@symerian/symi 2.3.4 → 2.4.1

package/dist/{pi-embedded-JXvD5-F-.js → pi-embedded-CAJ6egA1.js}

@@ -51787,7 +51787,7 @@ function isVoiceChannelType(type) {
 function createDefaultDeps() {
 	return {
 		sendMessageWhatsApp: async (...args) => {
-			const { sendMessageWhatsApp } = await import("./web-CryqKXT4.js");
+			const { sendMessageWhatsApp } = await import("./web-KYh3LgI0.js");
 			return await sendMessageWhatsApp(...args);
 		},
 		sendMessageTelegram: async (...args) => {
@@ -67069,7 +67069,7 @@ function loadWebLoginQr() {
 	return webLoginQrPromise;
 }
 function loadWebChannel() {
-	webChannelPromise ??= import("./web-CryqKXT4.js");
+	webChannelPromise ??= import("./web-KYh3LgI0.js");
 	return webChannelPromise;
 }
 function loadWhatsAppActions() {
@@ -72836,10 +72836,6 @@ 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"]);
@@ -72857,70 +72853,6 @@ 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 {
@@ -73001,20 +72933,6 @@ 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
@@ -73037,82 +72955,6 @@ 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
@@ -73389,6 +73231,74 @@ 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) => {
@@ -75372,6 +75282,89 @@ Use the message tool with buttons:
 [{"label": "✅ Proceed", "data": "/plan-proceed"}, {"label": "✏️ Modify", "data": "/plan-modify"}, {"label": "❌ Cancel", "data": "/plan-cancel"}]
 </plan_mode>`;
 
+//#endregion
+//#region src/agents/stream-monitor.ts
+/**
+* Create a stream monitor configured for the given model profile.
+*
+* - Claude (hasStructuredThinking: true): repetition detection only
+* - Gemma/vLLM (hasStructuredThinking: false): tool JSON + repetition detection
+*/
+function createStreamMonitor(profile, overrides) {
+	const detectToolJson = overrides?.detectToolJson ?? !profile.filters.hasStructuredThinking;
+	const detectRepetition = overrides?.detectRepetition ?? true;
+	const minBlockSize = overrides?.repetitionMinBlockSize ?? 150;
+	let accumulated = "";
+	let stopSignaled = false;
+	let stopReason;
+	return {
+		onChunk(delta) {
+			if (stopSignaled) return;
+			accumulated += delta;
+			if (detectToolJson && hasCompleteToolJson(accumulated)) {
+				stopSignaled = true;
+				stopReason = "complete-tool-json";
+				return;
+			}
+			if (detectRepetition && hasRepetition(accumulated, minBlockSize)) {
+				stopSignaled = true;
+				stopReason = "repetition-detected";
+				return;
+			}
+		},
+		shouldStop: () => stopSignaled,
+		getAccumulated: () => accumulated,
+		getStopReason: () => stopReason,
+		reset() {
+			accumulated = "";
+			stopSignaled = false;
+			stopReason = void 0;
+		}
+	};
+}
+/**
+* Detect if the accumulated text contains a syntactically complete JSON
+* object with tool_name and tool_args.  Used for models that output tool
+* calls as plain text JSON (not native structured tool calls).
+*/
+function hasCompleteToolJson(text) {
+	const hasToolKey = text.includes("tool_name") || text.includes("\"tool\"");
+	const hasArgsKey = text.includes("tool_args") || text.includes("\"args\"");
+	if (!hasToolKey || !hasArgsKey) return false;
+	const lastBrace = text.lastIndexOf("}");
+	if (lastBrace < 0) return false;
+	let depth = 0;
+	let start = -1;
+	for (let i = lastBrace; i >= 0; i--) {
+		if (text[i] === "}") depth++;
+		if (text[i] === "{") depth--;
+		if (depth === 0) {
+			start = i;
+			break;
+		}
+	}
+	if (start < 0) return false;
+	const candidate = text.slice(start, lastBrace + 1);
+	try {
+		const parsed = JSON.parse(candidate);
+		return typeof parsed === "object" && parsed !== null && (typeof parsed.tool_name === "string" || typeof parsed.tool === "string") && (typeof parsed.tool_args === "object" || typeof parsed.args === "object");
+	} catch {
+		return false;
+	}
+}
+/**
+* Detect if the accumulated text contains a repeated block, indicating
+* a looping cascade.  Returns true if a block of `minBlockSize` chars
+* from the first half appears again in the second half.
+*/
+function hasRepetition(text, minBlockSize) {
+	if (text.length < minBlockSize * 2.5) return false;
+	const sampleStart = Math.floor(text.length * .25);
+	const sample = text.slice(sampleStart, sampleStart + minBlockSize);
+	const searchStart = sampleStart + minBlockSize;
+	return text.indexOf(sample, searchStart) >= 0;
+}
+
 //#endregion
 //#region src/agents/symipulse-timeout.ts
 /**
@@ -76436,14 +76429,39 @@ 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,
@@ -76561,6 +76579,7 @@ async function runEmbeddedAttempt(params) {
 					return fn(...args);
 				});
 			};
+			const streamMonitor = createStreamMonitor(modelProfile);
 			const subscription = subscribeEmbeddedPiSession({
 				session: activeSession,
 				runId: params.runId,
@@ -76582,6 +76601,14 @@ async function runEmbeddedAttempt(params) {
 				onAgentEvent: (evt) => {
 					armSymipulseTimer();
 					params.onAgentEvent?.(evt);
+					if (evt.stream === "assistant" && typeof evt.data?.text === "string" && !aborted) {
+						streamMonitor.onChunk(evt.data.text);
+						if (streamMonitor.shouldStop()) {
+							const reason = streamMonitor.getStopReason();
+							log$2.warn(`stream monitor triggered abort: runId=${params.runId} reason=${reason}`);
+							abortRun(false, /* @__PURE__ */ new Error(`stream-monitor:${reason}`));
+						}
+					}
 				},
 				enforceFinalTag: params.enforceFinalTag,
 				config: params.config,