@snoai/memory 1.2.9 → 1.2.10

package/dist/{llm-client-fVrzPcyz.js → llm-client-ConEyQEs.js}

@@ -1,6 +1,6 @@
 import { ct as FIXED_PROTOCOL_VALUE_69, d as FIXED_EXTERNAL_VALUE_8, f as FIXED_EXTERNAL_VALUE_9 } from "./signed-registry-constants-yq3DlG3B.js";
 import { i as memoryOperationSignal, r as memoryOperationId, t as checkMemoryOperation } from "./operation-cancellation-CTKWwNHk.js";
-import { t as MODEL_CALLS } from "./model-call-table-BZJQkbKX.js";
+import { t as MODEL_CALLS } from "./model-call-table-BE97f37s.js";
 import { classifyLlmFailure, isCredentialFailure, isTransientLlmStatus } from "./internal/model/llm-failure.js";
 import { a as repairCommonJson, i as previewText, n as readModelReplyJson, r as extractJsonFromResponse } from "./model-reply-text-CKopStP2.js";
 import { resolveLlmRoute } from "./internal/model/llm-mode-routing.js";
@@ -165,7 +165,10 @@ function materializeBundledLlmixEndpoint(input) {
 	if (!path) throw new Error(`sno-station-mem llm-client: signed endpoint path missing from preset ${input.preset.preset}`);
 	return {
 		url: new URL(path, resolveEndpointBase(input.baseSource, input.preset.provider !== "sno-gpu")).toString(),
-		preset: input.preset
+		preset: input.preset.provider === "sno-gpu" && path === "/extract/profile/v1/completions" ? {
+			...input.preset,
+			model: "b-profile"
+		} : input.preset
 	};
 }
 async function resolveConfiguredBundledLlmixEndpoint(input) {
@@ -782,6 +785,7 @@ function normalizeMemoryLlmRequest(value) {
 	const emptyReplyAttempts = raw.emptyReplyAttempts;
 	if (emptyReplyAttempts !== void 0 && (!Number.isInteger(emptyReplyAttempts) || Number(emptyReplyAttempts) < 1 || Number(emptyReplyAttempts) > EMPTY_REPLY_ATTEMPTS)) throw new Error(`sno-station-mem llm-client: emptyReplyAttempts must be an integer from 1 to ${EMPTY_REPLY_ATTEMPTS}`);
 	const enableThinking = raw.enableThinking;
+	const takeProviderAttempt = raw.takeProviderAttempt;
 	if (enableThinking !== void 0 && typeof enableThinking !== "boolean") throw new Error("sno-station-mem llm-client: enableThinking must be a boolean");
 	const signal = raw.signal;
 	if (signal !== void 0 && !hasAbortSignalShape(signal)) throw new Error("sno-station-mem llm-client: signal must be an AbortSignal");
@@ -795,6 +799,7 @@ function normalizeMemoryLlmRequest(value) {
 		...maxTokens !== void 0 ? { maxTokens: Number(maxTokens) } : {},
 		...emptyReplyAttempts !== void 0 ? { emptyReplyAttempts: Number(emptyReplyAttempts) } : {},
 		...enableThinking !== void 0 ? { enableThinking } : {},
+		...typeof takeProviderAttempt === "function" ? { takeProviderAttempt: () => takeProviderAttempt() === true } : {},
 		...requestId !== void 0 ? { requestId } : {},
 		...typeof raw.promptTemplateHash === "string" && /^[a-f0-9]{64}$/.test(raw.promptTemplateHash) ? { promptTemplateHash: raw.promptTemplateHash } : {},
 		...typeof raw.extractionSkillHash === "string" && /^[a-f0-9]{64}$/.test(raw.extractionSkillHash) ? { extractionSkillHash: raw.extractionSkillHash } : {},
@@ -802,7 +807,9 @@ function normalizeMemoryLlmRequest(value) {
 	};
 }
 function textSystemContent(request) {
-	return MODEL_CALLS[request.callId].occasion === "conflictAdjudication" ? CONFLICT_ADJUDICATION_SYSTEM_CONTENT : REFLECTION_TEXT_SYSTEM_CONTENT;
+	const occasion = MODEL_CALLS[request.callId].occasion;
+	if (occasion === "conflictAdjudication") return CONFLICT_ADJUDICATION_SYSTEM_CONTENT;
+	return occasion === "memoryExtract" ? JSON_SYSTEM_CONTENT : REFLECTION_TEXT_SYSTEM_CONTENT;
 }
 /** Creates the LLM client facade with provider routing, key rotation, and JSON parsing. */
 function createLlmClient(config) {
@@ -832,8 +839,29 @@ function createLlmClient(config) {
 		}
 		return endpointPromise;
 	};
+	const providerBudget = new AsyncLocalStorage();
 	const pipeline = new CallPipeline({
-		dispatch: callProvider,
+		dispatch: async (context) => {
+			const budget = providerBudget.getStore();
+			if (!budget) return callProvider(context);
+			const timeoutMs = Math.floor(budget.deadlineMs - performance.now());
+			if (timeoutMs <= 0) throw Object.assign(/* @__PURE__ */ new Error("Request deadline exhausted before provider dispatch"), { statusCode: 499 });
+			if (budget.remaining <= 0 || budget.takeProviderAttempt && !budget.takeProviderAttempt()) throw Object.assign(/* @__PURE__ */ new Error("Provider request attempts exhausted"), { statusCode: 499 });
+			budget.remaining -= 1;
+			const providerConfig = context.config;
+			const dispatchConfig = {
+				...providerConfig,
+				timeoutMs,
+				...providerConfig.rawCompletion ? { rawCompletion: {
+					...providerConfig.rawCompletion,
+					timeoutMs
+				} } : {}
+			};
+			return callProvider({
+				...context,
+				config: dispatchConfig
+			});
+		},
 		maxRetries: 2,
 		retryBaseMs: 100,
 		retryMaxDelayMs: 500,
@@ -905,6 +933,7 @@ function createLlmClient(config) {
 		if (transport === "agent-host-seam") {
 			const agentPort = config.agentPort;
 			if (!agentPort) throw new LlmClientTerminalError("transport", `sno-station-mem: llm-client [${request.callId}] AgentLlmPort is unavailable`);
+			if (request.takeProviderAttempt && !request.takeProviderAttempt()) return null;
 			config.onTransportAttempt?.({
 				callId: request.callId,
 				destination,
@@ -1147,13 +1176,18 @@ function createLlmClient(config) {
 	* result is returned untouched — it means no call was made, not that the model said nothing.
 	*/
 	const requestContentRetryingEmpty = async (state, request, systemContent, deadlineMs, prompt = request.prompt) => {
-		let content = await requestContent(state, request, systemContent, prompt);
+		const budget = {
+			remaining: EMPTY_REPLY_ATTEMPTS,
+			deadlineMs,
+			takeProviderAttempt: request.takeProviderAttempt
+		};
+		let content = await providerBudget.run(budget, () => requestContent(state, request, systemContent, prompt));
 		let usage = state.usage;
 		try {
 			checkMemoryOperation();
 			const attemptLimit = request.emptyReplyAttempts ?? EMPTY_REPLY_ATTEMPTS;
 			for (let attempt = 2; attempt <= attemptLimit; attempt += 1) {
-				if (content === null || content.raw.trim().length > 0) break;
+				if (content === null || content.raw.trim().length > 0 || budget.remaining <= 0) break;
 				const remainingMs = deadlineMs - performance.now();
 				if (remainingMs <= 0) break;
 				log.debug("Retrying empty model reply", {
@@ -1167,10 +1201,10 @@ function createLlmClient(config) {
 					function: "requestContentRetryingEmpty",
 					site_id: "shared.llm-client.requestContentRetryingEmpty.cd874fefff"
 				});
-				content = await requestContent(state, {
+				content = await providerBudget.run(budget, () => requestContent(state, {
 					...request,
 					timeoutMs: remainingMs
-				}, systemContent, prompt);
+				}, systemContent, prompt));
 				usage = addUsage(usage, state.usage);
 				checkMemoryOperation();
 			}
@@ -1195,6 +1229,7 @@ function createLlmClient(config) {
 			if (snoError !== void 0) throw snoError;
 			return content;
 		}
+		const remainingMs = Math.floor(deadlineMs - performance.now());
 		const originalLastError = state.error;
 		const originalLastUsage = state.usage;
 		const source = snoError instanceof Error ? snoError.message : originalLastError ?? "";
@@ -1206,21 +1241,25 @@ function createLlmClient(config) {
 		else if (!config.apiKey?.trim()) snoFailure = "missing_key";
 		const operationId = config.operationId ?? memoryOperationId();
 		let hostFailure = "empty_reply";
-		if (!config.agentPort) hostFailure = "port_unavailable";
+		if (remainingMs <= 0) hostFailure = "deadline_exhausted";
+		else if (!config.agentPort) hostFailure = "port_unavailable";
 		else if ([
 			"E10",
 			"P1",
 			"REM1"
 		].includes(request.callId) && !request.hostPrompt) hostFailure = "host_prompt_missing";
 		else try {
-			const host = await requestContent(state, request, systemContent, request.hostPrompt ?? request.prompt, "host");
+			const host = await requestContent(state, {
+				...request,
+				timeoutMs: remainingMs
+			}, systemContent, request.hostPrompt ?? request.prompt, "host");
 			if (host?.raw.trim()) {
 				log.info("Sno model call answered by host", {
 					call_id: request.callId,
 					operation_id: operationId,
 					sno_failure: snoFailure,
 					host_outcome: "answered",
-					timeout_ms: resolveRequestTimeoutMs(request)
+					timeout_ms: remainingMs
 				}, {
 					event_name: "memory.llm_client.fallback",
 					file: "packages/memory/src/model/llm-client.ts",
@@ -1232,7 +1271,7 @@ function createLlmClient(config) {
 			}
 			if (host === null) hostFailure = "host_error";
 		} catch (error) {
-			if (request.signal?.aborted) throw error;
+			if (request.signal?.aborted || error instanceof ModelCallRefusedError) throw error;
 			hostFailure = error instanceof LlmClientTerminalError ? error.category : "transport";
 		}
 		state.error = originalLastError;