                                 }
+                                await writeTurnCompleteChunk(currentWirePayload.chatId);
+                                // Don't consume a turn iteration — actions aren't turns.
+                                turn--;
                             }
-                            // A no-op turn skips this block, and with it `followSessionPin`:
-                            // there is nothing to answer, so nothing to hand over.
-                            if ((!isAction || actionTurn) && !isNoOpTurn) {
+                            if (!isAction) {
                                 // Mint a scoped public access token once per turn, reused for
                                 // onChatStart, onTurnStart, onTurnComplete, and the turn-complete chunk.
                                 const currentRunId = ctx.run.id;
@@ -5430,8 +4891,6 @@ function chatAgent(options) {
                                             locals_js_1.locals.set(chatOverrideMessagesKey, undefined);
                                             accumulatedUIMessages = [...turnStartOverride];
                                             accumulatedMessages = await toModelMessages(turnStartOverride);
-                                            laneCompacted = false;
-                                            laneInjections = [];
                                             locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                         }
                                     }, {
@@ -5448,11 +4907,9 @@ function chatAgent(options) {
                                         },
                                     });
                                 }
-                                await followSessionPin(currentWirePayload.chatId, versionSkew);
                                 // chat.requestUpgrade() called in onTurnStart (or onValidateMessages) —
-                                // skip run() and hand over to a fresh run on the new version. The
-                                // successor picks the message up off session.in; the transport only
-                                // keeps reading.
+                                // skip run() and signal the transport to re-trigger the same message
+                                // on the new version.
                                 if (locals_js_1.locals.get(chatUpgradeRequestedKey)) {
                                     await writeUpgradeRequiredChunk();
                                     return "exit";
@@ -5492,12 +4949,7 @@ function chatAgent(options) {
                                     const lastAccumulated = accumulatedMessages[accumulatedMessages.length - 1];
                                     const bgQueue = locals_js_1.locals.get(chatBackgroundQueueKey);
                                     if (bgQueue && bgQueue.length > 0 && lastAccumulated?.role !== "tool") {
-                                        const injected = bgQueue.splice(0);
-                                        accumulatedMessages.push(...injected);
-                                        laneInjections.push({
-                                            afterId: accumulatedUIMessages.at(-1)?.id ?? "",
-                                            messages: injected,
-                                        });
+                                        accumulatedMessages.push(...bgQueue.splice(0));
                                     }
                                     if (isHeadStartFinalTurn) {
                                         // The synthesized partial UIMessage IS the response.
@@ -5516,9 +4968,6 @@ function chatAgent(options) {
                                         const preparedMessages = await applyPrepareMessages(accumulatedMessages, "run");
                                         runResult = await userRun({
                                             ...restWire,
-                                            // A turn requested by chat.turn() is not the action itself:
-                                            // a run() that short-circuits on "action" must still answer.
-                                            ...(actionTurn ? { trigger: "action-turn" } : {}),
                                             messages: preparedMessages,
                                             clientData,
                                             continuation,
@@ -5531,7 +4980,6 @@ function chatAgent(options) {
                                             signal: combinedSignal,
                                             cancelSignal,
                                             stopSignal,
-                                            streamText: createBoundStreamText(promptRegistry, agentSystem, agentCacheControl, agentSystemProviderOptions),
                                         });
                                     }
                                     // Auto-pipe if the run function returned a StreamTextResult or similar,
@@ -5645,21 +5093,7 @@ function chatAgent(options) {
                                 if (runOverride) {
                                     locals_js_1.locals.set(chatOverrideMessagesKey, undefined);
                                     accumulatedUIMessages = [...runOverride];
-                                    /**
-                                     * Steers the drain consumed are left out of the rebuild and
-                                     * appended by the reconciliation below instead, so the lane
-                                     * gets the form the model actually received rather than a
-                                     * reconversion of the UI message, and gets it once. A steer
-                                     * the edit removed is dropped from the pending list too, so
-                                     * the edit is honoured.
-                                     */
-                                    const overrideIds = new Set(runOverride.map((m) => m.id));
-                                    const pending = (locals_js_1.locals.get(chatPendingSteerKey) ?? []).filter((e) => overrideIds.has(e.ui.id));
-                                    locals_js_1.locals.set(chatPendingSteerKey, pending);
-                                    const pendingIds = new Set(pending.map((e) => e.ui.id));
-                                    accumulatedMessages = await toModelMessages(runOverride.filter((m) => !pendingIds.has(m.id)));
-                                    laneCompacted = false;
-                                    laneInjections = [];
+                                    accumulatedMessages = await toModelMessages(runOverride);
                                     locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                 }
                                 // Check if compaction set a model-only override (preserves UI messages).
@@ -5682,8 +5116,6 @@ function chatAgent(options) {
                                     accumulatedMessages = taskCompactionConfig?.compactModelMessages
                                         ? await taskCompactionConfig.compactModelMessages(compactEvent)
                                         : modelOnlyOverride;
-                                    laneCompacted = true;
-                                    laneInjections = [];
                                     // Apply UI messages: callback or default (preserve all)
                                     if (taskCompactionConfig?.compactUIMessages) {
                                         accumulatedUIMessages = (await taskCompactionConfig.compactUIMessages(compactEvent));
@@ -5691,15 +5123,6 @@ function chatAgent(options) {
                                 }
                                 // Determine if the user stopped generation this turn (not a full run cancel).
                                 const wasStopped = stopController.signal.aborted && !runSignal.aborted;
-                                // Give the model accumulator the steering messages the drain
-                                // consumed. Appended, never reconverted from the UI lane, so a
-                                // model-only compaction summary set just above survives; and done
-                                // before the response is appended so the order stays
-                                // steer-then-answer. Outside the `capturedResponseMessage`
-                                // branches below, so a turn that captured no response is covered.
-                                const steerTailThisTurn = reconcilePendingSteer({
-                                    turnNew: turnNewModelMessages,
-                                }).reduce((n, e) => n + e.model.length, 0) + reconcilePendingBackground();
                                 // Append the assistant's response (partial or complete) to the accumulator.
                                 // The onFinish callback fires even on abort/stop, so partial responses
                                 // from stopped generation are captured correctly.
@@ -5737,7 +5160,6 @@ function chatAgent(options) {
                                     const existingIdx = capturedResponseMessage.id
                                         ? accumulatedUIMessages.findIndex((m) => m.id === capturedResponseMessage.id)
                                         : -1;
-                                    const previousAtIdx = existingIdx !== -1 ? accumulatedUIMessages[existingIdx] : undefined;
                                     if (existingIdx !== -1) {
                                         accumulatedUIMessages[existingIdx] = capturedResponseMessage;
                                     }
@@ -5757,14 +5179,8 @@ function chatAgent(options) {
                                             stripProviderMetadata(capturedResponseMessage),
                                         ]);
                                         if (existingIdx !== -1) {
-                                            const ok = previousAtIdx !== undefined &&
-                                                (await replaceModelRun(accumulatedMessages, previousAtIdx, capturedResponseMessage, steerTailThisTurn));
-                                            if (!ok) {
-                                                v3_1.logger.warn("chat.agent: replaced response not found at the model lane tail; reconverting the lane");
-                                                accumulatedMessages = await toModelMessages(accumulatedUIMessages);
-                                                laneCompacted = false;
-                                                laneInjections = [];
-                                            }
+                                            // Reconvert all model messages since we replaced rather than appended
+                                            accumulatedMessages = await toModelMessages(accumulatedUIMessages);
                                         }
                                         else {
                                             accumulatedMessages.push(...responseModelMessages);
@@ -5872,8 +5288,6 @@ function chatAgent(options) {
                                                                 ],
                                                             },
                                                         ];
-                                                    laneCompacted = true;
-                                                    laneInjections = [];
                                                     // UI messages: callback or default (preserve all)
                                                     if (outerCompaction.compactUIMessages) {
                                                         accumulatedUIMessages = (await outerCompaction.compactUIMessages(outerCompactEvent));
@@ -5963,8 +5377,6 @@ function chatAgent(options) {
                                             locals_js_1.locals.set(chatOverrideMessagesKey, undefined);
                                             accumulatedUIMessages = [...override];
                                             accumulatedMessages = await toModelMessages(override);
-                                            laneCompacted = false;
-                                            laneInjections = [];
                                             locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                             // Update event so onTurnComplete sees compacted messages
                                             turnCompleteEvent.messages = accumulatedMessages;
@@ -6011,8 +5423,6 @@ function chatAgent(options) {
                                             locals_js_1.locals.set(chatOverrideMessagesKey, undefined);
                                             accumulatedUIMessages = [...turnCompleteOverride];
                                             accumulatedMessages = await toModelMessages(turnCompleteOverride);
-                                            laneCompacted = false;
-                                            laneInjections = [];
                                             locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
                                         }
                                     }, {
@@ -6066,18 +5476,13 @@ function chatAgent(options) {
                                 if (!hydrateMessages) {
                                     try {
                                         await tracer_js_1.tracer.startActiveSpan("snapshot.write", async () => {
-                                            lastSnapshotOutEventId =
-                                                turnCompleteResult?.lastEventId ?? lastSnapshotOutEventId;
-                                            await saveTranscript({
-                                                reason: "turn-complete",
+                                            const snapshotInCursor = chatInputRouter().resumeFloor();
+                                            await writeChatSnapshot(sessionIdForSnapshot, {
+                                                version: 1,
+                                                savedAt: Date.now(),
                                                 messages: accumulatedUIMessages,
-                                                turn,
-                                                trigger: storageTrigger(currentWirePayload.trigger),
-                                                clientData,
-                                                lastOutEventId: lastSnapshotOutEventId,
-                                                nonFinalIds: wasStopped && capturedResponseMessage?.id
-                                                    ? new Set([capturedResponseMessage.id])
-                                                    : undefined,
+                                                lastOutEventId: turnCompleteResult?.lastEventId,
+                                                lastInEventId: snapshotInCursor !== undefined ? String(snapshotInCursor) : undefined,
                                             });
                                         }, {
                                             attributes: {
@@ -6110,17 +5515,10 @@ function chatAgent(options) {
                                 currentWirePayload = bootInjectedQueue.shift();
                                 return "continue";
                             }
-                            // chat.requestUpgrade() was called — exit the loop; the handover
-                            // has already triggered a new run on the latest version.
+                            // chat.requestUpgrade() was called — exit the loop so the
+                            // transport triggers a new run on the latest version.
                             // chat.endRun() — same exit, no upgrade semantics.
-                            if (locals_js_1.locals.get(chatCloseRequestedKey)) {
-                                await performChatClose();
-                                return "exit";
-                            }
                             if (locals_js_1.locals.get(chatUpgradeRequestedKey) || locals_js_1.locals.get(chatEndRunRequestedKey)) {
-                                if (locals_js_1.locals.get(chatUpgradeRequestedKey)) {
-                                    await persistUpgradeHandoff();
-                                }
                                 return "exit";
                             }
                             // Wait for the next message — stay idle briefly, then suspend
@@ -6227,10 +5625,6 @@ function chatAgent(options) {
                             });
                             // Signal turn complete so the client knows this turn is done
                             errorTurnCompleteResult = await writeTurnCompleteChunk(currentWirePayload.chatId);
-                            // A later action's snapshot reuses this cursor, so it has to move
-                            // here too or that snapshot resumes from before the failed turn.
-                            lastSnapshotOutEventId =
-                                errorTurnCompleteResult?.lastEventId ?? lastSnapshotOutEventId;
                         }
                         catch {
                             // Best-effort — if stream write fails, let the run continue anyway
@@ -6275,47 +5669,15 @@ function chatAgent(options) {
                             : partialIdx === -1
                                 ? [...erroredUIMessages, partialResponse]
                                 : erroredUIMessages.map((m, i) => i === partialIdx ? partialResponse : m);
-                        /**
-                         * Seeded from the per-turn list, not just the wire message and the
-                         * partial, so a steering message the drain consumed is reported too.
-                         * An app persisting from `newUIMessages` would otherwise lose the
-                         * instruction whenever the turn it steered went on to fail.
-                         */
-                        const buildErroredNew = () => {
-                            const out = [];
-                            const addUnique = (m) => {
-                                if (m && !out.some((existing) => existing.id === m.id))
-                                    out.push(m);
-                            };
-                            addUnique(erroredWireMessage);
-                            for (const m of (locals_js_1.locals.get(chatTurnNewUIMessagesKey) ?? [])) {
-                                addUnique(m);
-                            }
-                            if (includePartial)
-                                addUnique(partialResponse);
-                            return out;
-                        };
-                        let erroredNewUIMessages = buildErroredNew();
+                        let erroredNewUIMessages = erroredWireMessage ? [erroredWireMessage] : [];
+                        if (includePartial) {
+                            erroredNewUIMessages.push(partialResponse);
+                        }
                         let erroredNewModelMessages = [];
-                        const reconciledSteer = reconcilePendingSteer();
-                        const backgroundTailThisTurn = reconcilePendingBackground();
                         if (!responseCommitted) {
                             try {
                                 if (erroredNewUIMessages.length > 0) {
-                                    /**
-                                     * Built in order from the recorded forms rather than by
-                                     * converting the UI list, so a steer appears in the delta as
-                                     * the model received it (what `prepare` produced), matching the
-                                     * lane. The wire message and partial are converted as before.
-                                     */
-                                    const steerModelById = new Map(reconciledSteer.map((e) => [e.ui.id, e.model]));
-                                    for (const m of erroredNewUIMessages) {
-                                        const recorded = steerModelById.get(m.id);
-                                        if (recorded)
-                                            erroredNewModelMessages.push(...recorded);
-                                        else
-                                            erroredNewModelMessages.push(...(await toModelMessages([stripProviderMetadata(m)])));
-                                    }
+                                    erroredNewModelMessages = await toModelMessages(erroredNewUIMessages.map((m) => stripProviderMetadata(m)));
                                 }
                                 if (erroredUIMessagesWithPartial !== accumulatedUIMessages) {
                                     if (partialIdx === -1) {
@@ -6323,14 +5685,7 @@ function chatAgent(options) {
                                         accumulatedMessages.push(...(await toModelMessages(appended.map((m) => stripProviderMetadata(m)))));
                                     }
                                     else {
-                                        const ok = await replaceModelRun(accumulatedMessages, erroredUIMessages[partialIdx], partialResponse, reconciledSteer.reduce((n, e) => n + e.model.length, 0) +
-                                            backgroundTailThisTurn);
-                                        if (!ok) {
-                                            v3_1.logger.warn("chat.agent: replaced partial not found at the model lane tail; reconverting the lane");
-                                            accumulatedMessages = await toModelMessages(erroredUIMessagesWithPartial);
-                                            laneCompacted = false;
-                                            laneInjections = [];
-                                        }
+                                        accumulatedMessages = await toModelMessages(erroredUIMessagesWithPartial);
                                     }
                                     accumulatedUIMessages = erroredUIMessagesWithPartial;
                                     locals_js_1.locals.set(chatCurrentUIMessagesKey, accumulatedUIMessages);
@@ -6339,7 +5694,7 @@ function chatAgent(options) {
                             catch {
                                 erroredNewModelMessages = [];
                                 erroredUIMessagesWithPartial = erroredUIMessages;
-                                erroredNewUIMessages = buildErroredNew().filter((m) => m !== partialResponse);
+                                erroredNewUIMessages = erroredWireMessage ? [erroredWireMessage] : [];
                             }
                         }
                         if (onTurnComplete) {
@@ -6393,14 +5748,13 @@ function chatAgent(options) {
                         // neither the snapshot nor the replayable `.in` tail.
                         if (!hydrateMessages) {
                             try {
-                                await saveTranscript({
-                                    reason: "turn-error",
+                                const errorSnapshotInCursor = chatInputRouter().resumeFloor();
+                                await writeChatSnapshot(sessionIdForSnapshot, {
+                                    version: 1,
+                                    savedAt: Date.now(),
                                     messages: erroredUIMessagesWithPartial,
-                                    turn,
-                                    trigger: storageTrigger(currentWirePayload.trigger),
-                                    clientData: turnClientData,
-                                    lastOutEventId: lastSnapshotOutEventId,
-                                    nonFinalIds: includePartial && partialResponse ? new Set([partialResponse.id]) : undefined,
+                                    lastOutEventId: errorTurnCompleteResult?.lastEventId,
+                                    lastInEventId: errorSnapshotInCursor !== undefined ? String(errorSnapshotInCursor) : undefined,
                                 });
                             }
                             catch (error) {
@@ -6410,15 +5764,8 @@ function chatAgent(options) {
                                 });
                             }
                         }
-                        if (locals_js_1.locals.get(chatCloseRequestedKey)) {
-                            await performChatClose();
-                            return;
-                        }
                         // chat.requestUpgrade() / chat.endRun() — exit after error turn too
                         if (locals_js_1.locals.get(chatUpgradeRequestedKey) || locals_js_1.locals.get(chatEndRunRequestedKey)) {
-                            if (locals_js_1.locals.get(chatUpgradeRequestedKey)) {
-                                await persistUpgradeHandoff();
-                            }
                             return;
                         }
                         // Drain remaining recovered turns before idling — a thrown
@@ -6441,12 +5788,6 @@ function chatAgent(options) {
                             return; // Timed out — end run gracefully
                         }
                         currentWirePayload = next.output;
-                        // Same close check the success path makes. Without it a close
-                        // record that lands after a failed turn is consumed as if it were
-                        // a turn payload, and the loop runs on against a closed session.
-                        if (currentWirePayload.trigger === "close") {
-                            return;
-                        }
                         // Continue to next iteration of the for loop
                     }
                     finally {
@@ -6455,11 +5796,6 @@ function chatAgent(options) {
                 }
             }
             finally {
-                // Safety net for a close requested on a path that exits without
-                // reaching one of the loop's close checks (a turn timeout, an OOM
-                // re-throw). `performChatClose` is idempotent, so the ordinary path
-                // having already run it costs nothing here.
-                await performChatClose();
                 // `stopSub` is registered post-preload so the close-during-preload
                 // early-return path may exit before it ever attached. Guard the
                 // cleanup so a missing subscription doesn't throw.
@@ -6797,22 +6133,15 @@ function isStopped() {
 // Version upgrade
 // ---------------------------------------------------------------------------
 /**
- * Hand the conversation over to another deployment.
- *
- * The handover happens immediately and server-side: a successor run is created
- * and picks the conversation up from `session.in`. The transport keeps reading
- * the same session output, so no client action is needed and nothing waits for
- * the next message.
- *
- * Without a target the session's pin is cleared, so the successor lands on the
- * latest deployed version; with `externalDeploymentId` the session is re-pinned
- * to that deployment.
+ * Request that the current run exits so the next message starts on the latest
+ * deployed version (via the standard continuation mechanism).
  *
  * When called from `onTurnStart` or `onValidateMessages`, `run()` is skipped
- * entirely and the successor answers the message that opened the turn.
+ * entirely — the run exits immediately and the transport re-triggers the
+ * same message on the new version.
  *
  * When called from `run()` or `chat.defer()`, the current turn completes
- * normally and the handover happens afterward.
+ * normally and the run exits afterward instead of waiting for the next message.
  *
  * Call from `onTurnStart`, `onValidateMessages`, `onChatResume`, `run()`,
  * or inside `chat.defer()`.
@@ -6832,37 +6161,8 @@ function isStopped() {
  * });
  * ```
  */
-function requestUpgrade(options) {
+function requestUpgrade() {
     locals_js_1.locals.set(chatUpgradeRequestedKey, true);
-    // Without a target the handoff clears the session's pin; with one it re-pins to that deployment.
-    const target = options?.externalDeploymentId?.trim();
-    if (target)
-        locals_js_1.locals.set(chatUpgradeExternalDeploymentIdKey, target);
-}
-/** @internal Requests a handoff when the session's pin no longer names this deployment. */
-async function followSessionPin(chatId, policy) {
-    if (!chatId) {
-        return;
-    }
-    const deployedExternalId = locals_js_1.locals.get(chatAgentRunContextKey)?.deployment?.externalId;
-    if (policy !== "hold" && !deployedExternalId) {
-        v3_1.logger.debug("chat.versionSkew: cannot follow the session pin", {
-            chatId,
-            reason: "the run context carries no deployment.externalId",
-        });
-    }
-    const target = await (0, chatVersionSkew_js_1.resolvePinToFollow)({
-        policy,
-        deployedExternalId,
-        upgradeAlreadyRequested: locals_js_1.locals.get(chatUpgradeRequestedKey) === true,
-        readPin: async () => (await sessions_js_1.sessions.retrieve(chatId, { retry: { maxAttempts: 2, randomize: true } }))
-            .triggerConfig,
-    });
-    if (!target) {
-        return;
-    }
-    v3_1.logger.info("chat.versionSkew: following the session pin", { chatId, target });
-    requestUpgrade({ externalDeploymentId: target });
 }
 /**
  * Hand off the current custom agent Session to a fresh run.
@@ -6901,31 +6201,20 @@ async function endAndContinue() {
     if ((locals_js_1.locals.get(chatActiveSessionIteratorsKey) ?? 0) > 0) {
         throw new Error("chat.endAndContinue() cannot be called while a chat.createSession() iterator is active. Close the iterator, then call chat.endAndContinue().");
     }
-    await performEndAndContinue({ reason: "continuation" });
+    await performEndAndContinue();
 }
 /** @internal Shared server handoff used by managed and custom agent loops. */
-async function performEndAndContinue(options) {
+async function performEndAndContinue() {
     const chatId = locals_js_1.locals.get(chatExternalIdKey);
     const callingRunId = locals_js_1.locals.get(chatAgentRunContextKey)?.run.id;
     if (!chatId || !callingRunId) {
         throw new Error("Cannot end and continue without an active chat agent run");
     }
-    const externalDeploymentId = options.externalDeploymentId;
     const apiClient = v3_1.apiClientManager.clientOrThrow();
-    const result = await apiClient.endAndContinueSession(chatId, {
+    await apiClient.endAndContinueSession(chatId, {
         callingRunId,
-        reason: options.reason,
-        ...(externalDeploymentId ? { externalDeploymentId } : {}),
+        reason: "upgrade",
     });
-    if (result?.pendingVersion !== true) {
-        return;
-    }
-    // The successor parked. Say so on `.out` while this run still can — the transport's
-    // subscription survives the swap, so the client learns without waiting for its next send.
-    const [error] = await (0, v3_1.tryCatch)(getChatSession().out.writeControl(v3_1.TRIGGER_CONTROL_SUBTYPE.PENDING_VERSION));
-    if (error) {
-        v3_1.logger.warn("could not signal a parked handoff", { chatId, error });
-    }
 }
 /**
  * Exit the run after the current turn completes, without waiting for the
@@ -6956,124 +6245,6 @@ async function performEndAndContinue(options) {
 function endRun() {
     locals_js_1.locals.set(chatEndRunRequestedKey, true);
 }
-/**
- * End the whole conversation, permanently. The session row is closed, further
- * appends are refused, and the run exits without scheduling a continuation.
- *
- * This is the session-level stop. {@link endRun} ends the current run and lets
- * the next message start a fresh one; `chat.close()` ends the session itself,
- * so there is no next message. Use it for a budget cap, a completed goal,
- * abuse detection, or a user signing out.
- *
- * In a `chat.agent`, call it from `run()`, `prepareStep`, or
- * `onBeforeTurnComplete`. Called mid-step, it aborts the in-flight
- * `streamText` the same way the stop signal does, so the partial response is
- * still captured and streamed. The turn then completes normally, a terminal
- * `session-closed` record carrying `reason` is written to the response stream,
- * and the loop exits.
- *
- * Prefer `onBeforeTurnComplete` over `onTurnComplete`. It carries the same
- * fields but still runs while the stream is open, so the closed state rides
- * out on the turn's final record. `onTurnComplete` runs after that record, so
- * a close decided there does not reach a reader that has already finished the
- * turn, and the user only finds out when their next message is refused.
- *
- * In a `chat.customAgent`, call it anywhere in your own loop. The close is
- * performed when `run()` returns, so it lands whether you break out of a
- * `chat.createSession` loop, return early, or hand-roll the loop entirely.
- *
- * Closing is one-way: a closed session cannot be reopened. Its transcript
- * stays readable.
- *
- * @example
- * ```ts
- * chat.agent({
- *   id: "budgeted-agent",
- *   onBeforeTurnComplete: async ({ usage }) => {
- *     if (await overBudget(usage)) {
- *       chat.close({ reason: "Monthly budget reached" });
- *     }
- *   },
- * });
- * ```
- */
-function close(options) {
-    if (!locals_js_1.locals.get(chatExternalIdKey)) {
-        throw new Error("chat.close() can only be called from inside a chat.agent() or chat.customAgent() run");
-    }
-    // Bound the reason once, here. It goes out on S2 record headers as well as
-    // the close API, and an oversized value would fail the turn-complete write
-    // that carries the turn boundary, costing the client far more than the
-    // reason text.
-    // Trailing high surrogate: the cut landed between the two halves of an
-    // astral character, and encoding the orphan to UTF-8 for a record header
-    // yields a replacement character. Drop it rather than ship mojibake.
-    const reason = options?.reason
-        ?.slice(0, CHAT_CLOSE_REASON_MAX_LENGTH)
-        .replace(/[\uD800-\uDBFF]$/, "");
-    locals_js_1.locals.set(chatCloseRequestedKey, reason ? { reason } : {});
-    // Mid-step call: unblock the in-flight streamText exactly like the stop
-    // signal, so the turn can reach its turn boundary instead of running the
-    // model out to completion after the decision to close has been made.
-    locals_js_1.locals.get(chatStopControllerKey)?.abort(reason ?? "closed");
-}
-/**
- * @internal Terminal close sequence, run once at whichever exit site observes
- * the close request. Writes the standalone `session-closed` record, then closes
- * the session row.
- *
- * The record lands after the turn's `turn-complete`, so a client reading that
- * turn's stream has already terminated on it and will not see this one. It is
- * there for a reconnect and for replay. What a live client reads is the
- * `session-closed` header stamped onto `turn-complete` itself by
- * `writeTurnCompleteChunk`, which fires whenever the close was decided before
- * the turn ended. A close decided from `onTurnComplete` is past that point, so
- * the client learns from the 409 on its next send.
- */
-async function performChatClose() {
-    const request = locals_js_1.locals.get(chatCloseRequestedKey);
-    if (!request || locals_js_1.locals.get(chatClosePerformedKey))
-        return;
-    const reason = request.reason;
-    // Two flags, not one. The record is a client-visible event and must not be
-    // written twice, but the row close is the part that actually ends the
-    // conversation: flagging it as done before it succeeds would let a transient
-    // failure leave the session open with no later call willing to retry.
-    if (!locals_js_1.locals.get(chatCloseRecordWrittenKey)) {
-        locals_js_1.locals.set(chatCloseRecordWrittenKey, true);
-        try {
-            const session = getChatSession();
-            await session.out.writeControl(v3_1.TRIGGER_CONTROL_SUBTYPE.SESSION_CLOSED, reason ? [[v3_1.SESSION_CLOSED_REASON_HEADER, reason]] : undefined);
-        }
-        catch (error) {
-            v3_1.logger.warn("chat.close: failed to write the session-closed record", {
-                error: error instanceof Error ? error.message : String(error),
-            });
-        }
-    }
-    const chatId = locals_js_1.locals.get(chatExternalIdKey);
-    if (!chatId)
-        return;
-    try {
-        await sessions_js_1.sessions.close(chatId, {
-            ...(reason ? { reason } : {}),
-            ...(locals_js_1.locals.get(chatAgentRunContextKey)?.run.id
-                ? { callingRunId: locals_js_1.locals.get(chatAgentRunContextKey).run.id }
-                : {}),
-        });
-        locals_js_1.locals.set(chatClosePerformedKey, true);
-    }
-    catch (error) {
-        // Deliberately NOT flagged as performed: the close API is idempotent, so a
-        // later exit site on this run gets to retry it. Losing every retry to a
-        // transient failure would leave the row open and the conversation alive.
-        // Non-fatal either way — the run still exits.
-        v3_1.logger.error("chat.close: failed to close the session", {
-            chatId,
-            error: error instanceof Error ? error.message : String(error),
-        });
-    }
-}
 // ---------------------------------------------------------------------------
 // Per-turn deferred work
 // ---------------------------------------------------------------------------
@@ -7135,18 +6306,9 @@ function chatDefer(promiseOrFn) {
  * ```
  */
 function injectBackgroundContext(messages) {
-    const systemBlocks = messages.filter((message) => message.role === "system");
-    const conversational = messages.filter((message) => message.role !== "system");
-    if (systemBlocks.length > 0) {
-        const instructions = locals_js_1.locals.get(chatInjectedInstructionsKey) ?? [];
-        instructions.push(...systemBlocks);
-        locals_js_1.locals.set(chatInjectedInstructionsKey, instructions);
-    }
-    if (conversational.length > 0) {
-        const queue = locals_js_1.locals.get(chatBackgroundQueueKey) ?? [];
-        queue.push(...conversational);
-        locals_js_1.locals.set(chatBackgroundQueueKey, queue);
-    }
+    const queue = locals_js_1.locals.get(chatBackgroundQueueKey) ?? [];
+    queue.push(...messages);
+    locals_js_1.locals.set(chatBackgroundQueueKey, queue);
 }
 // ---------------------------------------------------------------------------
 // Aborted message cleanup
@@ -7559,12 +6721,10 @@ class ChatMessageAccumulator {
         // a duplicate, mirroring the chat.agent accumulator.
         const existingIdx = this.uiMessages.findIndex((m) => m.id === response.id);
         if (existingIdx !== -1) {
-            const previous = this.uiMessages[existingIdx];
             this.uiMessages[existingIdx] = response;
             try {
-                if (!(await replaceModelRun(this.modelMessages, previous, response, 0))) {
-                    this.modelMessages = await toModelMessages(this.uiMessages.map((m) => stripProviderMetadata(m)));
-                }
+                // Reconvert all model messages since we replaced rather than appended.
+                this.modelMessages = await toModelMessages(this.uiMessages.map((m) => stripProviderMetadata(m)));
             }
             catch {
                 // Conversion failed — leave the existing model messages in place
@@ -7600,28 +6760,6 @@ class ChatMessageAccumulator {
         const modelMsgs = await toModelMessages([message]);
         this._steeringQueue.push({ uiMessage: message, modelMessages: modelMsgs });
     }
-    /**
-     * Record the messages a steering drain consumed.
-     *
-     * The drain only puts them in this step's prompt, so without this they
-     * shape one answer and then exist in neither lane: not in `uiMessages`,
-     * which is what an app persists from, and not in `modelMessages`, which is
-     * what every later turn sends.
-     *
-     * Both lanes are appended to. The model lane is never reconverted from the
-     * UI lane, because `compactIfNeeded` replaces it with a summary and leaves
-     * the UI lane whole: a reconversion would restore everything the summary
-     * replaced.
-     */
-    async absorbSteering(claimed, injected) {
-        const fresh = claimed.filter((m) => !this.uiMessages.some((e) => e.id === m.id));
-        if (fresh.length === 0)
-            return;
-        this.uiMessages.push(...fresh);
-        // Record what the model received. Only when the whole batch is new is
-        // `injected` known to describe exactly these messages.
-        this.modelMessages.push(...(injected && fresh.length === claimed.length ? injected : await toModelMessages(fresh)));
-    }
     /**
      * Get and clear unconsumed steering messages.
      */
@@ -7654,8 +6792,7 @@ class ChatMessageAccumulator {
             }
             // 2. Pending message injection
             if (pm && queue.length > 0) {
-                const { injected, claimed } = await drainSteeringQueue(pm, resultMessages ?? messages, steps, queue);
-                await this.absorbSteering(claimed, injected);
+                const injected = await drainSteeringQueue(pm, resultMessages ?? messages, steps, queue);
                 if (injected.length > 0) {
                     resultMessages = [...(resultMessages ?? messages), ...injected];
                 }
@@ -7844,7 +6981,7 @@ function trackActiveChatSessionIterator(iterator) {
  * ```
  */
 function createChatSession(payload, options) {
-    const { signal: runSignal, idleTimeoutInSeconds: sessionIdleTimeoutOpt, timeout = "1h", maxTurns = 100, compaction: sessionCompaction, pendingMessages: sessionPendingMessages, versionSkew: sessionVersionSkew, } = options;
+    const { signal: runSignal, idleTimeoutInSeconds: sessionIdleTimeoutOpt, timeout = "1h", maxTurns = 100, compaction: sessionCompaction, pendingMessages: sessionPendingMessages, } = options;
     const idleTimeoutInSeconds = sessionIdleTimeoutOpt ?? 30;
     return {
         [Symbol.asyncIterator]() {
@@ -7933,16 +7070,8 @@ function createChatSession(payload, options) {
                      * without suspending.
                      */
                     if (turn > 0) {
-                        if (locals_js_1.locals.get(chatCloseRequestedKey)) {
-                            await performChatClose();
-                            stop.cleanup();
-                            return { done: true, value: undefined };
-                        }
                         // chat.requestUpgrade() / chat.endRun() — exit before waiting
                         if (locals_js_1.locals.get(chatUpgradeRequestedKey) || locals_js_1.locals.get(chatEndRunRequestedKey)) {
-                            if (locals_js_1.locals.get(chatUpgradeRequestedKey)) {
-                                await persistUpgradeHandoff();
-                            }
                             stop.cleanup();
                             return { done: true, value: undefined };
                         }
@@ -8045,7 +7174,6 @@ function createChatSession(payload, options) {
                         }
                         accumulator.applyHandover(pendingHandoverSignal);
                     }
-                    await followSessionPin(currentPayload.chatId, sessionVersionSkew);
                     // chat.requestUpgrade() called before this turn — signal transport and exit
                     if (locals_js_1.locals.get(chatUpgradeRequestedKey)) {
                         await writeUpgradeRequiredChunk();
@@ -8262,8 +7390,7 @@ function createChatSession(payload, options) {
                                     }
                                 }
                                 if (sessionPendingMessages) {
-                                    const { injected, claimed } = await drainSteeringQueue(sessionPendingMessages, resultMessages ?? stepMsgs, steps, turnSteeringQueue);
-                                    await accumulator.absorbSteering(claimed, injected);
+                                    const injected = await drainSteeringQueue(sessionPendingMessages, resultMessages ?? stepMsgs, steps, turnSteeringQueue);
                                     if (injected.length > 0) {
                                         resultMessages = [...(resultMessages ?? stepMsgs), ...injected];
                                     }
@@ -8277,11 +7404,6 @@ function createChatSession(payload, options) {
                 async return() {
                     activeMsgSub?.off();
                     activeMsgSub = undefined;
-                    // Reached when the consumer leaves the `for await` early (`break`,
-                    // `return`, a throw). A `chat.close()` from the loop body would
-                    // otherwise be dropped: the exit that performs it lives in `next()`,
-                    // and `next()` is never called again.
-                    await performChatClose();
                     // `stop` only exists once next() has booted the iterator.
                     stop?.cleanup();
                     return { done: true, value: undefined };
@@ -8538,11 +7660,6 @@ function createChatStartSessionAction(taskId, options) {
         const maxAttempts = params.triggerConfig?.maxAttempts ?? options?.triggerConfig?.maxAttempts;
         const maxDuration = params.triggerConfig?.maxDuration ?? options?.triggerConfig?.maxDuration;
         const idleTimeoutInSeconds = params.triggerConfig?.idleTimeoutInSeconds ?? options?.triggerConfig?.idleTimeoutInSeconds;
-        // Only `undefined` means "not supplied": a per-call `null` (opt out) has to beat a pinning
-        // action default, which neither truthiness nor `??` would allow.
-        const externalDeploymentId = params.triggerConfig?.externalDeploymentId !== undefined
-            ? params.triggerConfig.externalDeploymentId
-            : options?.triggerConfig?.externalDeploymentId;
         const triggerConfig = {
             basePayload: {
                 messages: [],
@@ -8569,10 +7686,6 @@ function createChatStartSessionAction(taskId, options) {
                     lockToVersion: params.triggerConfig?.lockToVersion ?? options?.triggerConfig?.lockToVersion,
                 }
                 : {}),
-            ...(params.triggerConfig?.ttl !== undefined || options?.triggerConfig?.ttl !== undefined
-                ? { ttl: params.triggerConfig?.ttl ?? options?.triggerConfig?.ttl }
-                : {}),
-            ...(externalDeploymentId !== undefined ? { externalDeploymentId } : {}),
             ...(idleTimeoutInSeconds !== undefined ? { idleTimeoutInSeconds } : {}),
         };
         const startBody = {
@@ -8617,7 +7730,6 @@ function createChatStartSessionAction(taskId, options) {
             publicAccessToken,
             runId: created.runId,
             sessionId: created.id,
-            ...(created.pendingVersion ? { pendingVersion: true } : {}),
         };
     };
 }
@@ -8651,8 +7763,7 @@ async function callSessionsCreateWithOverride(args) {
     const init = {
         method: "POST",
         headers: overrideRequestHeaders(accessToken),
-        // This path bypasses `sessions.start`, so it resolves the pin itself.
-        body: JSON.stringify((0, externalDeploymentId_js_1.withResolvedExternalDeploymentId)(args.body)),
+        body: JSON.stringify(args.body),
     };
     const response = args.fetchOverride
         ? await args.fetchOverride(url, init, ctx)
@@ -8717,41 +7828,6 @@ async function mintPublicTokenWithOverride(args) {
         expirationTime: args.expirationTime,
     });
 }
-/**
- * Creates a server-side helper that reads a conversation from a transcript
- * storage, for rendering history before the chat connects. Works the same
- * for every storage, the platform default included, so the browser never
- * reads a store directly and the secret key stays on the server.
- *
- * Wrap it in a Next.js server action (or any server-side handler), scope it
- * to the authenticated user through `clientData`, and pass the result to
- * `useLoadTranscript` in the browser.
- *
- * @example
- * ```ts
- * // actions.ts
- * "use server";
- * import { chat, defaultStorage } from "@trigger.dev/sdk/ai";
- *
- * export const loadTranscript = chat.createLoadTranscriptAction(defaultStorage, { limit: 50 });
- * ```
- */
-function createChatLoadTranscriptAction(storage, options) {
-    return async (params) => {
-        if (!params.chatId) {
-            throw new Error("chat.createLoadTranscriptAction: params.chatId is required.");
-        }
-        if (options?.apiClient) {
-            const { apiClient, ...rest } = options;
-            return v3_1.apiClientManager.runWithConfig(apiClient, () => createChatLoadTranscriptAction(storage, rest)(params));
-        }
-        const limit = params.limit ?? options?.limit;
-        return storage.load({ chatId: params.chatId, clientData: params.clientData }, {
-            ...(limit !== undefined ? { limit } : {}),
-            ...(params.before !== undefined ? { before: params.before } : {}),
-        });
-    };
-}
 exports.chat = {
     /** Create a chat agent. See {@link chatAgent}. */
     agent: chatAgent,
@@ -8763,12 +7839,8 @@ exports.chat = {
     withClientData,
     /** Create a server-side helper for starting (or resuming) a Session for a chatId. See {@link createChatStartSessionAction}. */
     createStartSessionAction: createChatStartSessionAction,
-    /** Returns a server-side helper that reads a conversation from a transcript storage. */
-    createLoadTranscriptAction: createChatLoadTranscriptAction,
     /** Pipe a stream to the chat transport. See {@link pipeChat}. */
     pipe: pipeChat,
-    /** Return from `onAction` to run a turn on the edited history. See {@link chatTurn}. */
-    turn: chatTurn,
     /** Create a per-run typed local. See {@link chatLocal}. */
     local: chatLocal,
     /** Create a public access token for a chat task. See {@link createChatAccessToken}. */
@@ -8789,8 +7861,6 @@ exports.chat = {
     endAndContinue,
     /** Exit the run after the current turn completes, without any upgrade signal. See {@link endRun}. */
     endRun,
-    /** End the conversation permanently: close the session and exit the run. See {@link close}. */
-    close,
     /** Clean up aborted parts from a UIMessage. See {@link cleanupAbortedParts}. */
     cleanupAbortedParts,
     /** Register background work that runs in parallel with streaming. See {@link chatDefer}. */
@@ -8938,16 +8008,6 @@ async function writeTurnCompleteChunk(_chatId, publicAccessToken) {
     if (consumedCursor !== undefined) {
         extraHeaders.push([v3_1.SESSION_IN_CONSUMED_ID_HEADER, String(consumedCursor)]);
     }
-    // A close decided before this turn ended rides out on turn-complete. Readers
-    // terminate their stream on turn-complete, so a standalone record written
-    // after it only reaches a reconnect — this header is what a live client sees.
-    const pendingClose = locals_js_1.locals.get(chatCloseRequestedKey);
-    if (pendingClose) {
-        extraHeaders.push([v3_1.SESSION_CLOSED_HEADER, "true"]);
-        if (pendingClose.reason) {
-            extraHeaders.push([v3_1.SESSION_CLOSED_REASON_HEADER, pendingClose.reason]);
-        }
-    }
     const result = await session.out.writeControl(v3_1.TRIGGER_CONTROL_SUBTYPE.TURN_COMPLETE, extraHeaders);
     const T_N = result.lastEventId ? Number.parseInt(result.lastEventId, 10) : undefined;
     // 2. Trim back to the previous turn-complete, if we have one. Skipping on
@@ -9005,47 +8065,12 @@ async function writeTurnCompleteChunk(_chatId, publicAccessToken) {
  *
  * @internal
  */
-/**
- * Persists an upgrade requested after the turn has already run.
- *
- * The pre-turn sites reach {@link performEndAndContinue} through
- * {@link writeUpgradeRequiredChunk}, which is what clears (or re-points) the
- * session's stored `externalDeploymentId`. The post-turn exits had no such path,
- * so a `chat.requestUpgrade()` from `run()` or `chat.defer()` left the pin intact
- * and every continuation re-pinned to the deployment the agent asked to leave.
- *
- * No `upgrade-required` chunk is written here: the turn already produced its
- * answer, so there is nothing for a client to be told about.
- */
-async function persistUpgradeHandoff() {
-    const chatId = locals_js_1.locals.get(chatExternalIdKey);
-    const callingRunId = locals_js_1.locals.get(chatAgentRunContextKey)?.run.id;
-    if (!chatId || !callingRunId) {
-        return;
-    }
-    try {
-        await performEndAndContinue({
-            reason: "upgrade",
-            externalDeploymentId: locals_js_1.locals.get(chatUpgradeExternalDeploymentIdKey),
-        });
-    }
-    catch (error) {
-        v3_1.logger.warn("upgrade handoff failed; session keeps its current version pin", {
-            chatId,
-            callingRunId,
-            error,
-        });
-    }
-}
 async function writeUpgradeRequiredChunk() {
     const chatId = locals_js_1.locals.get(chatExternalIdKey);
     const callingRunId = locals_js_1.locals.get(chatAgentRunContextKey)?.run.id;
     if (chatId && callingRunId) {
         try {
-            await performEndAndContinue({
-                reason: "upgrade",
-                externalDeploymentId: locals_js_1.locals.get(chatUpgradeExternalDeploymentIdKey),
-            });
+            await performEndAndContinue();
         }
         catch (error) {
             // Non-fatal: the next `.in/append` re-triggers via the probe.