@symerian/symi 2.3.3 → 2.4.0

package/dist/{pi-tools.policy-juZV2uUv.js → pi-tools.policy-CIDBbw6x.js}

@@ -1,9 +1,9 @@
 import { C as resolveThreadParentSessionKey, l as resolveAgentIdFromSessionKey } from "./session-key-DCt45XZa.js";
-import { r as resolveAgentConfig } from "./agent-scope-BRwEc2pG.js";
+import { r as resolveAgentConfig } from "./agent-scope-CgUHAtCo.js";
 import { l as normalizeMessageChannel } from "./message-channel-aT-I_DTX.js";
-import { z as DEFAULT_SUBAGENT_MAX_SPAWN_DEPTH } from "./config-CJMcIRQM.js";
-import { T as matchesAnyGlobPattern, w as compileGlobPatterns, x as normalizeToolName, y as expandToolGroups } from "./sandbox-BzNrUHpt.js";
-import { o as resolveChannelGroupToolsPolicy, t as getChannelDock } from "./dock-BYiRZSgZ.js";
+import { z as DEFAULT_SUBAGENT_MAX_SPAWN_DEPTH } from "./config-CHwyw6l5.js";
+import { T as matchesAnyGlobPattern, w as compileGlobPatterns, x as normalizeToolName, y as expandToolGroups } from "./sandbox-C-T6hOMv.js";
+import { o as resolveChannelGroupToolsPolicy, t as getChannelDock } from "./dock-BhS32F6E.js";
 
 //#region src/agents/sandbox-tool-policy.ts
 function unionAllow(base, extra) {

package/dist/{plugin-auto-enable-Bbm7MwFV.js → plugin-auto-enable-BzX1uAKy.js}

@@ -1,7 +1,7 @@
 import { d as isRecord } from "./utils-B-0b9bGM.js";
 import { o as getChatChannelMeta, s as listChatChannels, u as normalizeChatChannelId } from "./registry-Cja8eT7G.js";
-import { l as normalizeProviderId } from "./model-selection-X1oLh3CE.js";
-import { t as hasAnyWhatsAppAuth } from "./accounts-Cd816n6l.js";
+import { l as normalizeProviderId } from "./model-selection-CyFvYcRt.js";
+import { t as hasAnyWhatsAppAuth } from "./accounts-DDuIpcln.js";
 import { n as getChannelPluginCatalogEntry, r as listChannelPluginCatalogEntries } from "./catalog-RiADx7MA.js";
 import { t as ensurePluginAllowlisted } from "./plugins-allowlist-sArtGE-d.js";
 

package/dist/{plugin-registry-DvWWctC9.js → plugin-registry-BDpdOTei.js}

@@ -2,7 +2,7 @@ import { t as __exportAll } from "./rolldown-runtime-Cbj13DAv.js";
 import { M as getActivePluginRegistry, o as createSubsystemLogger } from "./entry.js";
 import { c as resolveAgentWorkspaceDir, l as resolveDefaultAgentId } from "./agent-scope-D-jRCY0d.js";
 import { i as loadConfig } from "./config-DHBLS1Hl.js";
-import { b as loadSymiPlugins } from "./subagent-registry-CBklD0OL.js";
+import { b as loadSymiPlugins } from "./subagent-registry-CbTbJKLs.js";
 
 //#region src/cli/plugin-registry.ts
 var plugin_registry_exports = /* @__PURE__ */ __exportAll({ ensurePluginRegistryLoaded: () => ensurePluginRegistryLoaded });

package/dist/{plugin-registry-lWWSu9UD.js → plugin-registry-CZeHSx_c.js}

@@ -1,9 +1,9 @@
 import { t as __exportAll } from "./rolldown-runtime-Cbj13DAv.js";
-import { vt as loadSymiPlugins } from "./reply-DVxU0AMq.js";
+import { vt as loadSymiPlugins } from "./reply-Ca3slrA3.js";
 import { d as getActivePluginRegistry } from "./registry-Cja8eT7G.js";
 import { t as createSubsystemLogger } from "./subsystem-D9vIQve0.js";
-import { c as resolveAgentWorkspaceDir, l as resolveDefaultAgentId } from "./agent-scope-BRwEc2pG.js";
-import { i as loadConfig } from "./config-CJMcIRQM.js";
+import { c as resolveAgentWorkspaceDir, l as resolveDefaultAgentId } from "./agent-scope-CgUHAtCo.js";
+import { i as loadConfig } from "./config-CHwyw6l5.js";
 
 //#region src/cli/plugin-registry.ts
 var plugin_registry_exports = /* @__PURE__ */ __exportAll({ ensurePluginRegistryLoaded: () => ensurePluginRegistryLoaded });

package/dist/plugin-sdk/agents/model-aware-stream.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Model-Aware Stream Function — the single wrapper that replaces the
+ * 6-layer chain in applyExtraParamsToAgent().
+ *
+ * Handles ALL provider-specific concerns in ONE onPayload hook:
+ * - Injects stop, top_p, top_k, repeat_penalty (ignored by buildParams)
+ * - Fixes max_completion_tokens → max_tokens for vLLM
+ * - Removes unsupported fields (store) for vLLM
+ * - Chains caller's onPayload explicitly (no collision)
+ *
+ * @module
+ */
+import type { StreamFn } from "@mariozechner/pi-agent-core";
+import type { SymiConfig } from "../config/config.js";
+import type { ModelProfile } from "../config/model-profiles.js";
+export interface ModelAwareStreamOptions {
+    /** Base streaming function — streamSimple or createOllamaStreamFn. */
+    baseStreamFn?: StreamFn;
+    /** Resolved model profile (from resolveModelProfile). */
+    profile: ModelProfile;
+    /** Provider ID (e.g. "redsand", "anthropic"). */
+    provider: string;
+    /** Model ID (e.g. "gemma-4-31b", "claude-sonnet-4-6"). */
+    modelId: string;
+    /** Full Symi config — used for provider baseUrl detection. */
+    config?: SymiConfig;
+    /** Per-model params from symi.json (agents.defaults.models[key].params). */
+    configExtraParams?: Record<string, unknown>;
+    /** Runtime overrides (e.g. CLI --temperature). */
+    streamParamsOverride?: Record<string, unknown>;
+}
+/**
+ * Create a single stream function wrapper that handles all provider-specific
+ * param injection via one onPayload hook.
+ *
+ * Replaces: createStreamFnWithExtraParams, createVllmCompatWrapper,
+ * createOpenRouterHeadersWrapper.
+ */
+export declare function createModelAwareStreamFn(opts: ModelAwareStreamOptions): StreamFn;
+/** Detect vLLM/Gemma providers by model hints or baseUrl patterns. */
+export declare function isVllmProvider(provider: string, modelId: string, config?: SymiConfig): boolean;

package/dist/plugin-sdk/agents/model-token-filter.d.ts

@@ -11,5 +11,16 @@ declare function isGemmaModel(modelId: string): boolean;
  * Strip known control tokens from text.  Only applies to Gemma-family models;
  * other models pass through unchanged.
  */
+/**
+ * Strip known control tokens from text.  Only applies to Gemma-family models;
+ * other models pass through unchanged.  Used for streamed text output.
+ */
 export declare function stripModelTokens(text: string, modelId: string): string;
+/**
+ * Aggressively strip ALL token-like fragments from a string.  Used for tool
+ * call arguments where any `<|` corruption causes downstream failures.
+ * More aggressive than `stripModelTokens` — removes bare `<|` and `|>` fragments
+ * that don't match any known token template.
+ */
+export declare function stripModelTokensAggressive(text: string, modelId: string): string;
 export { isGemmaModel, GEMMA_TOKEN_PATTERNS };

package/dist/plugin-sdk/agents/pi-embedded-runner/extra-params.d.ts

@@ -11,6 +11,19 @@ export declare function resolveExtraParams(params: {
     provider: string;
     modelId: string;
 }): Record<string, unknown> | undefined;
+export declare function createOpenAIResponsesStoreWrapper(baseStreamFn: StreamFn | undefined): StreamFn;
+export declare function resolveAnthropicBetas(extraParams: Record<string, unknown> | undefined, provider: string, modelId: string): string[] | undefined;
+export declare function createAnthropicBetaHeadersWrapper(baseStreamFn: StreamFn | undefined, betas: string[]): StreamFn;
+/**
+ * Create a streamFn wrapper that injects tool_stream=true for Z.AI providers.
+ *
+ * Z.AI's API supports the `tool_stream` parameter to enable real-time streaming
+ * of tool call arguments and reasoning content. When enabled, the API returns
+ * progressive tool_call deltas, allowing users to see tool execution in real-time.
+ *
+ * @see https://docs.z.ai/api-reference#streaming
+ */
+export declare function createZaiToolStreamWrapper(baseStreamFn: StreamFn | undefined, enabled: boolean): StreamFn;
 /**
  * Apply extra params (like temperature) to an agent's streamFn.
  * Also adds OpenRouter app attribution headers when using the OpenRouter provider.

package/dist/plugin-sdk/{channel-web-BnrTC55Y.js → channel-web-uUypcT9Q.js}

@@ -1,6 +1,6 @@
 import { a as buildGroupHistoryKey, n as DEFAULT_MAIN_KEY, r as buildAgentMainSessionKey, s as normalizeAgentId } from "./session-key-C_0eELjb.js";
 import { f as readWebSelfId, o as getWebAuthAgeMs, r as resolveWhatsAppAccount } from "./accounts-D9zGZU5t.js";
-import { E as enqueueSystemEvent, G as resolveMentionGating, W as shouldAckReactionForWhatsApp, X as buildHistoryContextFromEntries, Y as DEFAULT_GROUP_HISTORY_LIMIT, _ as resolveEnvelopeFormatOptions, c as computeBackoff, d as formatDurationPrecise, f as dispatchReplyWithBufferedBlockDispatcher, g as formatInboundEnvelope, h as resolveInboundDebounceMs, ht as createDedupeCache, l as sleepWithAbort, m as createInboundDebouncer, n as normalizeGroupActivation, nt as buildMentionRegexes, r as parseActivationCommand, rt as normalizeMentionText, t as getReplyFromConfig, tt as recordPendingHistoryEntryIfEnabled, u as buildPairingReply, v as hasControlCommand, y as shouldComputeCommandAuthorized } from "./reply-kuMxu6Q4.js";
+import { E as enqueueSystemEvent, G as resolveMentionGating, W as shouldAckReactionForWhatsApp, X as buildHistoryContextFromEntries, Y as DEFAULT_GROUP_HISTORY_LIMIT, _ as resolveEnvelopeFormatOptions, c as computeBackoff, d as formatDurationPrecise, f as dispatchReplyWithBufferedBlockDispatcher, g as formatInboundEnvelope, h as resolveInboundDebounceMs, ht as createDedupeCache, l as sleepWithAbort, m as createInboundDebouncer, n as normalizeGroupActivation, nt as buildMentionRegexes, r as parseActivationCommand, rt as normalizeMentionText, t as getReplyFromConfig, tt as recordPendingHistoryEntryIfEnabled, u as buildPairingReply, v as hasControlCommand, y as shouldComputeCommandAuthorized } from "./reply-DTKTuCrn.js";
 import { A as normalizeE164, B as toWhatsappJid, K as logVerbose, N as resolveJidToE164, O as isSelfChatMode, R as sleep, Y as shouldLogVerbose, k as jidToE164, s as normalizeChatChannelId, tt as getChildLogger, x as clamp } from "./registry--_pGht6S.js";
 import { n as loadConfig } from "./config-DDkdiUOR.js";
 import { i as defaultRuntime, t as createSubsystemLogger } from "./subsystem-Coz2AgU8.js";

package/dist/plugin-sdk/index.js

@@ -1,6 +1,6 @@
 import { f as DEFAULT_ACCOUNT_ID, p as normalizeAccountId } from "./session-key-C_0eELjb.js";
 import { _ as createAccountListHelpers, i as resolveWhatsAppAuthDir, n as resolveDefaultWhatsAppAccountId, r as resolveWhatsAppAccount, t as listWhatsAppAccountIds } from "./accounts-D9zGZU5t.js";
-import { $ as clearHistoryEntriesIfEnabled, A as resolveDiscordUserAllowlist, B as logTypingFailure, C as stripMarkdown, Ct as unbindThreadBindingsBySessionKey, D as resolveSlackUserAllowlist, Dt as BLUEBUBBLES_ACTION_NAMES, Et as BLUEBUBBLES_ACTIONS, F as stringEnum, G as resolveMentionGating, H as removeAckReactionAfterReply, I as recordInboundSession, J as summarizeMapping, K as resolveMentionGatingWithBypass, L as resolveControlCommandGate, M as collectDiscordAuditChannelIds, N as formatDocsLink, O as resolveSlackChannelAllowlist, Ot as BLUEBUBBLES_GROUP_ACTIONS, P as optionalStringEnum, Q as clearHistoryEntries, R as logAckFailure, S as processLineMessage, St as listThreadBindingsBySessionKey, T as attachFooterText, Tt as resolveAllowlistMatchSimple, U as shouldAckReaction, V as createTypingCallbacks, W as shouldAckReactionForWhatsApp, Y as DEFAULT_GROUP_HISTORY_LIMIT, Z as buildPendingHistoryContextFromMap, _t as listDevicePairing, a as normalizeAccountId$1, at as isWSLEnv, bt as registerPluginHttpRoute, ct as DEFAULT_WEBHOOK_MAX_BODY_BYTES, dt as isRequestBodyLimitError, et as recordPendingHistoryEntry, ft as readJsonBodyWithLimit, gt as approveDevicePairing, ht as createDedupeCache, i as listLineAccountIds, it as isWSL2Sync, j as resolveDiscordChannelAllowlist, k as detectBinary, kt as CHANNEL_MESSAGE_ACTION_NAMES, lt as RequestBodyLimitError, mt as requestBodyErrorToText, o as resolveDefaultLineAccountId, ot as isWSLSync, pt as readRequestBodyWithLimit, q as mergeAllowlist, s as resolveLineAccount, st as DEFAULT_WEBHOOK_BODY_TIMEOUT_MS, tt as recordPendingHistoryEntryIfEnabled, ut as installRequestBodyLimitGuard, vt as rejectDevicePairing, w as createReceiptCard, wt as formatAllowlistMatchMeta, x as hasMarkdownToConvert, xt as autoBindSpawnedDiscordSubagent, yt as extractToolSend, z as logInboundDrop } from "./reply-kuMxu6Q4.js";
+import { $ as clearHistoryEntriesIfEnabled, A as resolveDiscordUserAllowlist, B as logTypingFailure, C as stripMarkdown, Ct as unbindThreadBindingsBySessionKey, D as resolveSlackUserAllowlist, Dt as BLUEBUBBLES_ACTION_NAMES, Et as BLUEBUBBLES_ACTIONS, F as stringEnum, G as resolveMentionGating, H as removeAckReactionAfterReply, I as recordInboundSession, J as summarizeMapping, K as resolveMentionGatingWithBypass, L as resolveControlCommandGate, M as collectDiscordAuditChannelIds, N as formatDocsLink, O as resolveSlackChannelAllowlist, Ot as BLUEBUBBLES_GROUP_ACTIONS, P as optionalStringEnum, Q as clearHistoryEntries, R as logAckFailure, S as processLineMessage, St as listThreadBindingsBySessionKey, T as attachFooterText, Tt as resolveAllowlistMatchSimple, U as shouldAckReaction, V as createTypingCallbacks, W as shouldAckReactionForWhatsApp, Y as DEFAULT_GROUP_HISTORY_LIMIT, Z as buildPendingHistoryContextFromMap, _t as listDevicePairing, a as normalizeAccountId$1, at as isWSLEnv, bt as registerPluginHttpRoute, ct as DEFAULT_WEBHOOK_MAX_BODY_BYTES, dt as isRequestBodyLimitError, et as recordPendingHistoryEntry, ft as readJsonBodyWithLimit, gt as approveDevicePairing, ht as createDedupeCache, i as listLineAccountIds, it as isWSL2Sync, j as resolveDiscordChannelAllowlist, k as detectBinary, kt as CHANNEL_MESSAGE_ACTION_NAMES, lt as RequestBodyLimitError, mt as requestBodyErrorToText, o as resolveDefaultLineAccountId, ot as isWSLSync, pt as readRequestBodyWithLimit, q as mergeAllowlist, s as resolveLineAccount, st as DEFAULT_WEBHOOK_BODY_TIMEOUT_MS, tt as recordPendingHistoryEntryIfEnabled, ut as installRequestBodyLimitGuard, vt as rejectDevicePairing, w as createReceiptCard, wt as formatAllowlistMatchMeta, x as hasMarkdownToConvert, xt as autoBindSpawnedDiscordSubagent, yt as extractToolSend, z as logInboundDrop } from "./reply-DTKTuCrn.js";
 import "./paths-DR2yt_mP.js";
 import "./github-copilot-token-D9X2phUj.js";
 import { D as resolveSlackAccount, E as resolveDefaultSlackAccountId, F as resolveDiscordAccount, M as listDiscordAccountIds, O as resolveSlackReplyToMode, P as resolveDefaultDiscordAccountId, S as resolveTelegramAccount, T as listSlackAccountIds, _ as normalizeWhatsAppTarget, a as listDiscordDirectoryPeersFromConfig, b as listTelegramAccountIds, c as listTelegramDirectoryGroupsFromConfig, d as listWhatsAppDirectoryPeersFromConfig, f as looksLikeSlackTargetId, g as isWhatsAppGroupJid, i as listDiscordDirectoryGroupsFromConfig, l as listTelegramDirectoryPeersFromConfig, o as listSlackDirectoryGroupsFromConfig, p as normalizeSlackMessagingTarget, s as listSlackDirectoryPeersFromConfig, u as listWhatsAppDirectoryGroupsFromConfig, w as listEnabledSlackAccounts, x as resolveDefaultTelegramAccountId } from "./plugins-BbAvhC25.js";
@@ -47,7 +47,7 @@ import "./paths-A0xdf3yk.js";
 import { h as onDiagnosticEvent, m as isDiagnosticsEnabled, p as emitDiagnosticEvent } from "./diagnostic-mFf4i4G9.js";
 import { n as extractOriginalFilename } from "./store-Do3t33-c.js";
 import { t as resolveWhatsAppOutboundTarget } from "./resolve-outbound-target-BkCUbYGV.js";
-import { r as resolveWhatsAppHeartbeatRecipients } from "./channel-web-BnrTC55Y.js";
+import { r as resolveWhatsAppHeartbeatRecipients } from "./channel-web-uUypcT9Q.js";
 import "./image-BOYy0Ump.js";
 import "./pi-model-discovery-LbcEa65a.js";
 import "./api-key-rotation-CVBMpnPc.js";

package/dist/plugin-sdk/{reply-kuMxu6Q4.js → reply-DTKTuCrn.js}

@@ -27119,7 +27119,7 @@ async function runWithImageModelFallback(params) {
 function createDefaultDeps() {
 	return {
 		sendMessageWhatsApp: async (...args) => {
-			const { sendMessageWhatsApp } = await import("./web-TMIhKQyF.js");
+			const { sendMessageWhatsApp } = await import("./web-Dh6Youzv.js");
 			return await sendMessageWhatsApp(...args);
 		},
 		sendMessageTelegram: async (...args) => {
@@ -44745,7 +44745,7 @@ function loadWebLoginQr() {
 	return webLoginQrPromise;
 }
 function loadWebChannel() {
-	webChannelPromise ??= import("./web-TMIhKQyF.js");
+	webChannelPromise ??= import("./web-Dh6Youzv.js");
 	return webChannelPromise;
 }
 function loadWhatsAppActions() {
@@ -49991,8 +49991,16 @@ const GEMMA_TOKEN_PATTERNS = [
 	/<start_of_turn>(?:model|user)?/g,
 	/<end_of_turn>/g,
 	/<\|(?:fim_prefix|fim_suffix|fim_middle)\|>/g,
-	/^\s*<bos>\s*/gm
+	/^\s*<bos>\s*/gm,
+	/<\|[^|>]{0,30}\|>/g,
+	/<\|[^>]{0,30}>/g
 ];
+/**
+* Aggressively strip all fragments of model control token corruption from
+* tool arguments.  This catches cases where the model produces raw garbage
+* like `<|<|"|<|<|\'` that doesn't match any known token template.
+*/
+const AGGRESSIVE_TOKEN_FRAGMENTS = [/<\|/g, /\|>/g];
 const GEMMA_MODEL_HINTS$1 = ["gemma", "redsand/"];
 function isGemmaModel$1(modelId) {
 	const lower = modelId.toLowerCase();
@@ -50002,6 +50010,10 @@ function isGemmaModel$1(modelId) {
 * Strip known control tokens from text.  Only applies to Gemma-family models;
 * other models pass through unchanged.
 */
+/**
+* Strip known control tokens from text.  Only applies to Gemma-family models;
+* other models pass through unchanged.  Used for streamed text output.
+*/
 function stripModelTokens(text, modelId) {
 	if (!text || !isGemmaModel$1(modelId)) return text;
 	let result = text;
@@ -50011,6 +50023,21 @@ function stripModelTokens(text, modelId) {
 	}
 	return result;
 }
+/**
+* Aggressively strip ALL token-like fragments from a string.  Used for tool
+* call arguments where any `<|` corruption causes downstream failures.
+* More aggressive than `stripModelTokens` — removes bare `<|` and `|>` fragments
+* that don't match any known token template.
+*/
+function stripModelTokensAggressive(text, modelId) {
+	if (!text || !isGemmaModel$1(modelId)) return text;
+	let result = stripModelTokens(text, modelId);
+	for (const re of AGGRESSIVE_TOKEN_FRAGMENTS) {
+		re.lastIndex = 0;
+		result = result.replace(re, "");
+	}
+	return result;
+}
 
 //#endregion
 //#region src/agents/pi-tools.abort.ts
@@ -61168,10 +61195,10 @@ function wrapToolWithArgSanitization(tool, modelId) {
 	};
 }
 function sanitizeToolArgs(params, modelId) {
-	if (!params || typeof params !== "object") return typeof params === "string" ? stripModelTokens(params, modelId) : params;
+	if (!params || typeof params !== "object") return typeof params === "string" ? stripModelTokensAggressive(params, modelId) : params;
 	if (Array.isArray(params)) return params.map((item) => sanitizeToolArgs(item, modelId));
 	const result = {};
-	for (const [key, value] of Object.entries(params)) if (typeof value === "string") result[key] = stripModelTokens(value, modelId);
+	for (const [key, value] of Object.entries(params)) if (typeof value === "string") result[key] = stripModelTokensAggressive(value, modelId);
 	else if (value && typeof value === "object") result[key] = sanitizeToolArgs(value, modelId);
 	else result[key] = value;
 	return result;
@@ -64420,10 +64447,6 @@ function resolveModelProfile(modelId, userProfiles) {
 
 //#endregion
 //#region src/agents/pi-embedded-runner/extra-params.ts
-const OPENROUTER_APP_HEADERS = {
-	"HTTP-Referer": "https://symi.ai",
-	"X-Title": "Symi"
-};
 const ANTHROPIC_CONTEXT_1M_BETA = "context-1m-2025-08-07";
 const ANTHROPIC_1M_MODEL_PREFIXES = ["claude-opus-4", "claude-sonnet-4"];
 const OPENAI_RESPONSES_APIS = new Set(["openai-responses"]);
@@ -64441,70 +64464,6 @@ function resolveExtraParams(params) {
 	const profile = resolveModelProfile(modelKey, params.cfg?.models?.profiles);
 	if (profile.params && Object.keys(profile.params).length > 0) return { ...profile.params };
 }
-/**
-* Resolve cacheRetention from extraParams, supporting both new `cacheRetention`
-* and legacy `cacheControlTtl` values for backwards compatibility.
-*
-* Mapping: "5m" → "short", "1h" → "long"
-*
-* Only applies to Anthropic provider (OpenRouter uses openai-completions API
-* with hardcoded cache_control, not the cacheRetention stream option).
-*
-* Defaults to "short" for Anthropic provider when not explicitly configured.
-*/
-function resolveCacheRetention(extraParams, provider) {
-	if (provider !== "anthropic") return;
-	const newVal = extraParams?.cacheRetention;
-	if (newVal === "none" || newVal === "short" || newVal === "long") return newVal;
-	const legacy = extraParams?.cacheControlTtl;
-	if (legacy === "5m") return "short";
-	if (legacy === "1h") return "long";
-	return "short";
-}
-function createStreamFnWithExtraParams(baseStreamFn, extraParams, provider) {
-	if (!extraParams || Object.keys(extraParams).length === 0) return;
-	const streamParams = {};
-	if (typeof extraParams.temperature === "number") streamParams.temperature = extraParams.temperature;
-	if (typeof extraParams.maxTokens === "number") streamParams.maxTokens = extraParams.maxTokens;
-	if (typeof extraParams.top_p === "number") streamParams.top_p = extraParams.top_p;
-	if (typeof extraParams.top_k === "number") streamParams.top_k = extraParams.top_k;
-	if (typeof extraParams.repeat_penalty === "number") streamParams.repeat_penalty = extraParams.repeat_penalty;
-	if (Array.isArray(extraParams.stop) && extraParams.stop.length > 0) streamParams.stop = extraParams.stop;
-	if (extraParams.chat_template_kwargs && typeof extraParams.chat_template_kwargs === "object") streamParams.chat_template_kwargs = extraParams.chat_template_kwargs;
-	const cacheRetention = resolveCacheRetention(extraParams, provider);
-	if (cacheRetention) streamParams.cacheRetention = cacheRetention;
-	if (Object.keys(streamParams).length === 0) return;
-	log$2.debug(`creating streamFn wrapper with params: ${JSON.stringify(streamParams)}`);
-	const sp = streamParams;
-	const payloadOverrides = {};
-	if (Array.isArray(sp.stop) && sp.stop.length > 0) payloadOverrides.stop = sp.stop;
-	if (typeof sp.top_p === "number") payloadOverrides.top_p = sp.top_p;
-	if (typeof sp.top_k === "number") payloadOverrides.top_k = sp.top_k;
-	if (typeof sp.repeat_penalty === "number") payloadOverrides.repeat_penalty = sp.repeat_penalty;
-	const hasPayloadOverrides = Object.keys(payloadOverrides).length > 0;
-	const underlying = baseStreamFn ?? streamSimple;
-	const wrappedStreamFn = (model, context, options) => {
-		if (hasPayloadOverrides) {
-			const originalOnPayload = options?.onPayload;
-			return underlying(model, context, {
-				...streamParams,
-				...options,
-				onPayload: (payload) => {
-					if (payload && typeof payload === "object") {
-						const p = payload;
-						for (const [key, value] of Object.entries(payloadOverrides)) if (p[key] === void 0) p[key] = value;
-					}
-					originalOnPayload?.(payload);
-				}
-			});
-		}
-		return underlying(model, context, {
-			...streamParams,
-			...options
-		});
-	};
-	return wrappedStreamFn;
-}
 function isDirectOpenAIBaseUrl(baseUrl) {
 	if (typeof baseUrl !== "string" || !baseUrl.trim()) return true;
 	try {
@@ -64585,20 +64544,6 @@ function createAnthropicBetaHeadersWrapper(baseStreamFn, betas) {
 	};
 }
 /**
-* Create a streamFn wrapper that adds OpenRouter app attribution headers.
-* These headers allow Symi to appear on OpenRouter's leaderboard.
-*/
-function createOpenRouterHeadersWrapper(baseStreamFn) {
-	const underlying = baseStreamFn ?? streamSimple;
-	return (model, context, options) => underlying(model, context, {
-		...options,
-		headers: {
-			...OPENROUTER_APP_HEADERS,
-			...options?.headers
-		}
-	});
-}
-/**
 * Create a streamFn wrapper that injects tool_stream=true for Z.AI providers.
 *
 * Z.AI's API supports the `tool_stream` parameter to enable real-time streaming
@@ -64621,82 +64566,6 @@ function createZaiToolStreamWrapper(baseStreamFn, enabled) {
 		});
 	};
 }
-const GEMMA_STOP_SEQUENCES = ["<end_of_turn>", "<eos>"];
-const VLLM_GEMMA_HINTS = ["gemma", "redsand"];
-function isVllmOrGemmaProvider(provider, modelId, cfg) {
-	const lower = `${provider}/${modelId}`.toLowerCase();
-	if (VLLM_GEMMA_HINTS.some((h) => lower.includes(h))) return true;
-	const baseUrl = ((cfg?.models?.providers?.[provider])?.baseUrl ?? "").toLowerCase();
-	return baseUrl.includes("coreweave") || baseUrl.includes("vllm");
-}
-/**
-* Wrap the stream function to fix vLLM/Gemma compat issues via onPayload:
-* 1. Replace `max_completion_tokens` with `max_tokens` (vLLM ignores the former)
-* 2. Inject Gemma stop sequences if none are set
-* 3. Remove unsupported fields (store, stream_options.include_usage on older vLLM)
-*/
-function createVllmCompatWrapper(baseStreamFn, extraParams) {
-	const underlying = baseStreamFn ?? streamSimple;
-	const configuredStop = Array.isArray(extraParams.stop) ? extraParams.stop : GEMMA_STOP_SEQUENCES;
-	return (model, context, options) => {
-		const originalOnPayload = options?.onPayload;
-		return underlying(model, context, {
-			...options,
-			onPayload: (payload) => {
-				if (payload && typeof payload === "object") {
-					const p = payload;
-					if (p.max_completion_tokens && !p.max_tokens) {
-						p.max_tokens = p.max_completion_tokens;
-						delete p.max_completion_tokens;
-					}
-					if (!p.stop && configuredStop.length > 0) p.stop = configuredStop;
-					delete p.store;
-				}
-				originalOnPayload?.(payload);
-			}
-		});
-	};
-}
-/**
-* Apply extra params (like temperature) to an agent's streamFn.
-* Also adds OpenRouter app attribution headers when using the OpenRouter provider.
-*
-* @internal Exported for testing
-*/
-function applyExtraParamsToAgent(agent, cfg, provider, modelId, extraParamsOverride) {
-	const extraParams = resolveExtraParams({
-		cfg,
-		provider,
-		modelId
-	});
-	const override = extraParamsOverride && Object.keys(extraParamsOverride).length > 0 ? Object.fromEntries(Object.entries(extraParamsOverride).filter(([, value]) => value !== void 0)) : void 0;
-	const merged = Object.assign({}, extraParams, override);
-	const wrappedStreamFn = createStreamFnWithExtraParams(agent.streamFn, merged, provider);
-	if (wrappedStreamFn) {
-		log$2.debug(`applying extraParams to agent streamFn for ${provider}/${modelId}`);
-		agent.streamFn = wrappedStreamFn;
-	}
-	const anthropicBetas = resolveAnthropicBetas(merged, provider, modelId);
-	if (anthropicBetas?.length) {
-		log$2.debug(`applying Anthropic beta header for ${provider}/${modelId}: ${anthropicBetas.join(",")}`);
-		agent.streamFn = createAnthropicBetaHeadersWrapper(agent.streamFn, anthropicBetas);
-	}
-	if (provider === "openrouter") {
-		log$2.debug(`applying OpenRouter app attribution headers for ${provider}/${modelId}`);
-		agent.streamFn = createOpenRouterHeadersWrapper(agent.streamFn);
-	}
-	if (provider === "zai" || provider === "z-ai") {
-		if (merged?.tool_stream !== false) {
-			log$2.debug(`enabling Z.AI tool_stream for ${provider}/${modelId}`);
-			agent.streamFn = createZaiToolStreamWrapper(agent.streamFn, true);
-		}
-	}
-	if (isVllmOrGemmaProvider(provider, modelId, cfg)) {
-		log$2.debug(`applying vLLM/Gemma compat wrapper for ${provider}/${modelId}`);
-		agent.streamFn = createVllmCompatWrapper(agent.streamFn, merged);
-	}
-	agent.streamFn = createOpenAIResponsesStoreWrapper(agent.streamFn);
-}
 
 //#endregion
 //#region src/utils/safe-json.ts
@@ -64973,6 +64842,74 @@ function createCacheTrace(params) {
 	};
 }
 
+//#endregion
+//#region src/agents/model-aware-stream.ts
+/**
+* Create a single stream function wrapper that handles all provider-specific
+* param injection via one onPayload hook.
+*
+* Replaces: createStreamFnWithExtraParams, createVllmCompatWrapper,
+* createOpenRouterHeadersWrapper.
+*/
+function createModelAwareStreamFn(opts) {
+	const underlying = opts.baseStreamFn ?? streamSimple;
+	const mergedParams = {
+		...opts.profile.params,
+		...opts.configExtraParams,
+		...opts.streamParamsOverride
+	};
+	const nativeOpts = {};
+	if (typeof mergedParams.temperature === "number") nativeOpts.temperature = mergedParams.temperature;
+	if (typeof mergedParams.max_tokens === "number") nativeOpts.maxTokens = mergedParams.max_tokens;
+	else if (typeof mergedParams.maxTokens === "number") nativeOpts.maxTokens = mergedParams.maxTokens;
+	const injections = {};
+	if (Array.isArray(mergedParams.stop) && mergedParams.stop.length > 0) injections.stop = mergedParams.stop;
+	if (typeof mergedParams.top_p === "number") injections.top_p = mergedParams.top_p;
+	if (typeof mergedParams.top_k === "number") injections.top_k = mergedParams.top_k;
+	if (typeof mergedParams.repeat_penalty === "number") injections.repeat_penalty = mergedParams.repeat_penalty;
+	const isVllm = isVllmProvider(opts.provider, opts.modelId, opts.config);
+	const isOpenRouter = opts.provider === "openrouter";
+	const extraHeaders = {};
+	if (isOpenRouter) {
+		extraHeaders["HTTP-Referer"] = "https://symi.ai";
+		extraHeaders["X-Title"] = "Symi";
+	}
+	const hasExtraHeaders = Object.keys(extraHeaders).length > 0;
+	const onPayloadHook = (payload, callerOnPayload) => {
+		if (payload && typeof payload === "object") {
+			const p = payload;
+			for (const [key, val] of Object.entries(injections)) if (p[key] === void 0) p[key] = val;
+			if (isVllm) {
+				if (p.max_completion_tokens && !p.max_tokens) {
+					p.max_tokens = p.max_completion_tokens;
+					delete p.max_completion_tokens;
+				}
+				delete p.store;
+			}
+		}
+		callerOnPayload?.(payload);
+	};
+	return (model, context, options) => {
+		const callerOnPayload = options?.onPayload;
+		return underlying(model, context, {
+			...nativeOpts,
+			...options,
+			onPayload: (payload) => onPayloadHook(payload, callerOnPayload),
+			...hasExtraHeaders ? { headers: {
+				...extraHeaders,
+				...options?.headers
+			} } : {}
+		});
+	};
+}
+/** Detect vLLM/Gemma providers by model hints or baseUrl patterns. */
+function isVllmProvider(provider, modelId, config) {
+	const lower = `${provider}/${modelId}`.toLowerCase();
+	if (["gemma", "redsand"].some((h) => lower.includes(h))) return true;
+	const baseUrl = ((config?.models?.providers?.[provider])?.baseUrl ?? "").toLowerCase();
+	return baseUrl.includes("coreweave") || baseUrl.includes("vllm");
+}
+
 //#endregion
 //#region src/auto-reply/reply/streaming-directives.ts
 const splitTrailingDirective = (text) => {
@@ -68020,14 +67957,39 @@ async function runEmbeddedAttempt(params) {
 				modelApi: params.model.api,
 				workspaceDir: params.workspaceDir
 			});
-			if (params.model.api === "ollama") {
-				const providerConfig = params.config?.models?.providers?.[params.model.provider];
-				const modelBaseUrl = typeof params.model.baseUrl === "string" ? params.model.baseUrl.trim() : "";
-				const providerBaseUrl = typeof providerConfig?.baseUrl === "string" ? providerConfig.baseUrl.trim() : "";
-				const ollamaBaseUrl = modelBaseUrl || providerBaseUrl || OLLAMA_NATIVE_BASE_URL;
-				activeSession.agent.streamFn = createOllamaStreamFn(ollamaBaseUrl);
-			} else activeSession.agent.streamFn = streamSimple;
-			applyExtraParamsToAgent(activeSession.agent, params.config, params.provider, params.modelId, params.streamParams);
+			{
+				let baseStreamFn = streamSimple;
+				if (params.model.api === "ollama") {
+					const providerConfig = params.config?.models?.providers?.[params.model.provider];
+					const modelBaseUrl = typeof params.model.baseUrl === "string" ? params.model.baseUrl.trim() : "";
+					const providerBaseUrl = typeof providerConfig?.baseUrl === "string" ? providerConfig.baseUrl.trim() : "";
+					baseStreamFn = createOllamaStreamFn(modelBaseUrl || providerBaseUrl || OLLAMA_NATIVE_BASE_URL);
+				}
+				const configExtraParams = resolveExtraParams({
+					cfg: params.config,
+					provider: params.provider,
+					modelId: params.modelId
+				});
+				activeSession.agent.streamFn = createModelAwareStreamFn({
+					baseStreamFn,
+					profile: modelProfile,
+					provider: params.provider,
+					modelId: params.modelId,
+					config: params.config,
+					configExtraParams,
+					streamParamsOverride: params.streamParams
+				});
+				const merged = {
+					...configExtraParams,
+					...params.streamParams
+				};
+				const anthropicBetas = resolveAnthropicBetas(merged, params.provider, params.modelId);
+				if (anthropicBetas?.length) activeSession.agent.streamFn = createAnthropicBetaHeadersWrapper(activeSession.agent.streamFn, anthropicBetas);
+				if (params.provider === "zai" || params.provider === "z-ai") {
+					if (merged?.tool_stream !== false) activeSession.agent.streamFn = createZaiToolStreamWrapper(activeSession.agent.streamFn, true);
+				}
+				activeSession.agent.streamFn = createOpenAIResponsesStoreWrapper(activeSession.agent.streamFn);
+			}
 			if (cacheTrace) {
 				cacheTrace.recordStage("session:loaded", {
 					messages: activeSession.messages,

package/dist/plugin-sdk/{web-TMIhKQyF.js → web-Dh6Youzv.js}

@@ -1,5 +1,5 @@
 import { a as WA_WEB_AUTH_DIR, g as webAuthExists, s as logWebSelfId, u as pickWebChannel } from "./accounts-D9zGZU5t.js";
-import "./reply-kuMxu6Q4.js";
+import "./reply-DTKTuCrn.js";
 import "./paths-DR2yt_mP.js";
 import "./github-copilot-token-D9X2phUj.js";
 import "./plugins-BbAvhC25.js";
@@ -45,7 +45,7 @@ import "./pi-embedded-helpers-BveUP4hk.js";
 import "./paths-A0xdf3yk.js";
 import "./diagnostic-mFf4i4G9.js";
 import "./store-Do3t33-c.js";
-import { n as monitorWebInbox, t as monitorWebChannel } from "./channel-web-BnrTC55Y.js";
+import { n as monitorWebInbox, t as monitorWebChannel } from "./channel-web-uUypcT9Q.js";
 import "./image-BOYy0Ump.js";
 import "./pi-model-discovery-LbcEa65a.js";
 import "./api-key-rotation-CVBMpnPc.js";

package/dist/{plugins-BAyxWQSz.js → plugins-DKDeQZF0.js}

@@ -4,8 +4,8 @@ import { h as normalizeAccountId, m as DEFAULT_ACCOUNT_ID } from "./session-key-
 import { c as normalizeAnyChannelId, f as requireActivePluginRegistry, n as CHAT_CHANNEL_ORDER } from "./registry-Cja8eT7G.js";
 import { t as isTruthyEnvValue } from "./env-DPCHYPeH.js";
 import { t as normalizeChatType } from "./chat-type-DyovJwCt.js";
-import { h as createAccountListHelpers } from "./accounts-Cd816n6l.js";
-import { i as resolveDefaultAgentBoundAccountId, r as listBoundAccountIds } from "./bindings-ChzePa8q.js";
+import { h as createAccountListHelpers } from "./accounts-DDuIpcln.js";
+import { i as resolveDefaultAgentBoundAccountId, r as listBoundAccountIds } from "./bindings-DFaVVCsf.js";
 import fs from "node:fs";
 
 //#region src/channels/plugins/account-action-gate.ts