@juspay/neurolink 11.15.8 → 11.16.0

package/dist/providers/googleVertex/client.js

@@ -29,7 +29,9 @@ import { convertZodToJsonSchema, inlineJsonSchema, ensureNestedSchemaTypes, } fr
 import { createNativeThinkingConfig } from "../../utils/thinkingConfig.js";
 import { TimeoutError, raceWithAbort, withTimeout, } from "../../utils/async/index.js";
 import { parseTimeout } from "../../utils/timeout.js";
-import { appendStepText, buildAbortedTurnMessage, buildContextCapMessage, buildToolLoopCapMessage, buildTurnStalledMessage, buildTurnTimeoutMessage, buildWrapupNudgeText, createContextGuard, createTurnClock, extractThoughtSignature, isAbortError, mapGeminiFinishReason, prependConversationMessages, resolveTurnStopReason, DedupExecuteMap, } from "../googleNativeGemini3/index.js";
+import { appendStepText, buildAbortedTurnMessage, buildContextCapMessage, buildDedupedEngineTools, buildToolLoopCapMessage, buildTurnStalledMessage, buildTurnTimeoutMessage, buildWrapupNudgeText, createContextGuard, createTurnClock, extractThoughtSignature, isAbortError, mapGeminiFinishReason, prependConversationMessages, resolveTurnStopReason, DedupExecuteMap, } from "../googleNativeGemini3/index.js";
+import { createGeminiLoopAdapter } from "../../core/geminiLoopAdapter.js";
+import { runAgenticLoop } from "../../core/loopEngine.js";
 import { createStreamChannel } from "../../core/streamChannel.js";
 import { toNativeToolDeclarations } from "../../core/nativeToolFormat.js";
 import { getAvailableInputTokens, getContextWindowSize, } from "../../constants/contextWindows.js";
@@ -371,9 +373,19 @@ const hasGoogleCredentials = () => {
             process.env.GOOGLE_AUTH_PRIVATE_KEY));
 };
 // Create Anthropic-specific Vertex settings for native @anthropic-ai/vertex-sdk
-const createVertexAnthropicSettings = async (region, timeoutMs) => {
+const createVertexAnthropicSettings = async (region, timeoutMs, direct, baseURL) => {
     const location = region || getVertexLocation();
-    const project = getVertexProjectId();
+    // Express-style auth carries its own credentials, so the ADC-derived project
+    // is neither available nor needed; asking for it would throw before the
+    // request is ever built. It cannot be EMPTY either — the SDK rejects a
+    // falsy projectId outright ("No projectId was given and it could not be
+    // resolved from credentials") — so a configured project is used when there
+    // is one and a placeholder stands in otherwise. The value only ever appears
+    // in the request path, which an endpoint reached this way is expected to
+    // route on its own.
+    const project = direct
+        ? direct.projectId?.trim() || "express"
+        : getVertexProjectId();
     return {
         projectId: project,
         region: location,
@@ -382,6 +394,26 @@ const createVertexAnthropicSettings = async (region, timeoutMs) => {
         // bound (a 429 with retry-after: 8549 sleeps 2.4h per retry, invisible
         // to fallback orchestration). Retries are the orchestrator's job.
         maxRetries: 0,
+        // Outside the express branch too: an endpoint override is about WHERE the
+        // request goes, not how it is authenticated, so a caller using ADC against
+        // a gateway needs it just as much.
+        ...(baseURL ? { baseURL } : {}),
+        ...(direct
+            ? {
+                // The token goes on the request directly. `accessToken` on the SDK's
+                // own options looks like it should do this and does not — the client
+                // stores it and never reads it for auth, so prepareOptions() still
+                // awaits Application Default Credentials and the call fails with a
+                // credentials error that names nothing useful. `authClient` is the
+                // option the SDK actually consults.
+                authClient: {
+                    getRequestHeaders: async () => ({
+                        Authorization: `Bearer ${direct.apiKey}`,
+                    }),
+                    projectId: null,
+                },
+            }
+            : {}),
     };
 };
 // Helper function to determine if a model is an Anthropic model
@@ -1338,10 +1370,16 @@ export class GoogleVertexProvider extends BaseProvider {
         // Convert Vercel AI SDK tools to @google/genai FunctionDeclarations
         let tools;
         const executeMap = new DedupExecuteMap();
+        let declarations;
         if (options.tools &&
             Object.keys(options.tools).length > 0 &&
             !options.disableTools) {
             const declared = toNativeToolDeclarations(options.tools, "functionDeclarations");
+            // Kept, not discarded: the shared adapter needs originalNameMap to
+            // translate sanitized wire names back, and buildDedupedEngineTools
+            // reads executeMap through it — which is the DedupExecuteMap that makes
+            // an identical repeated call answer from cache instead of re-running.
+            declarations = declared;
             tools = declared.toolsConfig;
             for (const [name, execute] of declared.executeMap) {
                 executeMap.set(name, execute);
@@ -1489,12 +1527,8 @@ export class GoogleVertexProvider extends BaseProvider {
         // hook can persist actual tool outputs rather than the placeholder
         // "success" string used by flushPendingToolData's default fallback.
         const toolExecutions = [];
-        let step = 0;
         // Track structured output from final_result tool (when using final_result pattern)
         let finalResultStructuredOutput;
-        // Track failed tools to prevent infinite retry loops
-        // Key: tool name, Value: { count: retry attempts, lastError: error message }
-        const failedTools = new Map();
         // In-loop context guard: stop calling tools when the accumulated
         // conversation approaches the model's context window instead of stepping
         // into a provider "prompt too long" rejection mid-loop.
@@ -1545,131 +1579,64 @@ export class GoogleVertexProvider extends BaseProvider {
             internalAbort.abort();
         }
         let wasAborted = false;
-        // One retry per turn for MALFORMED_FUNCTION_CALL steps (see the retry
-        // block after the step drain).
-        let malformedRetryCount = 0;
         // Step-cap flags declared in the outer scope so the terminal block (also
         // inside the try) and the finishReason mapping (after the finally) can
         // both read them.
         let hitStepLimit = false;
         let synthesizedFinalAnswer = false;
+        // How many steps the ENGINE actually took, reported back from the per-step
+        // hook. The terminal block compares it against maxSteps, and counting hook
+        // invocations instead would drift by the number of malformed retries.
+        let stepsTaken = 0;
         try {
             // Agentic loop for tool calling
-            while (step < maxSteps) {
-                if (effectiveSignal.aborted) {
-                    wasAborted = true;
-                    break;
-                }
-                // Context guard: stop the tool loop before the accumulated
-                // conversation crosses the window threshold — synthesize from what
-                // we have instead of stepping into a provider rejection.
-                if (contextGuard.shouldStop()) {
-                    // Parity upgrade: try to RECLAIM budget and keep going before
-                    // falling back to the historic stop-only behaviour. Ending the turn
-                    // early is safe but throws away work the model was mid-way through;
-                    // dropping the oldest complete tool exchanges usually buys enough
-                    // room to finish. Only when reclaiming changes nothing do we stop.
-                    const reclaimed = reclaimVertexLoopContext(currentContents, modelName, contextGuard.projectedNextPromptTokens);
-                    if (reclaimed) {
-                        contextGuard.resetAfterReclaim();
-                    }
-                    else {
-                        hitContextLimit = true;
-                        logger.warn(`[GoogleVertex] Gemini turn stopped by the context guard: ` +
-                            `projected prompt ~${contextGuard.projectedNextPromptTokens} tokens ` +
-                            `>= threshold ${contextGuard.thresholdTokens} (step ${step}) — synthesizing a final answer.`);
-                        break;
-                    }
-                }
-                step++;
-                turnClock.noteProgress();
-                // Mid-turn discovery sync: search_tools may have hydrated new tools
-                // into the live record during the previous step — advertise them in
-                // this step's request instead of leaving them invisible until the
-                // next turn (TOOL_NOT_FOUND).
-                this.refreshGeminiToolDeclarations(options.tools, tools?.[0]?.functionDeclarations, executeMap, failedTools);
-                logger.debug(`[GoogleVertex] Native SDK step ${step}/${maxSteps}`);
-                try {
-                    const stream = await client.models.generateContentStream({
-                        model: modelName,
-                        contents: currentContents,
-                        config: { ...config, abortSignal: effectiveSignal },
-                    });
-                    const stepFunctionCalls = [];
-                    // Capture raw response parts including thoughtSignature
-                    const rawResponseParts = [];
-                    // This step's own finish reason (vs the cross-step
-                    // lastFinishReason) — drives the single MALFORMED_FUNCTION_CALL
-                    // retry below.
-                    let stepFinishReason;
-                    // Per-step usage trackers for WRITE-THROUGH accumulation: within a
-                    // step's chunk stream the counts are latest-wins (promptTokenCount
-                    // arrives once in the final chunk; candidates/thoughts counts are
-                    // cumulative across chunks), so each chunk folds only the DELTA
-                    // over this step's previous value into the turn totals. The totals
-                    // are therefore correct at every point mid-drain — a step killed
-                    // mid-stream (abort / turn deadline / stall watchdog) still counts
-                    // the billed tokens it already reported.
-                    // The drain now lives in collectVertexStreamChunks. The hooks below
-                    // are the two couplings it had to this loop, plus the per-chunk
-                    // usage deltas — those must stay per-chunk so a step killed
-                    // mid-stream still bills what it reported.
-                    const collected = await collectVertexStreamChunks(stream, {
-                        push: (chunk) => {
-                            if (chunk.content) {
-                                incrementalTextChunks.push(chunk.content);
-                            }
-                        },
-                    }, {
-                        onProgress: () => turnClock.noteProgress(),
-                        onUsage: (input, output) => contextGuard.noteUsage(input, output),
-                        onUsageDelta: (counter, delta) => {
-                            if (counter === "input") {
-                                totalInputTokens += delta;
-                            }
-                            else if (counter === "output") {
-                                totalOutputTokens += delta;
-                            }
-                            else if (counter === "cacheRead") {
-                                totalCacheReadTokens += delta;
+            // The turn runs on the shared engine. The step cap, tool dispatch, the
+            // failure breaker, per-step usage accumulation, the single malformed
+            // retry and the pre-first-chunk provider retry all live there now; what
+            // stays here is everything the engine has no opinion about — the turn
+            // clock, the context guard, conversation-memory storage, the wrap-up
+            // nudge, and the terminal block below.
+            const engineAdapter = createGeminiLoopAdapter({
+                providerLabel: "GoogleVertex",
+                maxSteps,
+                // Ported verbatim: the same DEFAULT_TOOL_MAX_RETRIES threshold, plus
+                // the two rules this loop has always had and the engine did not.
+                toolFailureBreaker: {
+                    maxRetries: DEFAULT_TOOL_MAX_RETRIES,
+                    // Strikes are CONSECUTIVE here: a clean result clears the count, so
+                    // an argument-dependent soft error cannot accumulate its way to
+                    // disabling a tool that works.
+                    consecutive: true,
+                    // A result that reports failure without throwing — an MCP isError
+                    // payload, a proxy-blocked call resolving with { error } — counts
+                    // toward the breaker exactly as a throw does.
+                    classifyResultFailure: (output) => extractToolFailureText(output) ?? undefined,
+                },
+                liveTools: options.tools ?? {},
+                ...(declarations ? { declarations } : {}),
+                ...(useFinalResultTool
+                    ? {
+                        finalResultToolName: "final_result",
+                        onTerminalResult: (text) => {
+                            try {
+                                finalResultStructuredOutput = JSON.parse(text);
                             }
-                            else {
-                                totalReasoningTokens += delta;
+                            catch {
+                                /* the caller's coercion layer repairs a partial payload */
                             }
                         },
-                    });
-                    rawResponseParts.push(...collected.rawResponseParts);
-                    stepFunctionCalls.push(...collected.stepFunctionCalls);
-                    if (collected.finishReason) {
-                        stepFinishReason = collected.finishReason;
-                        lastFinishReason = collected.finishReason;
                     }
-                    // Extract text from raw parts after stream completes
-                    // This avoids SDK warning about non-text parts (thoughtSignature, functionCall)
-                    const stepText = rawResponseParts
-                        .filter((part) => typeof part.text === "string")
-                        .map((part) => part.text)
-                        .join("");
-                    // MALFORMED_FUNCTION_CALL is usually a transient formatting failure
-                    // (the model emitted an unparseable call): retry the step ONCE with
-                    // a corrective note instead of hard-ending the turn with empty
-                    // content — automated alert-RCA turns were dying at step 2-4 on
-                    // this, mislabeled as step-cap exits.
-                    if (stepFunctionCalls.length === 0 &&
-                        !stepText &&
-                        stepFinishReason === "MALFORMED_FUNCTION_CALL" &&
-                        malformedRetryCount < 1 &&
-                        !effectiveSignal.aborted) {
-                        malformedRetryCount++;
-                        logger.warn(`[GoogleVertex] Model returned MALFORMED_FUNCTION_CALL at step ${step}/${maxSteps}; retrying once with a corrective note.`);
-                        this.emitTurnEvent({ phase: "malformed-retry", step, maxSteps });
-                        if (rawResponseParts.length > 0) {
-                            currentContents.push({
-                                role: "model",
-                                parts: rawResponseParts,
-                            });
-                        }
-                        currentContents.push({
+                    : {}),
+                enableMalformedRetry: true,
+                buildMalformedRetryNote: (conversation, retriedStep) => {
+                    this.emitTurnEvent({
+                        phase: "malformed-retry",
+                        step: retriedStep + 1,
+                        maxSteps,
+                    });
+                    return [
+                        ...conversation,
+                        {
                             role: "user",
                             parts: [
                                 {
@@ -1678,310 +1645,206 @@ export class GoogleVertexProvider extends BaseProvider {
                                         "or answer in plain text.",
                                 },
                             ],
-                        });
-                        continue;
-                    }
-                    // If no function calls, we're done
-                    if (stepFunctionCalls.length === 0) {
-                        finalText = stepText;
-                        break;
-                    }
-                    // Check for final_result tool call - this is our structured output pattern
-                    if (useFinalResultTool) {
-                        const finalResultCall = stepFunctionCalls.find((call) => call.name === "final_result");
-                        if (finalResultCall) {
-                            // Extract the structured output from final_result arguments
-                            finalResultStructuredOutput = finalResultCall.args;
-                            logger.debug("[GoogleVertex] Received final_result tool call with structured output (stream)", {
-                                outputKeys: Object.keys(finalResultStructuredOutput),
-                            });
-                            // Return the structured output as JSON text
-                            finalText = JSON.stringify(finalResultStructuredOutput);
-                            break;
-                        }
-                    }
-                    // Execute function calls
-                    logger.debug(`[GoogleVertex] Executing ${stepFunctionCalls.length} function calls`);
-                    // Add model response with ALL parts (including thoughtSignature) to history
-                    // This preserves the thought_signature which is required for Gemini 3 multi-turn tool calling
-                    currentContents.push({
-                        role: "model",
-                        parts: rawResponseParts.length > 0
-                            ? rawResponseParts
-                            : stepFunctionCalls.map((fc) => ({
-                                functionCall: fc,
-                            })),
-                    });
-                    // Execute each function and collect responses (plus an optional
-                    // trailing wrap-up nudge text part).
-                    const functionResponses = [];
-                    // Per-step bookkeeping for conversation-memory storage.
-                    const stepStorageCalls = [];
-                    const stepStorageResults = [];
-                    // Note: tool:start / tool:end events are emitted by ToolsManager's
-                    // wrapped `execute` (see ToolsManager.ts:355) — no inline emit needed.
-                    for (const call of stepFunctionCalls) {
-                        // Honor a deadline/stall/caller abort BETWEEN tool executions —
-                        // without this check a multi-tool step keeps executing its whole
-                        // batch (up to N × toolTimeoutMs past the deadline) before the
-                        // while-top check finally breaks.
-                        if (effectiveSignal.aborted) {
-                            wasAborted = true;
-                            break;
-                        }
-                        allToolCalls.push({ toolName: call.name, args: call.args });
-                        stepStorageCalls.push({ toolName: call.name, args: call.args });
-                        // Check if this tool has already exceeded retry limit
-                        const failedInfo = failedTools.get(call.name);
-                        if (failedInfo && failedInfo.count >= DEFAULT_TOOL_MAX_RETRIES) {
-                            logger.warn(`[GoogleVertex] Tool "${call.name}" has exceeded retry limit (${DEFAULT_TOOL_MAX_RETRIES}), skipping execution`);
-                            const errorPayload = {
-                                error: `TOOL_PERMANENTLY_FAILED: The tool "${call.name}" has failed ${failedInfo.count} times and will not be retried. Last error: ${failedInfo.lastError}. Please proceed without using this tool or inform the user that this functionality is unavailable.`,
-                                status: "permanently_failed",
-                                do_not_retry: true,
-                            };
-                            functionResponses.push({
-                                functionResponse: {
-                                    name: call.name,
-                                    response: errorPayload,
-                                },
-                            });
-                            toolExecutions.push({
-                                name: call.name,
-                                input: call.args,
-                                output: errorPayload,
-                            });
-                            stepStorageResults.push({
-                                toolName: call.name,
-                                output: errorPayload,
-                            });
-                            continue;
+                        },
+                    ];
+                },
+                // The turn clock's per-chunk ping and the context guard's per-step
+                // prompt size both ride the drain, which is why this loop keeps its
+                // own collector rather than the adapter's default.
+                collectStep: (stream, channel) => collectVertexStreamChunks(stream, channel, {
+                    onProgress: () => turnClock.noteProgress(),
+                    onUsage: (input, output) => contextGuard.noteUsage(input, output),
+                    onUsageDelta: (counter, delta) => {
+                        if (counter === "input") {
+                            totalInputTokens += delta;
                         }
-                        let execute = executeMap.get(call.name);
-                        if (!execute) {
-                            // Snapshot miss: the tool may have been hydrated into the live
-                            // record by search_tools within this very step batch, or the
-                            // model called a deferred catalog tool directly by name.
-                            execute = this.resolveGeminiToolOnMiss(call.name, options.tools, tools?.[0]?.functionDeclarations, executeMap, failedTools);
+                        else if (counter === "output") {
+                            totalOutputTokens += delta;
                         }
-                        if (execute) {
-                            try {
-                                // AI SDK Tool execute requires (args, options) - provide minimal options
-                                const toolOptions = {
-                                    toolCallId: `${call.name}-${Date.now()}`,
-                                    messages: [],
-                                    abortSignal: effectiveSignal,
-                                };
-                                turnClock.noteProgress();
-                                // Bound the execute() await — a wedged tool costs one step
-                                // (error tool_result), not the whole turn — and race it
-                                // against the turn's abort so a deadline/caller abort is
-                                // observed IMMEDIATELY instead of after the tool settles.
-                                const result = await withTimeout(raceWithAbort(Promise.resolve(execute(call.args, toolOptions)), effectiveSignal), toolExecTimeoutMs, `Tool "${call.name}" execution timed out after ${toolExecTimeoutMs}ms`);
-                                turnClock.noteProgress();
-                                // Error-shaped success (MCP isError / { error } payloads —
-                                // e.g. proxy-blocked tools) counts toward the breaker too:
-                                // these fail without throwing, and only counting throws lets
-                                // the model grind on a blocked tool for the whole budget.
-                                const resultErrorText = extractToolFailureText(result);
-                                if (resultErrorText) {
-                                    const info = failedTools.get(call.name) || {
-                                        count: 0,
-                                        lastError: "",
-                                    };
-                                    info.count++;
-                                    info.lastError = resultErrorText;
-                                    failedTools.set(call.name, info);
-                                }
-                                else {
-                                    // Genuinely consecutive: a success clears the strike count
-                                    // (argument-dependent soft errors — file-not-found on
-                                    // different paths — must not disable a working tool).
-                                    failedTools.delete(call.name);
-                                }
-                                toolExecutions.push({
-                                    name: call.name,
-                                    input: call.args,
-                                    output: result,
-                                });
-                                functionResponses.push({
-                                    functionResponse: { name: call.name, response: { result } },
-                                });
-                                stepStorageResults.push({
-                                    toolName: call.name,
-                                    output: result,
-                                });
-                            }
-                            catch (error) {
-                                // An abort during tool execution ends the turn gracefully — it
-                                // must NOT be recorded as a spurious tool failure. Both checks
-                                // matter: effectiveSignal.aborted catches an abort WE triggered
-                                // (even if the throw isn't abort-shaped); isAbortError(error)
-                                // catches an abort-shaped throw (e.g. a tool raising AbortError)
-                                // even when the signal itself never fired.
-                                if (effectiveSignal.aborted || isAbortError(error)) {
-                                    wasAborted = true;
-                                    break;
-                                }
-                                turnClock.noteProgress();
-                                if (error instanceof TimeoutError) {
-                                    this.emitTurnEvent({
-                                        phase: "tool-timeout",
-                                        step,
-                                        maxSteps,
-                                        toolName: call.name,
-                                    });
-                                }
-                                const errorMessage = error instanceof Error ? error.message : "Unknown error";
-                                // Track this failure
-                                const currentFailInfo = failedTools.get(call.name) || {
-                                    count: 0,
-                                    lastError: "",
-                                };
-                                currentFailInfo.count++;
-                                currentFailInfo.lastError = errorMessage;
-                                failedTools.set(call.name, currentFailInfo);
-                                logger.warn(`[GoogleVertex] Tool "${call.name}" failed (attempt ${currentFailInfo.count}/${DEFAULT_TOOL_MAX_RETRIES}): ${errorMessage}`);
-                                // Determine if this is a permanent failure
-                                const isPermanentFailure = currentFailInfo.count >= DEFAULT_TOOL_MAX_RETRIES;
-                                const errorPayload = {
-                                    error: isPermanentFailure
-                                        ? `TOOL_PERMANENTLY_FAILED: The tool "${call.name}" has failed ${currentFailInfo.count} times with error: ${errorMessage}. This tool will not be retried. Please proceed without using this tool or inform the user that this functionality is unavailable.`
-                                        : `TOOL_EXECUTION_ERROR: ${errorMessage}. Retry attempt ${currentFailInfo.count}/${DEFAULT_TOOL_MAX_RETRIES}.`,
-                                    status: isPermanentFailure ? "permanently_failed" : "failed",
-                                    do_not_retry: isPermanentFailure,
-                                    retry_count: currentFailInfo.count,
-                                    max_retries: DEFAULT_TOOL_MAX_RETRIES,
-                                };
-                                functionResponses.push({
-                                    functionResponse: {
-                                        name: call.name,
-                                        response: errorPayload,
-                                    },
-                                });
-                                toolExecutions.push({
-                                    name: call.name,
-                                    input: call.args,
-                                    output: errorPayload,
-                                });
-                                stepStorageResults.push({
-                                    toolName: call.name,
-                                    output: errorPayload,
-                                });
-                            }
+                        else if (counter === "cacheRead") {
+                            totalCacheReadTokens += delta;
                         }
                         else {
-                            // Tool not found is a permanent error. Count it toward the
-                            // breaker too (parity with the Anthropic loops) — a model that
-                            // ignores the do_not_retry hint and keeps calling a
-                            // hallucinated/stale tool name must not burn the whole step
-                            // budget on TOOL_NOT_FOUND round-trips.
-                            const errorPayload = {
-                                error: `TOOL_NOT_FOUND: The tool "${call.name}" does not exist. Do not attempt to call this tool again.`,
-                                status: "permanently_failed",
-                                do_not_retry: true,
-                            };
-                            const notFoundInfo = failedTools.get(call.name) || {
-                                count: 0,
-                                lastError: "",
-                            };
-                            notFoundInfo.count++;
-                            notFoundInfo.lastError = errorPayload.error;
-                            failedTools.set(call.name, notFoundInfo);
-                            functionResponses.push({
-                                functionResponse: {
-                                    name: call.name,
-                                    response: errorPayload,
-                                },
-                            });
-                            toolExecutions.push({
-                                name: call.name,
-                                input: call.args,
-                                output: errorPayload,
-                            });
-                            stepStorageResults.push({
-                                toolName: call.name,
-                                output: errorPayload,
-                            });
+                            totalReasoningTokens += delta;
                         }
+                    },
+                }),
+                planReclaim: (conversation) => {
+                    if (!contextGuard.shouldStop()) {
+                        return undefined;
                     }
-                    // An abort inside the tool-exec loop only breaks that inner for-loop.
-                    // Break the while too so no further model call is issued and control
-                    // reaches the terminal step-cap handling below.
-                    if (wasAborted) {
-                        break;
+                    // Try to RECLAIM budget and keep going before falling back to the
+                    // historic stop-only behaviour. Ending the turn early is safe but
+                    // throws away work the model was mid-way through; dropping the
+                    // oldest complete tool exchanges usually buys enough room to finish.
+                    const working = [...conversation];
+                    if (reclaimVertexLoopContext(working, modelName, contextGuard.projectedNextPromptTokens)) {
+                        contextGuard.resetAfterReclaim();
+                        return { conversation: working };
                     }
-                    // Persist this step's tool calls/results into conversation memory.
-                    // Without this, tool_call / tool_result rows never reach Redis and
-                    // the chat-history UI loses every tool invocation.
-                    //
-                    // `thoughtSignature` rides as a sibling on the first call of the
-                    // step — Gemini 3 needs it to match thinking patterns when the
-                    // conversation is replayed on the next turn.
-                    if (stepStorageCalls.length > 0 || stepStorageResults.length > 0) {
-                        const stepThoughtSig = extractThoughtSignature(rawResponseParts);
-                        withTimeout(this.handleToolExecutionStorage(stepStorageCalls.map((c, i) => ({
-                            ...c,
-                            ...(i === 0 && stepThoughtSig
-                                ? { thoughtSignature: stepThoughtSig }
-                                : {}),
-                            stepIndex: step,
-                        })), stepStorageResults.map((r) => ({ ...r, stepIndex: step })), options, new Date()), TOOL_STORAGE_TIMEOUT_MS, "tool storage write timed out").catch((error) => {
-                            logger.warn("[GoogleVertex] Failed to store native Gemini stream tool executions", {
-                                error: error instanceof Error ? error.message : String(error),
+                    hitContextLimit = true;
+                    logger.warn(`[GoogleVertex] Gemini turn stopped by the context guard: ` +
+                        `projected prompt ~${contextGuard.projectedNextPromptTokens} tokens ` +
+                        `>= threshold ${contextGuard.thresholdTokens} — synthesizing a final answer.`);
+                    return { stop: true };
+                },
+                buildRequest: (conversation) => ({
+                    model: modelName,
+                    contents: conversation,
+                    config: { ...config, ...(tools ? { tools } : {}) },
+                }),
+                sendStep: async (request, signal) => {
+                    turnClock.noteProgress();
+                    const built = request;
+                    return client.models.generateContentStream({
+                        model: built.model,
+                        contents: built.contents,
+                        config: { ...(built.config ?? {}), abortSignal: signal },
+                    });
+                },
+            });
+            // Wrapped rather than configured: these fire once PER STEP, and
+            // buildToolResultMessages is the only hook that runs per step with
+            // exactly that step's results. Reading them off the turn's final result
+            // would batch every step into one late write and lose the per-step
+            // thought signature.
+            const adapter = {
+                ...engineAdapter,
+                // Counted HERE, not in buildToolResultMessages: this runs once per
+                // step exactly as the old `step++` at the top of the loop did,
+                // malformed retries included. Counting in the tool-result hook would
+                // skip the final text-only step and quietly report one step fewer in
+                // `stepsUsed`, which is a public field on the result.
+                buildStepRequest: (conversation, step) => {
+                    stepsTaken = step + 1;
+                    return engineAdapter.buildStepRequest(conversation, step);
+                },
+                buildToolResultMessages: (conversation, stepResult, toolResults, engineStep) => {
+                    const next = engineAdapter.buildToolResultMessages(conversation, stepResult, toolResults, engineStep);
+                    // Time-budget wrap-up nudge (twin of the Anthropic loops' soft step
+                    // nudge): with the turn deadline approaching, tell the model to
+                    // consolidate. Rides as a trailing text part on the tool-response
+                    // user turn.
+                    if (turnClock.shouldNudgeWrapup()) {
+                        const last = next[next.length - 1];
+                        if (last && Array.isArray(last.parts)) {
+                            last.parts.push({
+                                text: buildWrapupNudgeText(useFinalResultTool),
                             });
-                        });
+                        }
                     }
-                    // Time-budget wrap-up nudge (twin of the Anthropic loops' soft
-                    // step nudge): with the turn deadline approaching, tell the model
-                    // to consolidate. Rides as a trailing text part on the
-                    // tool-response user turn.
-                    if (turnClock.shouldNudgeWrapup()) {
-                        functionResponses.push({
-                            text: buildWrapupNudgeText(useFinalResultTool),
+                    // Persist this step's tool calls/results into conversation memory.
+                    // Without this, tool_call / tool_result rows never reach Redis and
+                    // the chat-history UI loses every tool invocation. `thoughtSignature`
+                    // rides as a sibling on the first call of the step — Gemini 3 needs
+                    // it to match thinking patterns when the conversation is replayed.
+                    const stepThoughtSig = extractThoughtSignature(stepResult.raw.rawResponseParts);
+                    withTimeout(this.handleToolExecutionStorage(toolResults.map((result, index) => ({
+                        toolName: result.name,
+                        args: result.args,
+                        ...(index === 0 && stepThoughtSig
+                            ? { thoughtSignature: stepThoughtSig }
+                            : {}),
+                        stepIndex: engineStep + 1,
+                    })), toolResults.map((result) => ({
+                        toolName: result.name,
+                        output: result.output,
+                        stepIndex: engineStep + 1,
+                    })), options, new Date()), TOOL_STORAGE_TIMEOUT_MS, "tool storage write timed out").catch((error) => {
+                        logger.warn("[GoogleVertex] Failed to store native Gemini stream tool executions", {
+                            error: error instanceof Error ? error.message : String(error),
                         });
-                    }
-                    // The @google/genai SDK only accepts "user" and "model" as valid
-                    // roles in contents — function/tool responses must use role: "user"
-                    // (matching the SDK's automaticFunctionCalling implementation and
-                    // the Google AI Studio path). Sending role: "function" was causing
-                    // native Vertex Gemini tool loops to be silently rejected by the
-                    // request validator.
-                    currentContents.push({
-                        role: "user",
-                        parts: functionResponses,
                     });
                     // Project this step's growth for the context guard: the appended
                     // tool results ride the next prompt (Gemini reports usage per call,
                     // but only for content it has already seen).
                     try {
-                        contextGuard.noteAppendedChars(JSON.stringify(functionResponses).length);
+                        const appended = next[next.length - 1];
+                        contextGuard.noteAppendedChars(JSON.stringify(appended?.parts ?? []).length);
                     }
                     catch {
                         /* estimation is best-effort — never break the loop */
                     }
-                }
-                catch (error) {
-                    // A mid-drain abort surfaces as an AbortError from the `for await`.
-                    // Break gracefully into the terminal block instead of re-throwing
-                    // (a re-throw would route the caller's abort into a second unbounded
-                    // fallback stream()). Dual check as with the inner catch:
-                    // effectiveSignal.aborted (a signal we tripped) OR isAbortError(error)
-                    // (an abort-shaped throw) — either means "stop", not a real failure.
-                    if (effectiveSignal.aborted || isAbortError(error)) {
-                        wasAborted = true;
-                        break;
+                    return next;
+                },
+            };
+            const { stream: engineStream, resultPromise } = runAgenticLoop(adapter, 
+            // A concrete parts array widens to the engine's `unknown[]` on its
+            // own; only the direction back needs an assertion.
+            currentContents, {
+                tools: buildDedupedEngineTools(declarations, options.tools, {
+                    toolTimeoutMs: toolExecTimeoutMs,
+                    abortSignal: effectiveSignal,
+                    onProgress: () => turnClock.noteProgress(),
+                }),
+                abortSignal: effectiveSignal,
+            });
+            // Collected, NOT forwarded to the consumer here. This loop replays the
+            // gathered text after the turn rather than streaming it live, and a
+            // characterization case pins exactly that — pushing to the consumer
+            // channel from inside the pump would make the turn stream live and break
+            // it.
+            const pump = (async () => {
+                for await (const chunk of engineStream) {
+                    if (chunk.content) {
+                        incrementalTextChunks.push(chunk.content);
                     }
-                    logger.error("[GoogleVertex] Native SDK error", error);
-                    throw this.handleProviderError(error);
                 }
+            })();
+            let engineResult;
+            let turnFailure;
+            try {
+                engineResult = await resultPromise;
+            }
+            catch (error) {
+                turnFailure = error;
+            }
+            // Drained unconditionally and tolerantly. When the turn ends by abort the
+            // channel rejects too, and re-awaiting a settled rejection here would
+            // rethrow the very error the branch below has already decided to absorb —
+            // which is what turned both turn-clock cases into failures instead of
+            // clean deadline exits.
+            await pump.catch(() => { });
+            if (turnFailure !== undefined) {
+                // A mid-drain abort surfaces as an AbortError. End gracefully into the
+                // terminal block instead of re-throwing — a re-throw would route the
+                // caller's abort into a second unbounded fallback stream().
+                if (effectiveSignal.aborted || isAbortError(turnFailure)) {
+                    wasAborted = true;
+                }
+                else {
+                    logger.error("[GoogleVertex] Native SDK error", turnFailure);
+                    throw this.handleProviderError(turnFailure);
+                }
+            }
+            if (engineResult) {
+                finalText = engineResult.text;
+                lastFinishReason = engineResult.rawStopReason ?? lastFinishReason;
+                for (const call of engineResult.toolCalls) {
+                    allToolCalls.push({ toolName: call.name, args: call.args });
+                }
+                for (const execution of engineResult.toolExecutions) {
+                    toolExecutions.push({
+                        name: execution.name,
+                        input: execution.input,
+                        output: execution.output,
+                    });
+                }
+                // Replace in place: `currentContents` is a const the terminal block
+                // and the synth call both read.
+                currentContents.length = 0;
+                currentContents.push(...engineResult.conversation);
+            }
+            if (effectiveSignal.aborted) {
+                wasAborted = true;
             }
             // Handle maxSteps termination / abort — the loop exited because the step
             // cap was reached (or the turn was aborted) while the model was still
             // calling tools. Surface a real answer instead of the canned placeholder
             // (Bug 1) and a meaningful finishReason (Bug 2).
-            if (!finalText && (step >= maxSteps || wasAborted || hitContextLimit)) {
-                hitStepLimit = step >= maxSteps && !wasAborted;
+            if (!finalText &&
+                (stepsTaken >= maxSteps || wasAborted || hitContextLimit)) {
+                hitStepLimit = stepsTaken >= maxSteps && !wasAborted;
                 const toolCallCount = allToolCalls.filter((tc) => tc.toolName !== "final_result").length;
                 // The consumer receives text via `incrementalTextChunks`; any text the
                 // model emitted across steps is already preserved there. Only produce a
@@ -2067,7 +1930,7 @@ export class GoogleVertexProvider extends BaseProvider {
         if (stopReason !== "completed") {
             this.emitTurnEvent({
                 phase: stopReason,
-                step,
+                step: stepsTaken,
                 maxSteps,
                 toolCallCount: allToolCalls.filter((tc) => tc.toolName !== "final_result").length,
                 elapsedMs: turnClock.elapsedMs(),
@@ -2146,7 +2009,7 @@ export class GoogleVertexProvider extends BaseProvider {
                 totalToolExecutions: externalToolCalls.length,
                 stopReason,
                 rawFinishReason: lastFinishReason,
-                stepsUsed: step,
+                stepsUsed: stepsTaken,
             },
         };
         // Add structured output if final_result tool was used
@@ -2311,8 +2174,12 @@ export class GoogleVertexProvider extends BaseProvider {
         // Convert Vercel AI SDK tools to @google/genai FunctionDeclarations
         let tools;
         const executeMap = new DedupExecuteMap();
+        let declarations;
         if (Object.keys(combinedTools).length > 0) {
             const declared = toNativeToolDeclarations(combinedTools, "functionDeclarations");
+            // Kept for the shared adapter: originalNameMap for name translation,
+            // executeMap for the per-turn dedup wrapper.
+            declarations = declared;
             tools = declared.toolsConfig;
             for (const [name, execute] of declared.executeMap) {
                 executeMap.set(name, execute);
@@ -2456,12 +2323,8 @@ export class GoogleVertexProvider extends BaseProvider {
         let lastFinishReason;
         const allToolCalls = [];
         const toolExecutions = [];
-        let step = 0;
         // Track structured output from final_result tool (when using final_result pattern)
         let finalResultStructuredOutput;
-        // Track failed tools to prevent infinite retry loops
-        // Key: tool name, Value: { count: retry attempts, lastError: error message }
-        const failedTools = new Map();
         // In-loop context guard: stop calling tools when the accumulated
         // conversation approaches the model's context window instead of stepping
         // into a provider "prompt too long" rejection mid-loop.
@@ -2505,125 +2368,63 @@ export class GoogleVertexProvider extends BaseProvider {
             internalAbort.abort();
         }
         let wasAborted = false;
-        // One retry per turn for MALFORMED_FUNCTION_CALL steps (see the retry
-        // block after the step drain).
-        let malformedRetryCount = 0;
         // Step-cap flags declared in the outer scope so the terminal block (also
         // inside the try) and the finishReason mapping (after the finally) can
         // both read them.
         let hitStepLimit = false;
         let synthesizedFinalAnswer = false;
+        // Steps the ENGINE took, reported back from the per-step hook; counting
+        // hook invocations would drift by the number of malformed retries.
+        let stepsTaken = 0;
         try {
             // Agentic loop for tool calling
-            while (step < maxSteps) {
-                if (effectiveSignal.aborted) {
-                    wasAborted = true;
-                    break;
-                }
-                // Context guard: stop the tool loop before the accumulated
-                // conversation crosses the window threshold — synthesize from what
-                // we have instead of stepping into a provider rejection.
-                if (contextGuard.shouldStop()) {
-                    // Parity upgrade: try to RECLAIM budget and keep going before
-                    // falling back to the historic stop-only behaviour. Ending the turn
-                    // early is safe but throws away work the model was mid-way through;
-                    // dropping the oldest complete tool exchanges usually buys enough
-                    // room to finish. Only when reclaiming changes nothing do we stop.
-                    const reclaimed = reclaimVertexLoopContext(currentContents, modelName, contextGuard.projectedNextPromptTokens);
-                    if (reclaimed) {
-                        contextGuard.resetAfterReclaim();
-                    }
-                    else {
-                        hitContextLimit = true;
-                        logger.warn(`[GoogleVertex] Gemini turn stopped by the context guard: ` +
-                            `projected prompt ~${contextGuard.projectedNextPromptTokens} tokens ` +
-                            `>= threshold ${contextGuard.thresholdTokens} (step ${step}) — synthesizing a final answer.`);
-                        break;
-                    }
-                }
-                step++;
-                turnClock.noteProgress();
-                // Mid-turn discovery sync — see the stream twin.
-                this.refreshGeminiToolDeclarations(combinedTools, tools?.[0]?.functionDeclarations, executeMap, failedTools);
-                logger.debug(`[GoogleVertex] Native SDK generate step ${step}/${maxSteps}`);
-                try {
-                    // Use generateContentStream and collect all chunks (same as GoogleAIStudio)
-                    const stream = await client.models.generateContentStream({
-                        model: modelName,
-                        contents: currentContents,
-                        config: { ...config, abortSignal: effectiveSignal },
-                    });
-                    const stepFunctionCalls = [];
-                    // Capture raw response parts including thoughtSignature
-                    const rawResponseParts = [];
-                    // This step's own finish reason (vs the cross-step
-                    // lastFinishReason) — drives the single MALFORMED_FUNCTION_CALL
-                    // retry below.
-                    let stepFinishReason;
-                    // Per-step usage trackers for WRITE-THROUGH accumulation: within a
-                    // step's chunk stream the counts are latest-wins (promptTokenCount
-                    // arrives once in the final chunk; candidates/thoughts counts are
-                    // cumulative across chunks), so each chunk folds only the DELTA
-                    // over this step's previous value into the turn totals. The totals
-                    // are therefore correct at every point mid-drain — a step killed
-                    // mid-stream (abort / turn deadline / stall watchdog) still counts
-                    // the billed tokens it already reported.
-                    // Same collector as the streaming twin. generate() returns one
-                    // result rather than streaming, so the channel is a no-op — the
-                    // stream loop uses it to fill incrementalTextChunks for replay, and
-                    // there is nothing to replay here. Everything else is identical,
-                    // including the per-chunk usage deltas that keep the turn totals
-                    // correct for a step killed mid-stream.
-                    const collected = await collectVertexStreamChunks(stream, { push: () => { } }, {
-                        onProgress: () => turnClock.noteProgress(),
-                        onUsage: (input, output) => contextGuard.noteUsage(input, output),
-                        onUsageDelta: (counter, delta) => {
-                            if (counter === "input") {
-                                totalInputTokens += delta;
-                            }
-                            else if (counter === "output") {
-                                totalOutputTokens += delta;
-                            }
-                            else if (counter === "cacheRead") {
-                                totalCacheReadTokens += delta;
+            // The turn runs on the shared engine. The step cap, tool dispatch, the
+            // failure breaker, per-step usage accumulation, the single malformed
+            // retry and the pre-first-chunk provider retry all live there now; what
+            // stays here is everything the engine has no opinion about — the turn
+            // clock, the context guard, conversation-memory storage, the wrap-up
+            // nudge, and the terminal block below.
+            const engineAdapter = createGeminiLoopAdapter({
+                providerLabel: "GoogleVertex",
+                maxSteps,
+                // Ported verbatim: the same DEFAULT_TOOL_MAX_RETRIES threshold, plus
+                // the two rules this loop has always had and the engine did not.
+                toolFailureBreaker: {
+                    maxRetries: DEFAULT_TOOL_MAX_RETRIES,
+                    // Strikes are CONSECUTIVE here: a clean result clears the count, so
+                    // an argument-dependent soft error cannot accumulate its way to
+                    // disabling a tool that works.
+                    consecutive: true,
+                    // A result that reports failure without throwing — an MCP isError
+                    // payload, a proxy-blocked call resolving with { error } — counts
+                    // toward the breaker exactly as a throw does.
+                    classifyResultFailure: (output) => extractToolFailureText(output) ?? undefined,
+                },
+                liveTools: options.tools ?? {},
+                ...(declarations ? { declarations } : {}),
+                ...(useFinalResultTool
+                    ? {
+                        finalResultToolName: "final_result",
+                        onTerminalResult: (text) => {
+                            try {
+                                finalResultStructuredOutput = JSON.parse(text);
                             }
-                            else {
-                                totalReasoningTokens += delta;
+                            catch {
+                                /* the caller's coercion layer repairs a partial payload */
                             }
                         },
-                    });
-                    rawResponseParts.push(...collected.rawResponseParts);
-                    stepFunctionCalls.push(...collected.stepFunctionCalls);
-                    if (collected.finishReason) {
-                        stepFinishReason = collected.finishReason;
-                        lastFinishReason = collected.finishReason;
                     }
-                    // Extract text from raw parts after stream completes
-                    // This avoids SDK warning about non-text parts (thoughtSignature, functionCall)
-                    const stepText = rawResponseParts
-                        .filter((part) => typeof part.text === "string")
-                        .map((part) => part.text)
-                        .join("");
-                    // MALFORMED_FUNCTION_CALL is usually a transient formatting failure
-                    // (the model emitted an unparseable call): retry the step ONCE with
-                    // a corrective note instead of hard-ending the turn with empty
-                    // content — automated alert-RCA turns were dying at step 2-4 on
-                    // this, mislabeled as step-cap exits.
-                    if (stepFunctionCalls.length === 0 &&
-                        !stepText &&
-                        stepFinishReason === "MALFORMED_FUNCTION_CALL" &&
-                        malformedRetryCount < 1 &&
-                        !effectiveSignal.aborted) {
-                        malformedRetryCount++;
-                        logger.warn(`[GoogleVertex] Model returned MALFORMED_FUNCTION_CALL at step ${step}/${maxSteps}; retrying once with a corrective note.`);
-                        this.emitTurnEvent({ phase: "malformed-retry", step, maxSteps });
-                        if (rawResponseParts.length > 0) {
-                            currentContents.push({
-                                role: "model",
-                                parts: rawResponseParts,
-                            });
-                        }
-                        currentContents.push({
+                    : {}),
+                enableMalformedRetry: true,
+                buildMalformedRetryNote: (conversation, retriedStep) => {
+                    this.emitTurnEvent({
+                        phase: "malformed-retry",
+                        step: retriedStep + 1,
+                        maxSteps,
+                    });
+                    return [
+                        ...conversation,
+                        {
                             role: "user",
                             parts: [
                                 {
@@ -2632,307 +2433,195 @@ export class GoogleVertexProvider extends BaseProvider {
                                         "or answer in plain text.",
                                 },
                             ],
-                        });
-                        continue;
-                    }
-                    // If no function calls, we're done
-                    if (stepFunctionCalls.length === 0) {
-                        finalText = stepText;
-                        break;
-                    }
-                    // Check for final_result tool call - this is our structured output pattern
-                    if (useFinalResultTool) {
-                        const finalResultCall = stepFunctionCalls.find((call) => call.name === "final_result");
-                        if (finalResultCall) {
-                            // Extract the structured output from final_result arguments
-                            finalResultStructuredOutput = finalResultCall.args;
-                            logger.debug("[GoogleVertex] Received final_result tool call with structured output (generate)", {
-                                outputKeys: Object.keys(finalResultStructuredOutput),
-                            });
-                            // Return the structured output as JSON text
-                            finalText = JSON.stringify(finalResultStructuredOutput);
-                            break;
-                        }
-                    }
-                    // Accumulate non-empty step text across steps so the
-                    // maxSteps-exhaustion exit can surface the prose the model produced
-                    // instead of a canned placeholder (Bug 1). Mirrors the Vertex-Claude
-                    // loop's text accumulation.
-                    accumulatedText = appendStepText(accumulatedText, stepText);
-                    // Execute function calls
-                    logger.debug(`[GoogleVertex] Generate executing ${stepFunctionCalls.length} function calls`);
-                    // Add model response with ALL parts (including thoughtSignature) to history
-                    // This preserves the thought_signature which is required for Gemini 3 multi-turn tool calling
-                    currentContents.push({
-                        role: "model",
-                        parts: rawResponseParts.length > 0
-                            ? rawResponseParts
-                            : stepFunctionCalls.map((fc) => ({
-                                functionCall: fc,
-                            })),
-                    });
-                    // Execute each function and collect responses (plus an optional
-                    // trailing wrap-up nudge text part).
-                    const functionResponses = [];
-                    const toolCallsBefore = allToolCalls.length;
-                    const toolExecsBefore = toolExecutions.length;
-                    // Note: tool:start / tool:end events are emitted by ToolsManager's
-                    // wrapped `execute` (see ToolsManager.ts:355) — no inline emit needed.
-                    for (const call of stepFunctionCalls) {
-                        // Honor a deadline/stall/caller abort BETWEEN tool executions —
-                        // without this check a multi-tool step keeps executing its whole
-                        // batch (up to N × toolTimeoutMs past the deadline) before the
-                        // while-top check finally breaks.
-                        if (effectiveSignal.aborted) {
-                            wasAborted = true;
-                            break;
-                        }
-                        allToolCalls.push({ toolName: call.name, args: call.args });
-                        // Check if this tool has already exceeded retry limit
-                        const failedInfo = failedTools.get(call.name);
-                        if (failedInfo && failedInfo.count >= DEFAULT_TOOL_MAX_RETRIES) {
-                            logger.warn(`[GoogleVertex] Tool "${call.name}" has exceeded retry limit (${DEFAULT_TOOL_MAX_RETRIES}), skipping execution`);
-                            const errorOutput = {
-                                error: `TOOL_PERMANENTLY_FAILED: The tool "${call.name}" has failed ${failedInfo.count} times and will not be retried. Last error: ${failedInfo.lastError}. Please proceed without using this tool or inform the user that this functionality is unavailable.`,
-                                status: "permanently_failed",
-                                do_not_retry: true,
-                            };
-                            toolExecutions.push({
-                                name: call.name,
-                                input: call.args,
-                                output: errorOutput,
-                            });
-                            functionResponses.push({
-                                functionResponse: {
-                                    name: call.name,
-                                    response: errorOutput,
-                                },
-                            });
-                            continue;
+                        },
+                    ];
+                },
+                // The turn clock's per-chunk ping and the context guard's per-step
+                // prompt size both ride the drain, which is why this loop keeps its
+                // own collector rather than the adapter's default.
+                collectStep: (stream, channel) => collectVertexStreamChunks(stream, channel, {
+                    onProgress: () => turnClock.noteProgress(),
+                    onUsage: (input, output) => contextGuard.noteUsage(input, output),
+                    onUsageDelta: (counter, delta) => {
+                        if (counter === "input") {
+                            totalInputTokens += delta;
                         }
-                        let execute = executeMap.get(call.name);
-                        if (!execute) {
-                            // Snapshot miss — see the stream twin.
-                            execute = this.resolveGeminiToolOnMiss(call.name, combinedTools, tools?.[0]?.functionDeclarations, executeMap, failedTools);
+                        else if (counter === "output") {
+                            totalOutputTokens += delta;
                         }
-                        if (execute) {
-                            try {
-                                // AI SDK Tool execute requires (args, options) - provide minimal options
-                                const toolOptions = {
-                                    toolCallId: `${call.name}-${Date.now()}`,
-                                    messages: [],
-                                    abortSignal: effectiveSignal,
-                                };
-                                turnClock.noteProgress();
-                                // Bound the execute() await — a wedged tool costs one step
-                                // (error tool_result), not the whole turn.
-                                const execResult = await withTimeout(raceWithAbort(Promise.resolve(execute(call.args, toolOptions)), effectiveSignal), toolExecTimeoutMs, `Tool "${call.name}" execution timed out after ${toolExecTimeoutMs}ms`);
-                                turnClock.noteProgress();
-                                // Error-shaped success (MCP isError / { error } payloads —
-                                // e.g. proxy-blocked tools) counts toward the breaker too:
-                                // these fail without throwing, and only counting throws lets
-                                // the model grind on a blocked tool for the whole budget.
-                                const resultErrorText = extractToolFailureText(execResult);
-                                if (resultErrorText) {
-                                    const info = failedTools.get(call.name) || {
-                                        count: 0,
-                                        lastError: "",
-                                    };
-                                    info.count++;
-                                    info.lastError = resultErrorText;
-                                    failedTools.set(call.name, info);
-                                }
-                                else {
-                                    // Genuinely consecutive: a success clears the strike count
-                                    // (argument-dependent soft errors — file-not-found on
-                                    // different paths — must not disable a working tool).
-                                    failedTools.delete(call.name);
-                                }
-                                // Track execution
-                                toolExecutions.push({
-                                    name: call.name,
-                                    input: call.args,
-                                    output: execResult,
-                                });
-                                functionResponses.push({
-                                    functionResponse: {
-                                        name: call.name,
-                                        response: { result: execResult },
-                                    },
-                                });
-                            }
-                            catch (error) {
-                                // An abort during tool execution ends the turn gracefully — it
-                                // must NOT be recorded as a spurious tool failure. Both checks
-                                // matter: effectiveSignal.aborted catches an abort WE triggered
-                                // (even if the throw isn't abort-shaped); isAbortError(error)
-                                // catches an abort-shaped throw (e.g. a tool raising AbortError)
-                                // even when the signal itself never fired.
-                                if (effectiveSignal.aborted || isAbortError(error)) {
-                                    wasAborted = true;
-                                    break;
-                                }
-                                turnClock.noteProgress();
-                                if (error instanceof TimeoutError) {
-                                    this.emitTurnEvent({
-                                        phase: "tool-timeout",
-                                        step,
-                                        maxSteps,
-                                        toolName: call.name,
-                                    });
-                                }
-                                const errorMessage = error instanceof Error ? error.message : "Unknown error";
-                                // Track this failure
-                                const currentFailInfo = failedTools.get(call.name) || {
-                                    count: 0,
-                                    lastError: "",
-                                };
-                                currentFailInfo.count++;
-                                currentFailInfo.lastError = errorMessage;
-                                failedTools.set(call.name, currentFailInfo);
-                                logger.warn(`[GoogleVertex] Tool "${call.name}" failed (attempt ${currentFailInfo.count}/${DEFAULT_TOOL_MAX_RETRIES}): ${errorMessage}`);
-                                // Determine if this is a permanent failure
-                                const isPermanentFailure = currentFailInfo.count >= DEFAULT_TOOL_MAX_RETRIES;
-                                const errorOutput = {
-                                    error: isPermanentFailure
-                                        ? `TOOL_PERMANENTLY_FAILED: The tool "${call.name}" has failed ${currentFailInfo.count} times with error: ${errorMessage}. This tool will not be retried. Please proceed without using this tool or inform the user that this functionality is unavailable.`
-                                        : `TOOL_EXECUTION_ERROR: ${errorMessage}. Retry attempt ${currentFailInfo.count}/${DEFAULT_TOOL_MAX_RETRIES}.`,
-                                    status: isPermanentFailure ? "permanently_failed" : "failed",
-                                    do_not_retry: isPermanentFailure,
-                                    retry_count: currentFailInfo.count,
-                                    max_retries: DEFAULT_TOOL_MAX_RETRIES,
-                                };
-                                toolExecutions.push({
-                                    name: call.name,
-                                    input: call.args,
-                                    output: errorOutput,
-                                });
-                                functionResponses.push({
-                                    functionResponse: {
-                                        name: call.name,
-                                        response: errorOutput,
-                                    },
-                                });
-                            }
+                        else if (counter === "cacheRead") {
+                            totalCacheReadTokens += delta;
                         }
                         else {
-                            // Tool not found is a permanent error
-                            const errorOutput = {
-                                error: `TOOL_NOT_FOUND: The tool "${call.name}" does not exist. Do not attempt to call this tool again.`,
-                                status: "permanently_failed",
-                                do_not_retry: true,
-                            };
-                            toolExecutions.push({
-                                name: call.name,
-                                input: call.args,
-                                output: errorOutput,
-                            });
-                            functionResponses.push({
-                                functionResponse: {
-                                    name: call.name,
-                                    response: errorOutput,
-                                },
-                            });
+                            totalReasoningTokens += delta;
                         }
+                    },
+                }),
+                planReclaim: (conversation) => {
+                    if (!contextGuard.shouldStop()) {
+                        return undefined;
                     }
-                    // An abort inside the tool-exec loop only breaks that inner for-loop.
-                    // Break the while too so no further model call is issued and control
-                    // reaches the terminal step-cap handling below.
-                    if (wasAborted) {
-                        break;
+                    // Try to RECLAIM budget and keep going before falling back to the
+                    // historic stop-only behaviour. Ending the turn early is safe but
+                    // throws away work the model was mid-way through; dropping the
+                    // oldest complete tool exchanges usually buys enough room to finish.
+                    const working = [...conversation];
+                    if (reclaimVertexLoopContext(working, modelName, contextGuard.projectedNextPromptTokens)) {
+                        contextGuard.resetAfterReclaim();
+                        return { conversation: working };
                     }
-                    // Persist this step's tool calls/results into conversation memory.
-                    // Without this, tool_call / tool_result rows never reach Redis and
-                    // the chat-history UI loses every tool invocation. The first call
-                    // of the step carries the step's `thoughtSignature` so Gemini 3 can
-                    // match thinking patterns on replay.
-                    const stepToolCalls = allToolCalls.slice(toolCallsBefore);
-                    const stepToolExecs = toolExecutions.slice(toolExecsBefore);
-                    if (stepToolCalls.length > 0 || stepToolExecs.length > 0) {
-                        const stepThoughtSig = extractThoughtSignature(rawResponseParts);
-                        withTimeout(this.handleToolExecutionStorage(stepToolCalls.map((tc, i) => ({
-                            toolName: tc.toolName,
-                            args: tc.args,
-                            ...(i === 0 && stepThoughtSig
-                                ? { thoughtSignature: stepThoughtSig }
-                                : {}),
-                            stepIndex: step,
-                        })), stepToolExecs.map((te) => ({
-                            toolName: te.name,
-                            output: te.output,
-                            stepIndex: step,
-                        })), options, new Date()), TOOL_STORAGE_TIMEOUT_MS, "tool storage write timed out").catch((error) => {
-                            logger.warn("[GoogleVertex] Failed to store native Gemini generate tool executions", {
-                                error: error instanceof Error ? error.message : String(error),
+                    hitContextLimit = true;
+                    logger.warn(`[GoogleVertex] Gemini turn stopped by the context guard: ` +
+                        `projected prompt ~${contextGuard.projectedNextPromptTokens} tokens ` +
+                        `>= threshold ${contextGuard.thresholdTokens} — synthesizing a final answer.`);
+                    return { stop: true };
+                },
+                buildRequest: (conversation) => ({
+                    model: modelName,
+                    contents: conversation,
+                    config: { ...config, ...(tools ? { tools } : {}) },
+                }),
+                sendStep: async (request, signal) => {
+                    turnClock.noteProgress();
+                    const built = request;
+                    return client.models.generateContentStream({
+                        model: built.model,
+                        contents: built.contents,
+                        config: { ...(built.config ?? {}), abortSignal: signal },
+                    });
+                },
+            });
+            // Wrapped rather than configured: these fire once PER STEP, and
+            // buildToolResultMessages is the only hook that runs per step with
+            // exactly that step's results. Reading them off the turn's final result
+            // would batch every step into one late write and lose the per-step
+            // thought signature.
+            const adapter = {
+                ...engineAdapter,
+                // Counted HERE, not in buildToolResultMessages: this runs once per
+                // step exactly as the old `step++` at the top of the loop did,
+                // malformed retries included. Counting in the tool-result hook would
+                // skip the final text-only step and quietly report one step fewer in
+                // `stepsUsed`, which is a public field on the result.
+                buildStepRequest: (conversation, step) => {
+                    stepsTaken = step + 1;
+                    return engineAdapter.buildStepRequest(conversation, step);
+                },
+                buildToolResultMessages: (conversation, stepResult, toolResults, engineStep) => {
+                    const next = engineAdapter.buildToolResultMessages(conversation, stepResult, toolResults, engineStep);
+                    // Time-budget wrap-up nudge (twin of the Anthropic loops' soft step
+                    // nudge): with the turn deadline approaching, tell the model to
+                    // consolidate. Rides as a trailing text part on the tool-response
+                    // user turn.
+                    if (turnClock.shouldNudgeWrapup()) {
+                        const last = next[next.length - 1];
+                        if (last && Array.isArray(last.parts)) {
+                            last.parts.push({
+                                text: buildWrapupNudgeText(useFinalResultTool),
                             });
-                        });
+                        }
                     }
-                    // Time-budget wrap-up nudge (twin of the Anthropic loops' soft
-                    // step nudge): with the turn deadline approaching, tell the model
-                    // to consolidate. Rides as a trailing text part on the
-                    // tool-response user turn.
-                    if (turnClock.shouldNudgeWrapup()) {
-                        functionResponses.push({
-                            text: buildWrapupNudgeText(useFinalResultTool),
+                    // Persist this step's tool calls/results into conversation memory.
+                    // Without this, tool_call / tool_result rows never reach Redis and
+                    // the chat-history UI loses every tool invocation. `thoughtSignature`
+                    // rides as a sibling on the first call of the step — Gemini 3 needs
+                    // it to match thinking patterns when the conversation is replayed.
+                    const stepThoughtSig = extractThoughtSignature(stepResult.raw.rawResponseParts);
+                    withTimeout(this.handleToolExecutionStorage(toolResults.map((result, index) => ({
+                        toolName: result.name,
+                        args: result.args,
+                        ...(index === 0 && stepThoughtSig
+                            ? { thoughtSignature: stepThoughtSig }
+                            : {}),
+                        stepIndex: engineStep + 1,
+                    })), toolResults.map((result) => ({
+                        toolName: result.name,
+                        output: result.output,
+                        stepIndex: engineStep + 1,
+                    })), options, new Date()), TOOL_STORAGE_TIMEOUT_MS, "tool storage write timed out").catch((error) => {
+                        logger.warn("[GoogleVertex] Failed to store native Gemini stream tool executions", {
+                            error: error instanceof Error ? error.message : String(error),
                         });
-                    }
-                    // The @google/genai SDK only accepts "user" and "model" as valid
-                    // roles in contents — function/tool responses must use role: "user"
-                    // (matching the SDK's automaticFunctionCalling implementation and
-                    // the Google AI Studio path). See note in executeNativeGemini3Stream.
-                    currentContents.push({
-                        role: "user",
-                        parts: functionResponses,
                     });
                     // Project this step's growth for the context guard: the appended
                     // tool results ride the next prompt (Gemini reports usage per call,
                     // but only for content it has already seen).
                     try {
-                        contextGuard.noteAppendedChars(JSON.stringify(functionResponses).length);
+                        const appended = next[next.length - 1];
+                        contextGuard.noteAppendedChars(JSON.stringify(appended?.parts ?? []).length);
                     }
                     catch {
                         /* estimation is best-effort — never break the loop */
                     }
+                    return next;
+                },
+            };
+            const { stream: engineStream, resultPromise } = runAgenticLoop(adapter, 
+            // A concrete parts array widens to the engine's `unknown[]` on its
+            // own; only the direction back needs an assertion.
+            currentContents, {
+                tools: buildDedupedEngineTools(declarations, options.tools, {
+                    toolTimeoutMs: toolExecTimeoutMs,
+                    abortSignal: effectiveSignal,
+                    onProgress: () => turnClock.noteProgress(),
+                }),
+                abortSignal: effectiveSignal,
+            });
+            // generate() returns one result rather than streaming, so the engine's
+            // chunks are drained and discarded — the answer comes off the turn's
+            // result. The drain still has to happen: leaving the channel unread
+            // would stall the engine once its buffer fills.
+            const pump = (async () => {
+                for await (const chunk of engineStream) {
+                    void chunk;
                 }
-                catch (error) {
-                    // A mid-drain abort surfaces as an AbortError from the `for await`.
-                    // Break gracefully into the terminal block instead of re-throwing
-                    // (a re-throw would route the caller's abort into a second unbounded
-                    // fallback generate()). Dual check as with the inner catch:
-                    // effectiveSignal.aborted (a signal we tripped) OR isAbortError(error)
-                    // (an abort-shaped throw) — either means "stop", not a real failure.
-                    if (effectiveSignal.aborted || isAbortError(error)) {
-                        wasAborted = true;
-                        break;
-                    }
-                    logger.error("[GoogleVertex] Native SDK generate error", {
-                        error,
-                        model: modelName,
-                        location: effectiveLocation,
-                        status: error?.status,
-                    });
-                    // Best-effort request context for formatProviderError —
-                    // this.modelName can be stale when options.model overrides the
-                    // instance default.
-                    try {
-                        if (error && typeof error === "object") {
-                            const e = error;
-                            e.requestModel = modelName;
-                            e.requestRegion = effectiveLocation;
-                        }
-                    }
-                    catch {
-                        /* frozen/sealed error — context stays best-effort */
-                    }
+            })();
+            let engineResult;
+            try {
+                engineResult = await resultPromise;
+            }
+            catch (error) {
+                await pump.catch(() => { });
+                // A mid-drain abort surfaces as an AbortError. End gracefully into the
+                // terminal block instead of re-throwing — a re-throw would route the
+                // caller's abort into a second unbounded fallback stream().
+                if (effectiveSignal.aborted || isAbortError(error)) {
+                    wasAborted = true;
+                }
+                else {
+                    logger.error("[GoogleVertex] Native SDK error", error);
                     throw this.handleProviderError(error);
                 }
             }
+            await pump;
+            if (engineResult) {
+                finalText = engineResult.text;
+                lastFinishReason = engineResult.rawStopReason ?? lastFinishReason;
+                for (const call of engineResult.toolCalls) {
+                    allToolCalls.push({ toolName: call.name, args: call.args });
+                }
+                for (const execution of engineResult.toolExecutions) {
+                    toolExecutions.push({
+                        name: execution.name,
+                        input: execution.input,
+                        output: execution.output,
+                    });
+                }
+                // Replace in place: `currentContents` is a const the terminal block
+                // and the synth call both read.
+                currentContents.length = 0;
+                currentContents.push(...engineResult.conversation);
+            }
+            if (effectiveSignal.aborted) {
+                wasAborted = true;
+            }
             // Handle maxSteps termination / abort — the loop exited because the step
             // cap was reached (or the turn was aborted) while the model was still
             // calling tools. Surface a real answer instead of the canned placeholder
             // (Bug 1) and a meaningful finishReason (Bug 2).
-            if (!finalText && (step >= maxSteps || wasAborted || hitContextLimit)) {
-                hitStepLimit = step >= maxSteps && !wasAborted;
+            if (!finalText &&
+                (stepsTaken >= maxSteps || wasAborted || hitContextLimit)) {
+                hitStepLimit = stepsTaken >= maxSteps && !wasAborted;
                 const toolCallCount = allToolCalls.filter((tc) => tc.toolName !== "final_result").length;
                 if (accumulatedText) {
                     // Prefer the prose the model already produced across steps.
@@ -3015,7 +2704,7 @@ export class GoogleVertexProvider extends BaseProvider {
         if (stopReason !== "completed") {
             this.emitTurnEvent({
                 phase: stopReason,
-                step,
+                step: stepsTaken,
                 maxSteps,
                 toolCallCount: allToolCalls.filter((tc) => tc.toolName !== "final_result").length,
                 elapsedMs: turnClock.elapsedMs(),
@@ -3037,7 +2726,7 @@ export class GoogleVertexProvider extends BaseProvider {
             finishReason: resolvedFinishReason,
             stopReason,
             rawFinishReason: lastFinishReason,
-            stepsUsed: step,
+            stepsUsed: stepsTaken,
             usage: {
                 input: adjustedInputTokens,
                 // Thinking tokens are billed at the output rate but Gemini does NOT
@@ -3185,7 +2874,20 @@ export class GoogleVertexProvider extends BaseProvider {
      */
     async createAnthropicVertexClient(timeoutMs) {
         const mod = await getAnthropicVertexModule();
-        const settings = await createVertexAnthropicSettings(this.location, timeoutMs);
+        const expressApiKey = this.resolveExpressApiKey();
+        const directBaseURL = this.baseURL?.trim() || process.env.GOOGLE_VERTEX_BASE_URL?.trim();
+        const settings = await createVertexAnthropicSettings(this.location, timeoutMs, expressApiKey
+            ? {
+                apiKey: expressApiKey,
+                ...(this.projectId ? { projectId: this.projectId } : {}),
+            }
+            : undefined, directBaseURL);
+        // One assertion, at the one place the shapes genuinely differ. The SDK
+        // declares `authClient` as its full `AuthClient` (24-plus members) while
+        // `prepareOptions()` only ever calls `getRequestHeaders()` and reads
+        // `projectId`. Naming the narrow surface in our own type and widening it
+        // here is the honest version of that gap; the alternative is standing up a
+        // real AuthClient to satisfy a contract the SDK does not exercise.
         const client = new mod.AnthropicVertex(settings);
         // The vertex SDK eagerly starts Google ADC resolution in its constructor
         // (`this._authClientPromise = this._auth.getClient()`) and only awaits it
@@ -5545,7 +5247,7 @@ export class GoogleVertexProvider extends BaseProvider {
         if (stopReason !== "completed") {
             this.emitTurnEvent({
                 phase: stopReason,
-                step,
+                step: step,
                 maxSteps,
                 toolCallCount: externalToolCalls.length,
                 elapsedMs: turnClock.elapsedMs(),