@remnic/core 9.3.627 → 9.3.629

package/dist/types.d.ts

@@ -1,3 +1,3 @@
 import './types-BliCnURB.js';
 import './index-DJ9QWMw-.js';
-export { y as AccessTrackingEntry, A as ActiveRecallChatType, d as ActiveRecallModelFallbackPolicy, b as ActiveRecallPromptStyle, a as ActiveRecallQueryMode, c as ActiveRecallThinking, a6 as AgentAccessAuthToken, b2 as AgentAccessHttpConfig, b3 as AgentDefaultsConfig, b4 as AgentPersona, f as AgentPersonaModelConfig, b5 as BehaviorLoopAdjustment, e as BehaviorLoopPolicyState, b6 as BehaviorSignalDirection, B as BehaviorSignalEvent, b7 as BehaviorSignalType, aH as BootstrapOptions, aI as BootstrapResult, a2 as BriefingActiveThread, b8 as BriefingCalendarSourceError, aO as BriefingConfig, $ as BriefingFocus, _ as BriefingFollowup, aP as BriefingOpenCommitment, a4 as BriefingRecentEntity, a3 as BriefingResult, Z as BriefingSections, aQ as BriefingWindow, b9 as BufferEntryState, k as BufferState, o as BufferSurpriseEvent, N as BufferTurn, a1 as CalendarEvent, a0 as CalendarSource, ba as CaptureMode, S as Checkpoint, bb as CodexCliReasoningEffort, bc as CodexCompactionFlushMode, aR as CodexCompatConfig, bd as CodexConnectorConfig, aG as CodingContext, aL as CodingModeConfig, be as CompressionGuidelineActivationState, a8 as CompressionGuidelineOptimizerActionSummary, a7 as CompressionGuidelineOptimizerEventCounts, a9 as CompressionGuidelineOptimizerRuleUpdate, bf as CompressionGuidelineOptimizerSourceWindow, C as CompressionGuidelineOptimizerState, bg as ConfidenceTier, bh as ConsolidationAction, bi as ConsolidationItem, aJ as ConsolidationObservation, ag as ConsolidationResult, s as ContinuityImprovementLoop, r as ContinuityIncidentCloseInput, p as ContinuityIncidentOpenInput, q as ContinuityIncidentRecord, bj as ContinuityIncidentState, bk as ContinuityLoopCadence, u as ContinuityLoopReviewInput, bl as ContinuityLoopStatus, t as ContinuityLoopUpsertInput, bm as ContradictionScanConfig, as as ConversationThread, bn as CronConversationRecallMode, bo as CronRecallMode, aS as DEFAULT_RECALL_DISCLOSURE, ah as DaySummaryResult, bp as DreamingConfig, bq as DreamingNarrativePromptStyle, br as DreamsDeepSleepConfig, bs as DreamsLightSleepConfig, au as DreamsPhase, bt as DreamsPhaseStatus, bu as DreamsPhasesConfig, bv as DreamsRemConfig, av as DreamsRunResult, at as DreamsStatusResult, bw as EngramTraceEvent, w as EntityActivityEntry, x as EntityFile, ab as EntityMention, v as EntityRelationship, ac as EntitySchemaDefinition, ad as EntitySchemaSectionDefinition, E as EntityStructuredSection, aK as EntityTimelineEntry, aT as ExtractedFact, bx as ExtractedProcedureStep, by as ExtractedQuestion, bz as ExtractedReasoningTrace, bA as ExtractedReasoningTraceStep, bB as ExtractedRelationship, bC as ExtractionPassSource, af as ExtractionResult, F as FileHygieneConfig, G as GatewayConfig, bD as GitHubLiveConnectorConfig, bE as GmailLiveConnectorConfig, bF as GoogleDriveLiveConnectorConfig, bG as HeartbeatConfig, bH as HeartbeatDetectionMode, U as HourlySummary, J as IdentityInjectionMode, ae as ImportanceLevel, I as ImportanceScore, W as LifecycleState, L as LiveConnectorsConfig, bI as LlmTraceCallback, bJ as LlmTraceEvent, ak as MemoryActionEligibilityContext, aU as MemoryActionEligibilitySource, m as MemoryActionEvent, bK as MemoryActionOutcome, bL as MemoryActionPolicyDecision, al as MemoryActionPolicyResult, bM as MemoryActionStatus, aa as MemoryActionType, M as MemoryCategory, g as MemoryFile, i as MemoryFrontmatter, ai as MemoryIntent, n as MemoryLifecycleEvent, am as MemoryLifecycleEventType, an as MemoryLifecycleStateSummary, h as MemoryLink, bN as MemoryLinkType, aV as MemoryObservation, bO as MemoryOsPresetName, z as MemoryProjectionCurrentState, aW as MemoryScope, j as MemoryStatus, D as MemorySummary, l as MetaState, bP as ModelApi, bQ as ModelDefinitionConfig, bR as ModelProviderAuthMode, ao as ModelProviderConfig, bS as NamespacePolicy, ap as NativeKnowledgeConfig, bT as NativeKnowledgeFolderRuleConfig, bU as NativeKnowledgeObsidianVaultConfig, bV as NativeKnowledgeOpenClawWorkspaceConfig, bW as NotionLiveConnectorConfig, P as PluginConfig, Y as PolicyClass, bX as PrincipalFromSessionKeyMode, bY as PrincipalRule, bZ as ProceduralConfig, aj as QmdSearchExplain, Q as QmdSearchResult, b_ as QuestionEntry, aX as RECALL_DISCLOSURE_LEVELS, b$ as ReasoningEffort, K as RecallDisclosure, c0 as RecallPipelineConfig, H as RecallPlanMode, c1 as RecallSectionConfig, R as RecallTierExplain, c2 as RecallTraceEvent, c3 as RelevanceFeedback, aq as RetrievalTier, c4 as SPECULATIVE_TTL_DAYS, c5 as ScoredEntity, a5 as SecretRef, c6 as SemanticChunkingConfigShape, V as SessionObserverBandConfig, c7 as SignalLevel, ar as SignalScanResult, c8 as SlotBehaviorConfig, c9 as SlotMismatchMode, T as TopicScore, O as TranscriptEntry, ca as TriggerMode, X as VerificationState, cb as confidenceTier, b1 as isRecallDisclosure } from './types-D5VRAI04.js';
+export { y as AccessTrackingEntry, A as ActiveRecallChatType, d as ActiveRecallModelFallbackPolicy, b as ActiveRecallPromptStyle, a as ActiveRecallQueryMode, c as ActiveRecallThinking, a6 as AgentAccessAuthToken, b2 as AgentAccessHttpConfig, b3 as AgentDefaultsConfig, b4 as AgentPersona, f as AgentPersonaModelConfig, b5 as BehaviorLoopAdjustment, e as BehaviorLoopPolicyState, b6 as BehaviorSignalDirection, B as BehaviorSignalEvent, b7 as BehaviorSignalType, aH as BootstrapOptions, aI as BootstrapResult, V as BriefingActiveThread, b8 as BriefingCalendarSourceError, aO as BriefingConfig, Q as BriefingFocus, O as BriefingFollowup, aP as BriefingOpenCommitment, X as BriefingRecentEntity, W as BriefingResult, N as BriefingSections, aQ as BriefingWindow, b9 as BufferEntryState, k as BufferState, o as BufferSurpriseEvent, Y as BufferTurn, U as CalendarEvent, S as CalendarSource, ba as CaptureMode, $ as Checkpoint, bb as CodexCliReasoningEffort, bc as CodexCompactionFlushMode, aR as CodexCompatConfig, bd as CodexConnectorConfig, aG as CodingContext, aL as CodingModeConfig, be as CompressionGuidelineActivationState, a8 as CompressionGuidelineOptimizerActionSummary, a7 as CompressionGuidelineOptimizerEventCounts, a9 as CompressionGuidelineOptimizerRuleUpdate, bf as CompressionGuidelineOptimizerSourceWindow, C as CompressionGuidelineOptimizerState, bg as ConfidenceTier, bh as ConsolidationAction, bi as ConsolidationItem, aJ as ConsolidationObservation, ag as ConsolidationResult, s as ContinuityImprovementLoop, r as ContinuityIncidentCloseInput, p as ContinuityIncidentOpenInput, q as ContinuityIncidentRecord, bj as ContinuityIncidentState, bk as ContinuityLoopCadence, u as ContinuityLoopReviewInput, bl as ContinuityLoopStatus, t as ContinuityLoopUpsertInput, bm as ContradictionScanConfig, as as ConversationThread, bn as CronConversationRecallMode, bo as CronRecallMode, aS as DEFAULT_RECALL_DISCLOSURE, ah as DaySummaryResult, bp as DreamingConfig, bq as DreamingNarrativePromptStyle, br as DreamsDeepSleepConfig, bs as DreamsLightSleepConfig, au as DreamsPhase, bt as DreamsPhaseStatus, bu as DreamsPhasesConfig, bv as DreamsRemConfig, av as DreamsRunResult, at as DreamsStatusResult, bw as EngramTraceEvent, w as EntityActivityEntry, x as EntityFile, ab as EntityMention, v as EntityRelationship, ac as EntitySchemaDefinition, ad as EntitySchemaSectionDefinition, E as EntityStructuredSection, aK as EntityTimelineEntry, aT as ExtractedFact, bx as ExtractedProcedureStep, by as ExtractedQuestion, bz as ExtractedReasoningTrace, bA as ExtractedReasoningTraceStep, bB as ExtractedRelationship, bC as ExtractionPassSource, af as ExtractionResult, F as FileHygieneConfig, G as GatewayConfig, bD as GitHubLiveConnectorConfig, bE as GmailLiveConnectorConfig, bF as GoogleDriveLiveConnectorConfig, bG as HeartbeatConfig, bH as HeartbeatDetectionMode, a0 as HourlySummary, J as IdentityInjectionMode, ae as ImportanceLevel, I as ImportanceScore, a2 as LifecycleState, L as LiveConnectorsConfig, bI as LlmTraceCallback, bJ as LlmTraceEvent, ak as MemoryActionEligibilityContext, aU as MemoryActionEligibilitySource, m as MemoryActionEvent, bK as MemoryActionOutcome, bL as MemoryActionPolicyDecision, al as MemoryActionPolicyResult, bM as MemoryActionStatus, aa as MemoryActionType, M as MemoryCategory, g as MemoryFile, i as MemoryFrontmatter, ai as MemoryIntent, n as MemoryLifecycleEvent, am as MemoryLifecycleEventType, an as MemoryLifecycleStateSummary, h as MemoryLink, bN as MemoryLinkType, aV as MemoryObservation, bO as MemoryOsPresetName, z as MemoryProjectionCurrentState, aW as MemoryScope, j as MemoryStatus, D as MemorySummary, l as MetaState, bP as ModelApi, bQ as ModelDefinitionConfig, bR as ModelProviderAuthMode, ao as ModelProviderConfig, bS as NamespacePolicy, ap as NativeKnowledgeConfig, bT as NativeKnowledgeFolderRuleConfig, bU as NativeKnowledgeObsidianVaultConfig, bV as NativeKnowledgeOpenClawWorkspaceConfig, bW as NotionLiveConnectorConfig, P as PluginConfig, a4 as PolicyClass, bX as PrincipalFromSessionKeyMode, bY as PrincipalRule, bZ as ProceduralConfig, aj as QmdSearchExplain, Z as QmdSearchResult, b_ as QuestionEntry, aX as RECALL_DISCLOSURE_LEVELS, b$ as ReasoningEffort, K as RecallDisclosure, c0 as RecallPipelineConfig, H as RecallPlanMode, c1 as RecallSectionConfig, R as RecallTierExplain, c2 as RecallTraceEvent, c3 as RelevanceFeedback, aq as RetrievalTier, c4 as SPECULATIVE_TTL_DAYS, c5 as ScoredEntity, a5 as SecretRef, c6 as SemanticChunkingConfigShape, a1 as SessionObserverBandConfig, c7 as SignalLevel, ar as SignalScanResult, c8 as SlotBehaviorConfig, c9 as SlotMismatchMode, T as TopicScore, _ as TranscriptEntry, ca as TriggerMode, a3 as VerificationState, cb as confidenceTier, b1 as isRecallDisclosure } from './types-BCF2wqKa.js';

package/dist/utility-runtime.d.ts

@@ -1,5 +1,5 @@
 import { TierRoutingPolicy } from './tier-routing.js';
-import './types-D5VRAI04.js';
+import './types-BCF2wqKa.js';
 import './types-BliCnURB.js';
 import './index-DJ9QWMw-.js';
 import './lifecycle.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/core",
-  "version": "9.3.627",
+  "version": "9.3.629",
   "description": "Framework-agnostic Remnic memory engine — orchestrator, storage, extraction, search, trust zones",
   "type": "module",
   "main": "dist/index.js",

package/src/access-service.ts

@@ -1814,6 +1814,13 @@ export class EngramAccessService {
       openaiApiKey: config.openaiApiKey,
       openaiBaseUrl: config.openaiBaseUrl,
       model: config.model,
+      // Without a direct OpenAI key, route follow-ups through the configured
+      // LLM chain (gateway model source or local LLM) — same fallback every
+      // other LLM feature uses. A configured key keeps its precedence so
+      // existing deployments are unchanged.
+      followupGenerator: config.openaiApiKey
+        ? undefined
+        : this.orchestrator.briefingChainFollowupGenerator,
     });
 
     return {

package/src/briefing.ts

@@ -19,6 +19,7 @@ import { readFile } from "node:fs/promises";
 import os from "node:os";
 import path from "node:path";
 import { log } from "./logger.js";
+import { extractJsonCandidates } from "./json-extract.js";
 import { normalizeEntityName, StorageManager } from "./storage.js";
 import { readEnvVar, resolveHomeDir } from "./runtime/env.js";
 import type {
@@ -865,7 +866,8 @@ export async function buildBriefing(options: BuildBriefingOptions): Promise<Brie
   if (maxFollowups === 0 || options.allowLlm === false) {
     followupsUnavailableReason = "LLM follow-ups disabled by configuration";
   } else if (!options.openaiApiKey && !options.followupGenerator) {
-    followupsUnavailableReason = "OPENAI_API_KEY not configured";
+    followupsUnavailableReason =
+      'no LLM configured for follow-ups (set OPENAI_API_KEY, enable a local LLM, or use modelSource "gateway")';
   } else {
     try {
       const generator = options.followupGenerator ?? buildOpenAiFollowupGenerator({
@@ -1200,6 +1202,70 @@ function buildOpenAiFollowupGenerator(cfg: {
   };
 }
 
+/**
+ * Minimal chat-completion surface shared by `FallbackLlmClient` (gateway
+ * model chain) and `LocalLlmClient` (Ollama / OpenAI-compatible local
+ * endpoints). Matches `Orchestrator.fastLlmForRerank` so briefing
+ * follow-ups can ride the same routing as other fast-tier operations.
+ */
+export interface BriefingChainLlmClient {
+  chatCompletion(
+    messages: Array<{ role: "system" | "user" | "assistant"; content: string }>,
+    options?: {
+      temperature?: number;
+      maxTokens?: number;
+      timeoutMs?: number;
+      operation?: string;
+      priority?: "background" | "recall-critical";
+    },
+  ): Promise<{ content: string } | null>;
+}
+
+/**
+ * Build a follow-up generator backed by the configured LLM chain
+ * (gateway model source or local LLM) instead of a direct OpenAI key.
+ *
+ * Local models frequently wrap JSON in code fences or prose, so the
+ * response is run through `extractJsonCandidates` and the first candidate
+ * that parses into a valid `{ followups: [...] }` shape wins. Throws when
+ * the chain returns nothing or no candidate parses — `buildBriefing`
+ * catches and surfaces the message via `followupsUnavailableReason`.
+ */
+export function buildChainFollowupGenerator(
+  client: BriefingChainLlmClient,
+): BriefingFollowupGenerator {
+  return async ({ sections, windowLabel, maxFollowups }) => {
+    const prompt = buildFollowupPrompt(sections, windowLabel, maxFollowups);
+    const response = await client.chatCompletion(
+      [
+        { role: "system", content: FOLLOWUP_INSTRUCTIONS },
+        { role: "user", content: prompt },
+      ],
+      {
+        temperature: 0.2,
+        maxTokens: 512,
+        operation: "briefing-followups",
+        priority: "background",
+      },
+    );
+    if (!response?.content) {
+      throw new Error("LLM chain returned no response");
+    }
+    const candidates = extractJsonCandidates(response.content);
+    let lastError: unknown;
+    for (const candidate of candidates) {
+      try {
+        return parseFollowupResponse(candidate, maxFollowups);
+      } catch (err) {
+        lastError = err;
+      }
+    }
+    throw new Error(
+      `LLM chain response contained no valid followups JSON: ${stringifyError(lastError ?? "no JSON candidates found")}`,
+    );
+  };
+}
+
 const FOLLOWUP_INSTRUCTIONS = `You suggest short follow-up prompts for a daily context briefing.
 Return strict JSON of the form { "followups": [{ "text": "...", "rationale": "..." }] }.
 Rules:

package/src/index.ts

@@ -507,6 +507,8 @@ export {
   briefingFilename,
   FileCalendarSource,
   BRIEFING_FORMAT_ALLOWED,
+  buildChainFollowupGenerator,
+  type BriefingChainLlmClient,
   type BuildBriefingOptions,
   type BriefingFollowupGenerator,
   type ParsedBriefingWindow,

package/src/orchestrator.ts

@@ -1,4 +1,8 @@
 import { log } from "./logger.js";
+import {
+  buildChainFollowupGenerator,
+  type BriefingFollowupGenerator,
+} from "./briefing.js";
 import path from "node:path";
 import os from "node:os";
 import { createHash, randomBytes } from "node:crypto";
@@ -2516,6 +2520,28 @@ export class Orchestrator {
     };
   }
 
+  /**
+   * Build a briefing follow-up generator backed by the configured LLM chain
+   * (gateway model source or local LLM). Returns `undefined` when no chain
+   * is available so `buildBriefing` can surface a clear unavailable reason
+   * instead of failing at call time. Used by the access service and CLI as
+   * the fallback when no direct `openaiApiKey` is configured, so briefing
+   * follow-ups ride the same routing as every other fast-tier LLM feature.
+   */
+  get briefingChainFollowupGenerator(): BriefingFollowupGenerator | undefined {
+    // Plugin mode gates on `localLlmEnabled` alone: `LocalLlmClient.chatCompletion`
+    // returns null when the master switch is off, so `localLlmFastEnabled` by
+    // itself cannot serve requests (Cursor review on PR #1463).
+    const chainAvailable =
+      this.config.modelSource === "gateway"
+        ? this._fastGatewayLlm?.isAvailable(
+            this.config.fastGatewayAgentId || this.config.gatewayAgentId || undefined,
+          ) === true
+        : this.config.localLlmEnabled;
+    if (!chainAvailable) return undefined;
+    return buildChainFollowupGenerator(this.fastLlmForRerank);
+  }
+
   async initialize(): Promise<void> {
     // Recreate the deferred-ready gate on every initialize() call.
     // The same Orchestrator instance may be reused across stop/start cycles