@revealui/ai 0.3.0 → 0.4.0

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AAKH;;;;GAIG;AACH,wBAAgB,cAAc,IAAI,OAAO,CASxC;AAGD,cAAc,gBAAgB,CAAC;AAE/B,cAAc,kBAAkB,CAAC;AAIjC,cAAc,uBAAuB,CAAC;AAEtC,cAAc,sBAAsB,CAAC;AAErC,cAAc,iBAAiB,CAAC;AAChC,cAAc,0BAA0B,CAAC;AACzC,cAAc,yBAAyB,CAAC;AACxC,cAAc,kCAAkC,CAAC;AACjD,cAAc,wBAAwB,CAAC;AACvC,cAAc,oCAAoC,CAAC;AAEnD,cAAc,mBAAmB,CAAC;AAElC,cAAc,0BAA0B,CAAC;AACzC,cAAc,6BAA6B,CAAC;AAC5C,cAAc,uCAAuC,CAAC;AACtD,cAAc,iCAAiC,CAAC;AAChD,cAAc,4BAA4B,CAAC;AAC3C,cAAc,sCAAsC,CAAC;AACrD,cAAc,iCAAiC,CAAC;AAEhD,cAAc,mBAAmB,CAAC;AAElC,cAAc,sBAAsB,CAAC;AAErC,cAAc,iBAAiB,CAAC;AAChC,cAAc,yBAAyB,CAAC;AACxC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC;AACpC,cAAc,yBAAyB,CAAC;AACxC,cAAc,sBAAsB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AAGH,cAAc,gBAAgB,CAAC;AAE/B,cAAc,kBAAkB,CAAC;AAIjC,cAAc,uBAAuB,CAAC;AAEtC,cAAc,sBAAsB,CAAC;AAErC,cAAc,iBAAiB,CAAC;AAChC,cAAc,0BAA0B,CAAC;AACzC,cAAc,yBAAyB,CAAC;AACxC,cAAc,oCAAoC,CAAC;AACnD,cAAc,kCAAkC,CAAC;AACjD,cAAc,wBAAwB,CAAC;AACvC,cAAc,oCAAoC,CAAC;AAEnD,cAAc,mBAAmB,CAAC;AAElC,cAAc,0BAA0B,CAAC;AACzC,cAAc,6BAA6B,CAAC;AAC5C,cAAc,uCAAuC,CAAC;AACtD,cAAc,iCAAiC,CAAC;AAChD,cAAc,4BAA4B,CAAC;AAC3C,cAAc,sCAAsC,CAAC;AACrD,cAAc,iCAAiC,CAAC;AAEhD,cAAc,mBAAmB,CAAC;AAElC,cAAc,sBAAsB,CAAC;AAErC,cAAc,iBAAiB,CAAC;AAChC,cAAc,yBAAyB,CAAC;AACxC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,wBAAwB,CAAC;AACvC,cAAc,4BAA4B,CAAC;AAC3C,cAAc,uBAAuB,CAAC;AACtC,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,qBAAqB,CAAC;AACpC,cAAc,yBAAyB,CAAC;AACxC,cAAc,sBAAsB,CAAC"}

package/dist/index.js

@@ -45,21 +45,6 @@
  *
  * @packageDocumentation
  */
-import { isFeatureEnabled } from '@revealui/core/features';
-import { logger } from '@revealui/core/observability/logger';
-/**
- * Check if the AI package is licensed for use.
- * Returns false with a warning log if no Pro/Enterprise license is active.
- * Call at application startup to verify licensing.
- */
-export function checkAiLicense() {
-    if (!isFeatureEnabled('ai')) {
-        logger.warn('[@revealui/ai] AI features require a Pro or Enterprise license. ' +
-            'Visit https://revealui.com/pricing for details.');
-        return false;
-    }
-    return true;
-}
 // Re-export A2A protocol
 export * from './a2a/index.js';
 // Re-export audit trail
@@ -74,6 +59,7 @@ export * from './ingestion/index.js';
 export * from './llm/client.js';
 export * from './llm/provider-health.js';
 export * from './llm/providers/base.js';
+export * from './llm/providers/inference-snaps.js';
 export * from './llm/providers/openai-compat.js';
 export * from './llm/token-counter.js';
 export * from './llm/workspace-provider-config.js';
@@ -96,6 +82,9 @@ export * from './tools/base.js';
 export * from './tools/deduplicator.js';
 export * from './tools/document-summarizer.js';
 export * from './tools/mcp-adapter.js';
+export * from './tools/mcp-elicitation.js';
+export * from './tools/mcp-events.js';
+export * from './tools/mcp-sampling.js';
 export * from './tools/memory/index.js';
 export * from './tools/registry.js';
 export * from './tools/ticket-tools.js';

package/dist/llm/client.d.ts

@@ -4,12 +4,13 @@
  * Single interface for all LLM providers with fallback and rate limiting
  */
 import type { Database } from '@revealui/db/client';
+import { CircuitBreaker } from '@revealui/resilience';
 import type { AuditStore } from '../audit/store.js';
 import type { ProviderHealthMonitor } from './provider-health.js';
 import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMResponse, LLMStreamOptions, Message } from './providers/base.js';
 import { type CacheStats, type ResponseCacheOptions } from './response-cache.js';
 import { type SemanticCacheOptions, type SemanticCacheStats } from './semantic-cache.js';
-export type LLMProviderType = 'vultr' | 'groq' | 'ollama' | 'huggingface' | 'inference-snaps';
+export type LLMProviderType = 'groq' | 'ollama' | 'huggingface' | 'inference-snaps';
 export interface LLMClientConfig {
     provider: LLMProviderType;
     apiKey: string;
@@ -46,6 +47,10 @@ export interface LLMClientConfig {
     semanticCacheOptions?: SemanticCacheOptions;
     /** Optional health monitor  -  records latency + error rate per provider */
     healthMonitor?: ProviderHealthMonitor;
+    /** Circuit breaker failure threshold before tripping (default: 5) */
+    circuitBreakerFailureThreshold?: number;
+    /** Circuit breaker reset timeout in ms before half-open probe (default: 30000) */
+    circuitBreakerResetTimeout?: number;
 }
 export declare class LLMClient {
     private provider;
@@ -56,6 +61,8 @@ export declare class LLMClient {
     private responseCache?;
     private semanticCache?;
     private healthMonitor?;
+    private circuitBreaker;
+    private fallbackCircuitBreaker?;
     /** Tracks the last resolved API key so we only recreate the provider when it changes */
     private currentApiKey;
     constructor(config: LLMClientConfig);
@@ -78,6 +85,21 @@ export declare class LLMClient {
         tokens: number;
         estimatedCostUsd: number;
     };
+    /**
+     * Get circuit breaker stats for the primary and fallback providers.
+     */
+    getCircuitBreakerStats(): {
+        primary: {
+            name: string;
+            state: string;
+            stats: ReturnType<CircuitBreaker['getStats']>;
+        };
+        fallback?: {
+            name: string;
+            state: string;
+            stats: ReturnType<CircuitBreaker['getStats']>;
+        };
+    };
     /**
      * Get the provider health monitor if one was configured.
      */
@@ -107,13 +129,20 @@ export declare class LLMClient {
  * Create an LLM client from environment variables.
  *
  * When LLM_PROVIDER is not set, auto-detects the provider by checking env vars
- * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA.
+ * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA. If none are set, defaults
+ * to Inference Snaps at http://localhost:9090/v1 (Ubuntu local) and emits a
+ * one-line stderr warning so the implicit default is discoverable in logs.
  *
  * All providers use OpenAI-compatible APIs. No proprietary provider SDKs.
  *
+ * Canonical Inference Snaps is the reference local provider on Ubuntu — offline,
+ * silicon-optimized, no API key required. See `providers/inference-snaps.ts` for
+ * install docs (`sudo snap install gemma3`, etc.).
+ *
  * Provider defaults:
- *   groq   → qwen/qwen3-32b
- *   ollama → gemma4:e2b
+ *   inference-snaps → gemma3       (base URL defaults to http://localhost:9090/v1)
+ *   groq            → qwen/qwen3-32b
+ *   ollama          → gemma4:e2b   (base URL defaults to http://localhost:11434)
  */
 export declare function createLLMClientFromEnv(): LLMClient;
 /**

package/dist/llm/client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/llm/client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAKH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAIpD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAClE,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,qBAAqB,CAAC;AAS7B,OAAO,EAAE,KAAK,UAAU,EAAiB,KAAK,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAChG,OAAO,EAEL,KAAK,oBAAoB,EACzB,KAAK,kBAAkB,EACxB,MAAM,qBAAqB,CAAC;AAG7B,MAAM,MAAM,eAAe,GAAG,OAAO,GAAG,MAAM,GAAG,QAAQ,GAAG,aAAa,GAAG,iBAAiB,CAAC;AAE9F,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,eAAe,CAAC;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;IACjC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,gBAAgB,CAAC,EAAE,eAAe,CAAC;IACnC,SAAS,CAAC,EAAE;QACV,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,cAAc,CAAC,EAAE,MAAM,CAAC;KACzB,CAAC;IACF,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B,wEAAwE;IACxE,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,6BAA6B;IAC7B,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,iEAAiE;IACjE,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,6BAA6B;IAC7B,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,4EAA4E;IAC5E,aAAa,CAAC,EAAE,qBAAqB,CAAC;CACvC;AAQD,qBAAa,SAAS;IACpB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,gBAAgB,CAAC,CAAc;IACvC,OAAO,CAAC,qBAAqB,CAAC,CAAc;IAC5C,OAAO,CAAC,MAAM,CAAkB;IAChC,OAAO,CAAC,cAAc,CAAiB;IACvC,OAAO,CAAC,aAAa,CAAC,CAAgB;IACtC,OAAO,CAAC,aAAa,CAAC,CAAgB;IACtC,OAAO,CAAC,aAAa,CAAC,CAAwB;IAC9C,wFAAwF;IACxF,OAAO,CAAC,aAAa,CAAS;gBAElB,MAAM,EAAE,eAAe;IA8CnC,OAAO,CAAC,cAAc;IAuBtB;;;OAGG;YACW,uBAAuB;IAmBrC,OAAO,CAAC,cAAc;IAoCtB,OAAO,CAAC,aAAa;IAMf,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAmHzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IA2B5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IA0BvF;;;OAGG;IACH,eAAe,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,gBAAgB,EAAE,MAAM,CAAA;KAAE;IAIlF;;OAEG;IACH,gBAAgB,IAAI,qBAAqB,GAAG,SAAS;IAIrD;;;;OAIG;IACH,qBAAqB,IAAI,UAAU,GAAG,SAAS;IAI/C;;OAEG;IACH,kBAAkB,IAAI,IAAI;IAI1B;;;;OAIG;IACH,qBAAqB,IAAI,kBAAkB,GAAG,SAAS;IAIvD;;OAEG;IACH,kBAAkB,IAAI,IAAI;CAG3B;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,sBAAsB,IAAI,SAAS,CAoElD;AAED;;;;;;;;;;;;GAYG;AACH,wBAAsB,sBAAsB,CAC1C,MAAM,EAAE,MAAM,EACd,EAAE,EAAE,QAAQ,EACZ,UAAU,CAAC,EAAE,UAAU,GACtB,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC,CAiD3B"}
+{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/llm/client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAKH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAGpD,OAAO,EACL,cAAc,EAGf,MAAM,sBAAsB,CAAC;AAE9B,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAClE,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,qBAAqB,CAAC;AAQ7B,OAAO,EAAE,KAAK,UAAU,EAAiB,KAAK,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAChG,OAAO,EAEL,KAAK,oBAAoB,EACzB,KAAK,kBAAkB,EACxB,MAAM,qBAAqB,CAAC;AAG7B,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,QAAQ,GAAG,aAAa,GAAG,iBAAiB,CAAC;AAEpF,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,eAAe,CAAC;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;IACjC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,gBAAgB,CAAC,EAAE,eAAe,CAAC;IACnC,SAAS,CAAC,EAAE;QACV,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,cAAc,CAAC,EAAE,MAAM,CAAC;KACzB,CAAC;IACF,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B,wEAAwE;IACxE,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,6BAA6B;IAC7B,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,iEAAiE;IACjE,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,6BAA6B;IAC7B,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,4EAA4E;IAC5E,aAAa,CAAC,EAAE,qBAAqB,CAAC;IACtC,qEAAqE;IACrE,8BAA8B,CAAC,EAAE,MAAM,CAAC;IACxC,kFAAkF;IAClF,0BAA0B,CAAC,EAAE,MAAM,CAAC;CACrC;AAQD,qBAAa,SAAS;IACpB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,gBAAgB,CAAC,CAAc;IACvC,OAAO,CAAC,qBAAqB,CAAC,CAAc;IAC5C,OAAO,CAAC,MAAM,CAAkB;IAChC,OAAO,CAAC,cAAc,CAAiB;IACvC,OAAO,CAAC,aAAa,CAAC,CAAgB;IACtC,OAAO,CAAC,aAAa,CAAC,CAAgB;IACtC,OAAO,CAAC,aAAa,CAAC,CAAwB;IAC9C,OAAO,CAAC,cAAc,CAAiB;IACvC,OAAO,CAAC,sBAAsB,CAAC,CAAiB;IAChD,wFAAwF;IACxF,OAAO,CAAC,aAAa,CAAS;gBAElB,MAAM,EAAE,eAAe;IA0DnC,OAAO,CAAC,cAAc;IAoBtB;;;OAGG;YACW,uBAAuB;IAmBrC,OAAO,CAAC,cAAc;IAoCtB,OAAO,CAAC,aAAa;IAMf,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAwHzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IA8B5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAkCvF;;;OAGG;IACH,eAAe,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,gBAAgB,EAAE,MAAM,CAAA;KAAE;IAIlF;;OAEG;IACH,sBAAsB,IAAI;QACxB,OAAO,EAAE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,UAAU,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC,CAAA;SAAE,CAAC;QACxF,QAAQ,CAAC,EAAE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,UAAU,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC,CAAA;SAAE,CAAC;KAC3F;IAkBD;;OAEG;IACH,gBAAgB,IAAI,qBAAqB,GAAG,SAAS;IAIrD;;;;OAIG;IACH,qBAAqB,IAAI,UAAU,GAAG,SAAS;IAI/C;;OAEG;IACH,kBAAkB,IAAI,IAAI;IAI1B;;;;OAIG;IACH,qBAAqB,IAAI,kBAAkB,GAAG,SAAS;IAIvD;;OAEG;IACH,kBAAkB,IAAI,IAAI;CAG3B;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,sBAAsB,IAAI,SAAS,CAiElD;AAED;;;;;;;;;;;;GAYG;AACH,wBAAsB,sBAAsB,CAC1C,MAAM,EAAE,MAAM,EACd,EAAE,EAAE,QAAQ,EACZ,UAAU,CAAC,EAAE,UAAU,GACtB,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC,CAiD3B"}

package/dist/llm/client.js

@@ -5,11 +5,11 @@
  */
 import { decryptApiKey } from '@revealui/db/crypto';
 import { tenantProviderConfigs, userApiKeys } from '@revealui/db/schema';
+import { CircuitBreaker, CircuitBreakerOpenError, } from '@revealui/resilience';
 import { and, eq } from 'drizzle-orm';
 import { GroqProvider } from './providers/groq.js';
 import { InferenceSnapsProvider, } from './providers/inference-snaps.js';
 import { OllamaProvider } from './providers/ollama.js';
-import { VultrProvider } from './providers/vultr.js';
 import { ResponseCache } from './response-cache.js';
 import { SemanticCache, } from './semantic-cache.js';
 import { estimateRequest as _estimateRequestTokens } from './token-counter.js';
@@ -22,6 +22,8 @@ export class LLMClient {
     responseCache;
     semanticCache;
     healthMonitor;
+    circuitBreaker;
+    fallbackCircuitBreaker;
     /** Tracks the last resolved API key so we only recreate the provider when it changes */
     currentApiKey;
     constructor(config) {
@@ -42,6 +44,17 @@ export class LLMClient {
         }
         // Wire health monitor if provided
         this.healthMonitor = config.healthMonitor;
+        // Per-provider circuit breakers — isolate outages so one provider's failure
+        // doesn't block calls to a different provider
+        const cbConfig = {
+            failureThreshold: config.circuitBreakerFailureThreshold ?? 5,
+            resetTimeout: config.circuitBreakerResetTimeout ?? 30_000,
+            successThreshold: 2,
+        };
+        this.circuitBreaker = new CircuitBreaker(cbConfig);
+        if (config.fallbackProvider) {
+            this.fallbackCircuitBreaker = new CircuitBreaker(cbConfig);
+        }
         // Wire dedicated embed provider if supplied
         this.embedProviderOverride = config.embedProvider;
         // Create primary provider
@@ -65,8 +78,6 @@ export class LLMClient {
     }
     createProvider(type, config) {
         switch (type) {
-            case 'vultr':
-                return new VultrProvider(config);
             case 'groq':
                 return new GroqProvider(config);
             case 'ollama':
@@ -185,7 +196,7 @@ export class LLMClient {
         const callStart = Date.now();
         try {
             this.recordRequest();
-            const response = await this.provider.chat(messages, options);
+            const response = await this.circuitBreaker.execute(() => this.provider.chat(messages, options));
             this.healthMonitor?.recordCall(this.config.provider, Date.now() - callStart);
             // Store in semantic cache (if enabled)
             if (this.semanticCache) {
@@ -215,11 +226,14 @@ export class LLMClient {
             this.healthMonitor?.recordCall(this.config.provider, Date.now() - callStart, error instanceof Error ? error : new Error(String(error)));
             // Try fallback if available
             if (this.fallbackProvider && this.config.fallbackProvider) {
+                const fp = this.fallbackProvider;
                 const fallbackStart = Date.now();
                 try {
-                    const fallbackResponse = await this.fallbackProvider.chat(messages, options);
+                    const fb = this.fallbackCircuitBreaker
+                        ? await this.fallbackCircuitBreaker.execute(() => fp.chat(messages, options))
+                        : await fp.chat(messages, options);
                     this.healthMonitor?.recordCall(this.config.fallbackProvider, Date.now() - fallbackStart);
-                    return fallbackResponse;
+                    return fb;
                 }
                 catch (fallbackError) {
                     this.healthMonitor?.recordCall(this.config.fallbackProvider, Date.now() - fallbackStart, fallbackError instanceof Error ? fallbackError : new Error(String(fallbackError)));
@@ -238,13 +252,16 @@ export class LLMClient {
         const embedProvider = this.embedProviderOverride ?? this.provider;
         try {
             this.recordRequest();
-            return await embedProvider.embed(text, options);
+            return await this.circuitBreaker.execute(() => embedProvider.embed(text, options));
         }
         catch (error) {
             // Try fallback if available (only when using the primary provider path)
             if (!this.embedProviderOverride && this.fallbackProvider) {
+                const fp = this.fallbackProvider;
                 try {
-                    return await this.fallbackProvider.embed(text, options);
+                    return this.fallbackCircuitBreaker
+                        ? await this.fallbackCircuitBreaker.execute(() => fp.embed(text, options))
+                        : await fp.embed(text, options);
                 }
                 catch {
                     throw new Error(`Both primary and fallback providers failed: ${error instanceof Error ? error.message : String(error)}`);
@@ -260,6 +277,11 @@ export class LLMClient {
             throw new Error('Rate limit exceeded');
         }
         try {
+            // Circuit breaker check — streaming can't use execute() wrapper
+            // so we check state and record outcomes manually
+            if (this.circuitBreaker.isOpen()) {
+                throw new CircuitBreakerOpenError(`llm-${this.config.provider}`);
+            }
             this.recordRequest();
             yield* this.provider.stream(messages, options);
         }
@@ -267,6 +289,9 @@ export class LLMClient {
             // Try fallback if available
             if (this.fallbackProvider) {
                 try {
+                    if (this.fallbackCircuitBreaker?.isOpen()) {
+                        throw new CircuitBreakerOpenError(`llm-${this.config.fallbackProvider}`);
+                    }
                     yield* this.fallbackProvider.stream(messages, options);
                 }
                 catch {
@@ -285,6 +310,25 @@ export class LLMClient {
     estimateRequest(messages) {
         return _estimateRequestTokens(messages, this.config.model ?? '');
     }
+    /**
+     * Get circuit breaker stats for the primary and fallback providers.
+     */
+    getCircuitBreakerStats() {
+        return {
+            primary: {
+                name: `llm-${this.config.provider}`,
+                state: this.circuitBreaker.getState(),
+                stats: this.circuitBreaker.getStats(),
+            },
+            fallback: this.fallbackCircuitBreaker && this.config.fallbackProvider
+                ? {
+                    name: `llm-${this.config.fallbackProvider}`,
+                    state: this.fallbackCircuitBreaker.getState(),
+                    stats: this.fallbackCircuitBreaker.getStats(),
+                }
+                : undefined,
+        };
+    }
     /**
      * Get the provider health monitor if one was configured.
      */
@@ -324,13 +368,20 @@ export class LLMClient {
  * Create an LLM client from environment variables.
  *
  * When LLM_PROVIDER is not set, auto-detects the provider by checking env vars
- * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA.
+ * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA. If none are set, defaults
+ * to Inference Snaps at http://localhost:9090/v1 (Ubuntu local) and emits a
+ * one-line stderr warning so the implicit default is discoverable in logs.
  *
  * All providers use OpenAI-compatible APIs. No proprietary provider SDKs.
  *
+ * Canonical Inference Snaps is the reference local provider on Ubuntu — offline,
+ * silicon-optimized, no API key required. See `providers/inference-snaps.ts` for
+ * install docs (`sudo snap install gemma3`, etc.).
+ *
  * Provider defaults:
- *   groq   → qwen/qwen3-32b
- *   ollama → gemma4:e2b
+ *   inference-snaps → gemma3       (base URL defaults to http://localhost:9090/v1)
+ *   groq            → qwen/qwen3-32b
+ *   ollama          → gemma4:e2b   (base URL defaults to http://localhost:11434)
  */
 export function createLLMClientFromEnv() {
     // Auto-detect provider when LLM_PROVIDER is not explicitly set
@@ -348,18 +399,14 @@ export function createLLMClientFromEnv() {
         provider = 'ollama';
     }
     else {
-        throw new Error('No LLM provider configured. Set one of: OLLAMA_BASE_URL (local Ollama), ' +
-            'INFERENCE_SNAPS_BASE_URL (local snap), GROQ_API_KEY (cloud). ' +
-            'Alternatively, set LLM_PROVIDER explicitly.');
+        // Zero-config Ubuntu default: assume Inference Snaps on the standard
+        // local port. Operator sees the warning once at boot.
+        provider = 'inference-snaps';
     }
     let apiKey;
     let baseURL;
     let defaultModel;
-    if (provider === 'vultr') {
-        apiKey = process.env.VULTR_API_KEY;
-        baseURL = process.env.VULTR_BASE_URL;
-    }
-    else if (provider === 'huggingface') {
+    if (provider === 'huggingface') {
         apiKey = process.env.HF_TOKEN;
         baseURL = process.env.HF_MODEL_URL;
     }
@@ -377,12 +424,14 @@ export function createLLMClientFromEnv() {
     }
     else if (provider === 'inference-snaps') {
         apiKey = 'inference-snaps'; // inference-snaps ignores the API key
-        baseURL = process.env.INFERENCE_SNAPS_BASE_URL;
+        // Defaults to Canonical's Inference Snap local service on port 9090; override
+        // via INFERENCE_SNAPS_BASE_URL when the snap listens on a non-default port.
+        baseURL = process.env.INFERENCE_SNAPS_BASE_URL ?? 'http://localhost:9090/v1';
         defaultModel = 'gemma3';
     }
     if (!apiKey) {
         throw new Error(`API key not found for provider "${provider}". Set the corresponding env var ` +
-            `(INFERENCE_SNAPS_BASE_URL, GROQ_API_KEY, OLLAMA_BASE_URL, VULTR_API_KEY, or HF_TOKEN).`);
+            `(INFERENCE_SNAPS_BASE_URL, GROQ_API_KEY, OLLAMA_BASE_URL, or HF_TOKEN).`);
     }
     return new LLMClient({
         provider,

package/dist/llm/key-validator.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"key-validator.d.ts","sourceRoot":"","sources":["../../src/llm/key-validator.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,MAAM,MAAM,wBAAwB,GAAG;IAAE,KAAK,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,KAAK,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAezF;;;;;;;;GAQG;AACH,wBAAsB,mBAAmB,CACvC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,wBAAwB,CAAC,CAuEnC"}
+{"version":3,"file":"key-validator.d.ts","sourceRoot":"","sources":["../../src/llm/key-validator.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,MAAM,MAAM,wBAAwB,GAAG;IAAE,KAAK,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,KAAK,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAezF;;;;;;;;GAQG;AACH,wBAAsB,mBAAmB,CACvC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,wBAAwB,CAAC,CA2CnC"}

package/dist/llm/key-validator.js

@@ -41,22 +41,6 @@ export async function validateProviderKey(provider, apiKey) {
                 // Any other non-OK status (429, 500 etc.)  -  treat as reachable but unknown
                 return { valid: false, error: `Groq validation failed: HTTP ${res.status}` };
             }
-            case 'anthropic': {
-                // Anthropic has no free read endpoint. Validate by key format only.
-                // Valid Anthropic keys start with "sk-ant-api".
-                if (!apiKey.startsWith('sk-ant-api')) {
-                    return { valid: false, error: 'Anthropic API key must start with "sk-ant-api"' };
-                }
-                return { valid: true };
-            }
-            case 'openai': {
-                // Validate by format  -  keys start with "sk-"
-                // (Per LLM policy, OpenAI API calls are blocked until we have revenue.)
-                if (!apiKey.startsWith('sk-')) {
-                    return { valid: false, error: 'OpenAI API key must start with "sk-"' };
-                }
-                return { valid: true };
-            }
             case 'huggingface': {
                 const res = await probeFetch('https://huggingface.co/api/whoami-v2', {
                     headers: { Authorization: `Bearer ${apiKey}` },
@@ -68,17 +52,6 @@ export async function validateProviderKey(provider, apiKey) {
                 }
                 return { valid: false, error: `HuggingFace validation failed: HTTP ${res.status}` };
             }
-            case 'vultr': {
-                // Vultr Serverless Inference API (OpenAI-compatible)
-                const res = await probeFetch('https://api.vultrinference.com/v1/models', {
-                    headers: { Authorization: `Bearer ${apiKey}` },
-                });
-                if (res.ok)
-                    return { valid: true };
-                if (res.status === 401)
-                    return { valid: false, error: 'Invalid Vultr API key' };
-                return { valid: false, error: `Vultr validation failed: HTTP ${res.status}` };
-            }
             case 'ollama': {
                 // Ollama is local  -  we cannot reliably probe it from the server.
                 // Accept the key as-is (Ollama doesn't use API keys anyway).

package/dist/llm/providers/inference-snaps.d.ts

@@ -4,11 +4,13 @@
  * Local inference via Canonical's inference-snaps OpenAI-compatible API.
  * No API key required. Zero cost, fully offline, hardware-optimized.
  *
- * Supported models (snaps):
- *   gemma3          -  general LLM + vision (text/image in, text out)
- *   deepseek-r1     -  reasoning LLM
- *   qwen-vl         -  vision-language model (image + text)
- *   nemotron-nano   -  general LLM (reasoning + non-reasoning)
+ * Supported models (snaps), per Canonical's official catalog at
+ * https://documentation.ubuntu.com/inference-snaps/reference/snaps/ :
+ *   gemma3                 -  general LLM + vision (text/image in, text out)
+ *   deepseek-r1            -  reasoning LLM (chat completions)
+ *   qwen-vl                -  vision-language model (image + text)
+ *   nemotron-3-nano        -  general LLM (reasoning + non-reasoning, text only)
+ *   nemotron-3-nano-omni   -  multimodal (text/image/video/audio in, text out)
  *
  * Install a model:
  *   sudo snap install gemma3

package/dist/llm/providers/inference-snaps.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"inference-snaps.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/inference-snaps.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,4BAA6B,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IACrF,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC;IAChB,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,sBAAuB,YAAW,WAAW;IACxD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,4BAA4B;IAahD,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}
+{"version":3,"file":"inference-snaps.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/inference-snaps.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,4BAA6B,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IACrF,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC;IAChB,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,sBAAuB,YAAW,WAAW;IACxD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,4BAA4B;IAahD,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}

package/dist/llm/providers/inference-snaps.js

@@ -4,11 +4,13 @@
  * Local inference via Canonical's inference-snaps OpenAI-compatible API.
  * No API key required. Zero cost, fully offline, hardware-optimized.
  *
- * Supported models (snaps):
- *   gemma3          -  general LLM + vision (text/image in, text out)
- *   deepseek-r1     -  reasoning LLM
- *   qwen-vl         -  vision-language model (image + text)
- *   nemotron-nano   -  general LLM (reasoning + non-reasoning)
+ * Supported models (snaps), per Canonical's official catalog at
+ * https://documentation.ubuntu.com/inference-snaps/reference/snaps/ :
+ *   gemma3                 -  general LLM + vision (text/image in, text out)
+ *   deepseek-r1            -  reasoning LLM (chat completions)
+ *   qwen-vl                -  vision-language model (image + text)
+ *   nemotron-3-nano        -  general LLM (reasoning + non-reasoning, text only)
+ *   nemotron-3-nano-omni   -  multimodal (text/image/video/audio in, text out)
  *
  * Install a model:
  *   sudo snap install gemma3

package/dist/llm/providers/openai-compat.d.ts

@@ -2,7 +2,7 @@
  * OpenAI-Compatible Provider
  *
  * Base implementation for any LLM API that follows the OpenAI chat/completions
- * format. Used by: Ollama, Groq, Inference Snaps, Vultr.
+ * format. Used by: Ollama, Groq, Inference Snaps.
  * NOT for direct OpenAI usage  -  RevealUI uses open-source models only.
  */
 import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';

package/dist/llm/providers/openai-compat.js

@@ -2,7 +2,7 @@
  * OpenAI-Compatible Provider
  *
  * Base implementation for any LLM API that follows the OpenAI chat/completions
- * format. Used by: Ollama, Groq, Inference Snaps, Vultr.
+ * format. Used by: Ollama, Groq, Inference Snaps.
  * NOT for direct OpenAI usage  -  RevealUI uses open-source models only.
  */
 const authorizationHeader = 'Authorization';

package/dist/llm/response-cache.d.ts

@@ -5,8 +5,7 @@
  * Provides 100% cost savings on cache hits vs Anthropic's 90% token-level savings.
  *
  * Use cases:
- * - Vultr (doesn't have prompt caching)
- * - OpenAI (doesn't have prompt caching)
+ * - Groq (doesn't have prompt caching)
  * - Any provider without built-in caching
  * - Supplement to Anthropic prompt caching for exact duplicates
  *

package/dist/llm/response-cache.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"response-cache.d.ts","sourceRoot":"","sources":["../../src/llm/response-cache.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAIH,OAAO,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAE7E,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;CACH;AAED,MAAM,WAAW,oBAAoB;IACnC,yDAAyD;IACzD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,wDAAwD;IACxD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,uDAAuD;IACvD,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,qBAAa,aAAa;IACxB,OAAO,CAAC,KAAK,CAAmC;IAChD,OAAO,CAAC,KAAK,CAAsD;IACnE,OAAO,CAAC,WAAW,CAAU;gBAEjB,OAAO,GAAE,oBAAyB;IAqB9C;;;;;;;OAOG;IACH,WAAW,CACT,QAAQ,EAAE,OAAO,EAAE,EACnB,OAAO,CAAC,EAAE;QACR,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;QACzB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,GACA,MAAM;IAqBT;;;;OAIG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS;IAiB5C;;OAEG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,EAAE,QAAQ,EAAE,cAAc,GAAG,IAAI;IAShD;;OAEG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO;IAIzB;;OAEG;IACH,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO;IAI5B;;OAEG;IACH,KAAK,IAAI,IAAI;IASb;;OAEG;IACH,QAAQ,IAAI,UAAU;IAatB;;OAEG;IACH,UAAU,IAAI,IAAI;IAQlB;;OAEG;IACH,IAAI,IAAI,IAAI,MAAM,CAEjB;IAED;;OAEG;IACH,IAAI,OAAO,IAAI,MAAM,CAEpB;CACF;AAQD;;;;;;;;;GASG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,CAAC,EAAE,oBAAoB,GAAG,aAAa,CAKpF;AAED;;GAEG;AACH,wBAAgB,wBAAwB,IAAI,IAAI,CAI/C;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,6BAA6B,CAC3C,KAAK,EAAE,UAAU,EACjB,OAAO,EAAE;IACP,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB,GACA;IACD,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;CACzB,CAcA"}
+{"version":3,"file":"response-cache.d.ts","sourceRoot":"","sources":["../../src/llm/response-cache.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAIH,OAAO,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAE7E,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;CACH;AAED,MAAM,WAAW,oBAAoB;IACnC,yDAAyD;IACzD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,wDAAwD;IACxD,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,uDAAuD;IACvD,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,qBAAa,aAAa;IACxB,OAAO,CAAC,KAAK,CAAmC;IAChD,OAAO,CAAC,KAAK,CAAsD;IACnE,OAAO,CAAC,WAAW,CAAU;gBAEjB,OAAO,GAAE,oBAAyB;IAqB9C;;;;;;;OAOG;IACH,WAAW,CACT,QAAQ,EAAE,OAAO,EAAE,EACnB,OAAO,CAAC,EAAE;QACR,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;QACzB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,GACA,MAAM;IAqBT;;;;OAIG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS;IAiB5C;;OAEG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,EAAE,QAAQ,EAAE,cAAc,GAAG,IAAI;IAShD;;OAEG;IACH,GAAG,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO;IAIzB;;OAEG;IACH,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO;IAI5B;;OAEG;IACH,KAAK,IAAI,IAAI;IASb;;OAEG;IACH,QAAQ,IAAI,UAAU;IAatB;;OAEG;IACH,UAAU,IAAI,IAAI;IAQlB;;OAEG;IACH,IAAI,IAAI,IAAI,MAAM,CAEjB;IAED;;OAEG;IACH,IAAI,OAAO,IAAI,MAAM,CAEpB;CACF;AAQD;;;;;;;;;GASG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,CAAC,EAAE,oBAAoB,GAAG,aAAa,CAKpF;AAED;;GAEG;AACH,wBAAgB,wBAAwB,IAAI,IAAI,CAI/C;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,6BAA6B,CAC3C,KAAK,EAAE,UAAU,EACjB,OAAO,EAAE;IACP,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB,GACA;IACD,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;CACzB,CAcA"}

package/dist/llm/response-cache.js

@@ -5,8 +5,7 @@
  * Provides 100% cost savings on cache hits vs Anthropic's 90% token-level savings.
  *
  * Use cases:
- * - Vultr (doesn't have prompt caching)
- * - OpenAI (doesn't have prompt caching)
+ * - Groq (doesn't have prompt caching)
  * - Any provider without built-in caching
  * - Supplement to Anthropic prompt caching for exact duplicates
  *

package/dist/llm/server.d.ts

@@ -10,5 +10,5 @@ export * from './providers/groq.js';
 export * from './providers/inference-snaps.js';
 export * from './providers/ollama.js';
 export * from './providers/openai-compat.js';
-export * from './providers/vultr.js';
+export * from './workspace-provider-config.js';
 //# sourceMappingURL=server.d.ts.map

package/dist/llm/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/llm/server.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,cAAc,aAAa,CAAC;AAG5B,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,uBAAuB,CAAC;AACtC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,sBAAsB,CAAC"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/llm/server.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,cAAc,aAAa,CAAC;AAG5B,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,uBAAuB,CAAC;AACtC,cAAc,8BAA8B,CAAC;AAI7C,cAAc,gCAAgC,CAAC"}

package/dist/llm/server.js

@@ -12,4 +12,6 @@ export * from './providers/groq.js';
 export * from './providers/inference-snaps.js';
 export * from './providers/ollama.js';
 export * from './providers/openai-compat.js';
-export * from './providers/vultr.js';
+// Export per-workspace provider registry (used by admin inference-config route
+// to hydrate per-site config at boot + apply changes immediately on PUT)
+export * from './workspace-provider-config.js';

package/dist/memory/services/index.d.ts

@@ -5,4 +5,6 @@
  */
 export type { EntityType } from './node-id-service.js';
 export { NodeIdService } from './node-id-service.js';
+export type { Contradiction, ReconciledMemory, ReconciliationResult, SharedFactInput, } from './reconciliation-service.js';
+export { buildReconciliationPrompt, parseReconciliationResponse, reconcileHeuristic, } from './reconciliation-service.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/memory/services/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/memory/services/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/memory/services/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,YAAY,EACV,aAAa,EACb,gBAAgB,EAChB,oBAAoB,EACpB,eAAe,GAChB,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,yBAAyB,EACzB,2BAA2B,EAC3B,kBAAkB,GACnB,MAAM,6BAA6B,CAAC"}

package/dist/memory/services/index.js

@@ -4,3 +4,4 @@
  * Business logic services for memory operations.
  */
 export { NodeIdService } from './node-id-service.js';
+export { buildReconciliationPrompt, parseReconciliationResponse, reconcileHeuristic, } from './reconciliation-service.js';

package/dist/memory/services/reconciliation-service.d.ts

@@ -0,0 +1,64 @@
+/**
+ * Reconciliation Service - Fact resolution for multi-agent memory.
+ *
+ * Reads shared facts from a coordination session, groups related facts,
+ * deduplicates, and resolves contradictions. Produces reconciled memories
+ * for Layer 3.
+ *
+ * Layer 3 of the multi-agent shared memory architecture.
+ *
+ * LICENSING BOUNDARY:
+ * - `reconcileHeuristic()` is the production path — no external API calls,
+ *   safe for all tiers including free customers.
+ * - `buildReconciliationPrompt()` and `parseReconciliationResponse()` are
+ *   reference implementations for internal/developer use only. They build
+ *   a prompt string but DO NOT call any LLM API. If wired to an LLM provider,
+ *   the caller is responsible for API pricing and usage compliance.
+ * - Customer-facing reconciliation MUST use the heuristic path or a
+ *   self-hosted model (e.g. Ollama). Do not proxy third-party LLM APIs
+ *   to end users without a commercial API agreement with the provider.
+ */
+export interface SharedFactInput {
+    id: string;
+    agentId: string;
+    content: string;
+    factType: string;
+    confidence: number;
+    tags: string[];
+    sourceRef?: Record<string, unknown> | null;
+}
+export interface ReconciledMemory {
+    content: string;
+    type: string;
+    sourceFactIds: string[];
+    confidence: number;
+}
+export interface Contradiction {
+    factIds: string[];
+    description: string;
+    resolution: string;
+}
+export interface ReconciliationResult {
+    canonicalFacts: ReconciledMemory[];
+    contradictions: Contradiction[];
+    duplicates: string[][];
+    summary: string;
+}
+/**
+ * Build the reconciliation prompt for an LLM.
+ * The prompt asks the model to analyze facts from multiple agents and
+ * produce a structured reconciliation result.
+ */
+export declare function buildReconciliationPrompt(facts: SharedFactInput[], scratchpadContent?: Record<string, unknown>): string;
+/**
+ * Parse the LLM response into a structured reconciliation result.
+ * Maps 1-based fact indices back to actual fact IDs.
+ */
+export declare function parseReconciliationResponse(response: string, facts: SharedFactInput[]): ReconciliationResult;
+/**
+ * Simple heuristic reconciliation (no LLM required).
+ * Deduplicates by normalized content and groups by type.
+ * Use this as a fallback when no LLM provider is configured.
+ */
+export declare function reconcileHeuristic(facts: SharedFactInput[]): ReconciliationResult;
+//# sourceMappingURL=reconciliation-service.d.ts.map

package/dist/memory/services/reconciliation-service.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"reconciliation-service.d.ts","sourceRoot":"","sources":["../../../src/memory/services/reconciliation-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAAC;CAC5C;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,aAAa;IAC5B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,oBAAoB;IACnC,cAAc,EAAE,gBAAgB,EAAE,CAAC;IACnC,cAAc,EAAE,aAAa,EAAE,CAAC;IAChC,UAAU,EAAE,MAAM,EAAE,EAAE,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED;;;;GAIG;AACH,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,eAAe,EAAE,EACxB,iBAAiB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC1C,MAAM,CAuCR;AAED;;;GAGG;AACH,wBAAgB,2BAA2B,CACzC,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,eAAe,EAAE,GACvB,oBAAoB,CAsCtB;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,eAAe,EAAE,GAAG,oBAAoB,CAsCjF"}