@revealui/ai 0.6.3 → 0.7.0

package/README.md

@@ -16,7 +16,7 @@ AI system for RevealUI - memory, LLM, orchestration, and tools.
 ## Features
 
 - **Memory System**: CRDT-based persistent memory (Working, Episodic, Semantic)
-- **LLM Integration**: Provider abstractions for Anthropic, GROQ, Ollama, Canonical Inference Snaps, and more
+- **LLM Integration**: Provider abstractions for Ollama and Canonical Inference Snaps (open-model defaults), plus Groq, Anthropic, OpenAI, and HuggingFace as bring-your-own-key cloud adapters. Every adapter calls the vendor's OpenAI-compatible HTTP endpoint with your own key. No proprietary Anthropic or OpenAI SDK.
 - **Agent Orchestration**: Runtime and execution engine for AI agents
 - **Tool Calling**: Tool registry + standard-MCP-client integration (Stage 5.1a)
 - **Vector Search**: Semantic search with pgvector
@@ -64,8 +64,9 @@ const provider = new InferenceSnapsProvider({
 const client = new LLMClient({ provider })
 ```
 
-Cloud providers (Anthropic, OpenAI, GROQ) remain supported; the local
-inference path is the documented default for self-hosted deployments.
+Groq, Anthropic, OpenAI, and HuggingFace remain supported as pluggable,
+bring-your-own-key cloud providers. The local inference path is the
+documented default for self-hosted deployments.
 
 ## MCP tool integration
 
@@ -313,7 +314,7 @@ const state = await persistence.loadCRDTState(crdtId, 'lww_register')
 
 ### LLM Integration
 
-Provider abstractions and unified client for Anthropic, GROQ, and Ollama.
+Provider abstractions and unified client for Ollama, Canonical Inference Snaps, Groq, Anthropic, OpenAI, and HuggingFace.
 
 ```typescript
 import { LLMClient, createLLMClientFromEnv } from '@revealui/ai/llm/client'

package/dist/a2a/card.d.ts

@@ -8,6 +8,15 @@
  */
 import type { AgentDefinition } from '@revealui/contracts';
 import { type A2AAgentCard } from '@revealui/contracts';
+/**
+ * Built-in platform agents defer model selection to the inference provider the
+ * account configures on the API Keys page. The dispatch runtime resolves the
+ * real model from that provider, so the definition itself names no vendor. An
+ * empty value means "use the configured provider's model". Claude and other
+ * model ids stay valid where an operator wires a compatible provider; they are
+ * no longer the hardcoded default.
+ */
+export declare const PROVIDER_RESOLVED_MODEL = "";
 declare class AgentCardRegistry {
     private readonly defs;
     constructor();

package/dist/a2a/card.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"card.d.ts","sourceRoot":"","sources":["../../src/a2a/card.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,KAAK,YAAY,EAAyB,MAAM,qBAAqB,CAAC;AAwI/E,cAAM,iBAAiB;IACrB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAsC;;IAO3D,QAAQ,CAAC,GAAG,EAAE,eAAe,GAAG,IAAI;IAIpC,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;IAIpC,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,GAAG,SAAS,CAAC,CAAC,GAAG,OAAO;IAOzF,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS;IAIpD,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,IAAI;IAM9D,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE;IAI1C,QAAQ,IAAI,eAAe,EAAE;IAI7B,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;CAG9B;AAED,qDAAqD;AACrD,eAAO,MAAM,iBAAiB,mBAA0B,CAAC;AAEzD,YAAY,EAAE,eAAe,EAAE,CAAC"}
+{"version":3,"file":"card.d.ts","sourceRoot":"","sources":["../../src/a2a/card.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,KAAK,YAAY,EAAyB,MAAM,qBAAqB,CAAC;AAM/E;;;;;;;GAOG;AACH,eAAO,MAAM,uBAAuB,KAAK,CAAC;AAoI1C,cAAM,iBAAiB;IACrB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAsC;;IAO3D,QAAQ,CAAC,GAAG,EAAE,eAAe,GAAG,IAAI;IAIpC,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;IAIpC,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,GAAG,SAAS,CAAC,CAAC,GAAG,OAAO;IAOzF,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS;IAIpD,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,IAAI;IAM9D,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE;IAI1C,QAAQ,IAAI,eAAe,EAAE;IAI7B,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;CAG9B;AAED,qDAAqD;AACrD,eAAO,MAAM,iBAAiB,mBAA0B,CAAC;AAEzD,YAAY,EAAE,eAAe,EAAE,CAAC"}

package/dist/a2a/card.js

@@ -10,6 +10,15 @@ import { agentDefinitionToCard } from '@revealui/contracts';
 // =============================================================================
 // Built-in agent definitions
 // =============================================================================
+/**
+ * Built-in platform agents defer model selection to the inference provider the
+ * account configures on the API Keys page. The dispatch runtime resolves the
+ * real model from that provider, so the definition itself names no vendor. An
+ * empty value means "use the configured provider's model". Claude and other
+ * model ids stay valid where an operator wires a compatible provider; they are
+ * no longer the hardcoded default.
+ */
+export const PROVIDER_RESOLVED_MODEL = '';
 /** RevealUI platform meta-agent  -  "The Creator" */
 const THE_CREATOR_DEF = {
     id: 'revealui-creator',
@@ -18,7 +27,7 @@ const THE_CREATOR_DEF = {
     description: 'The RevealUI platform agent. Scaffolds new AI agents, manages agent lifecycles, ' +
         'orchestrates multi-agent workflows, and acts as the primary interface for AI capabilities ' +
         'on the RevealUI platform.',
-    model: 'claude-opus-4-6',
+    model: PROVIDER_RESOLVED_MODEL,
     systemPrompt: 'You are The Creator, the meta-agent for RevealUI. You design, configure, and deploy ' +
         'purpose-built AI agents for RevealUI users. You have access to agent scaffolding tools, ' +
         'the RevealUI admin, and the billing system.',
@@ -74,7 +83,7 @@ const TICKET_AGENT_DEF = {
     name: 'Ticket Agent',
     description: 'Handles support tickets, resolves user issues, and escalates when needed. ' +
         'Uses the RevealUI admin to create and update tickets.',
-    model: 'claude-sonnet-4-6',
+    model: PROVIDER_RESOLVED_MODEL,
     systemPrompt: 'You are the RevealUI Ticket Agent. You help users resolve issues by creating tickets, ' +
         'searching for solutions, and escalating complex problems to the support team.',
     tools: [

package/dist/embeddings/index.d.ts

@@ -7,6 +7,7 @@
  * Note: Groq and Anthropic do not support embeddings  -  use Ollama for embedding tasks.
  */
 import z from 'zod/v4';
+import { type LLMClient } from '../llm/client.js';
 declare const EmbeddingSchema: z.ZodObject<{
     model: z.ZodString;
     vector: z.ZodArray<z.ZodNumber>;
@@ -23,6 +24,13 @@ export interface GenerateEmbeddingOptions {
      */
     model?: string;
     cache?: boolean;
+    /**
+     * Pre-resolved LLM client (GAP-360). When supplied — by an authenticated
+     * caller that ran `resolveLLMClientForRequest` — embeddings use it (per-user
+     * BYOK / hosted resolution). When omitted, embeddings fall back to the
+     * env-configured client, preserving self-hosted and internal-caller behavior.
+     */
+    client?: Pick<LLMClient, 'embed'>;
 }
 /**
  * Generate an embedding for the given text using the configured LLM provider.

package/dist/embeddings/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/embeddings/index.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,CAAC,MAAM,QAAQ,CAAC;AAGvB,QAAA,MAAM,eAAe;;;;;iBAgBjB,CAAC;AACL,KAAK,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AAEjD,MAAM,WAAW,wBAAwB;IACvC;;;;;OAKG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAsB,iBAAiB,CACrC,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,wBAA6B,GACrC,OAAO,CAAC,SAAS,CAAC,CA4BpB;AAED;;;;;;GAMG;AACH,wBAAsB,kBAAkB,CACtC,KAAK,EAAE,MAAM,EAAE,EACf,OAAO,GAAE,wBAA6B,GACrC,OAAO,CAAC,SAAS,EAAE,CAAC,CAMtB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/embeddings/index.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,CAAC,MAAM,QAAQ,CAAC;AACvB,OAAO,EAA0B,KAAK,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE1E,QAAA,MAAM,eAAe;;;;;iBAgBjB,CAAC;AACL,KAAK,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AAEjD,MAAM,WAAW,wBAAwB;IACvC;;;;;OAKG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;CACnC;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAsB,iBAAiB,CACrC,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,wBAA6B,GACrC,OAAO,CAAC,SAAS,CAAC,CA6BpB;AAED;;;;;;GAMG;AACH,wBAAsB,kBAAkB,CACtC,KAAK,EAAE,MAAM,EAAE,EACf,OAAO,GAAE,wBAA6B,GACrC,OAAO,CAAC,SAAS,EAAE,CAAC,CAMtB"}

package/dist/embeddings/index.js

@@ -41,8 +41,9 @@ export async function generateEmbedding(text, options = {}) {
     if (!text || typeof text !== 'string' || text.trim().length === 0) {
         throw new Error('Text must be a non-empty string');
     }
-    // Use unified LLM client  -  auto-detects provider from env vars
-    const client = createLLMClientFromEnv();
+    // Prefer the caller-resolved client (BYOK / hosted). Fall back to the
+    // env-configured client for self-hosted and internal (no-user) callers.
+    const client = options.client ?? createLLMClientFromEnv();
     // Ask client to embed  -  each provider uses its own default model when model is undefined
     const result = await client.embed(text, model ? { model } : undefined);
     // If provider returned batch, pick first

package/dist/index.d.ts

@@ -54,6 +54,7 @@ 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/resolve.js';
 export * from './llm/token-counter.js';
 export * from './llm/workspace-provider-config.js';
 export * from './memory/index.js';

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"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"}
+{"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,kBAAkB,CAAC;AACjC,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

@@ -61,6 +61,7 @@ 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/resolve.js';
 export * from './llm/token-counter.js';
 export * from './llm/workspace-provider-config.js';
 // Re-export memory system

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

@@ -53,7 +53,7 @@ export declare function estimateCacheSavings(inputTokens: number, cacheHitRate:
  *
  * @example
  * ```ts
- * const response = await client.chat(messages, { enableCache: true })
+ * const response = await client.chat(messages, { cacheHint: true })
  * const stats = formatCacheStats(response.usage)
  * console.log(stats)
  * // "Cache: 45% read (2,500 tokens), 10% created (500 tokens)"
@@ -83,7 +83,7 @@ export declare function shouldCache(content: string, minTokens?: number): boolea
  *   ],
  * })
  *
- * const response = await client.chat(conversation, { enableCache: true })
+ * const response = await client.chat(conversation, { cacheHint: true })
  * ```
  */
 export declare function createCachedConversation(config: {
@@ -92,31 +92,17 @@ export declare function createCachedConversation(config: {
     messages: Message[];
 }): Message[];
 /**
- * Pricing information for Anthropic Claude models (as of 2024)
- * Prices are per million tokens
+ * @deprecated Use `MODEL_PRICING` from `./token-counter.js`. Retained as a thin derived
+ * view (the Anthropic-with-cache subset) so existing importers keep one source of truth.
  */
 export declare const ANTHROPIC_PRICING: {
-    readonly 'claude-3-5-sonnet-20241022': {
-        readonly input: 3;
-        readonly output: 15;
-        readonly cacheWrite: 3.75;
-        readonly cacheRead: 0.3;
-    };
-    readonly 'claude-3-5-haiku-20241022': {
-        readonly input: 1;
-        readonly output: 5;
-        readonly cacheWrite: 1.25;
-        readonly cacheRead: 0.1;
-    };
-    readonly 'claude-3-opus-20240229': {
-        readonly input: 15;
-        readonly output: 75;
-        readonly cacheWrite: 18.75;
-        readonly cacheRead: 1.5;
-    };
+    readonly 'claude-3-5-sonnet-20241022': import("./token-counter.js").ModelPricing;
+    readonly 'claude-3-5-haiku-20241022': import("./token-counter.js").ModelPricing;
+    readonly 'claude-3-opus-20240229': import("./token-counter.js").ModelPricing;
 };
 /**
- * Calculate actual cost of a request with caching
+ * Calculate actual cost of a request with caching, using the unified MODEL_PRICING table.
+ * Unknown models price at zero (matching `estimateCost`).
  *
  * @example
  * ```ts
@@ -128,12 +114,12 @@ export declare const ANTHROPIC_PRICING: {
  *   cacheReadTokens: 5000,
  * })
  *
- * console.log(`Request cost: $${cost.toFixed(4)}`)
+ * console.log(`Request cost: $${cost.total.toFixed(4)}`)
  * console.log(`Savings vs no cache: $${cost.savings.toFixed(4)}`)
  * ```
  */
 export declare function calculateCacheCost(usage: {
-    model: keyof typeof ANTHROPIC_PRICING;
+    model: string;
     promptTokens: number;
     completionTokens: number;
     cacheCreationTokens?: number;

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

@@ -1 +1 @@
-{"version":3,"file":"cache-utils.d.ts","sourceRoot":"","sources":["../../src/llm/cache-utils.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAEnD;;;GAGG;AACH,wBAAgB,SAAS,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAKnD;AAED;;;;;;;;;GASG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAK9D;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,qBAAqB,EAAE,MAAM,GAC5B,MAAM,CAeR;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG,MAAM,GAAG,IAAI,CAqBhB;AAED;;;GAGG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,SAAO,GAAG,OAAO,CAItE;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,wBAAwB,CAAC,MAAM,EAAE;IAC/C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,EAAE,OAAO,EAAE,CAAC;CACrB,GAAG,OAAO,EAAE,CAwBZ;AAED;;;GAGG;AACH,eAAO,MAAM,iBAAiB;;;;;;;;;;;;;;;;;;;CAmBpB,CAAC;AAEX;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE;IACxC,KAAK,EAAE,MAAM,OAAO,iBAAiB,CAAC;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAwBxE"}
+{"version":3,"file":"cache-utils.d.ts","sourceRoot":"","sources":["../../src/llm/cache-utils.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAGnD;;;GAGG;AACH,wBAAgB,SAAS,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAKnD;AAED;;;;;;;;;GASG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAK9D;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,qBAAqB,EAAE,MAAM,GAC5B,MAAM,CAeR;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG,MAAM,GAAG,IAAI,CAqBhB;AAED;;;GAGG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,SAAO,GAAG,OAAO,CAItE;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,wBAAwB,CAAC,MAAM,EAAE;IAC/C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,EAAE,OAAO,EAAE,CAAC;CACrB,GAAG,OAAO,EAAE,CAwBZ;AAWD;;;GAGG;AACH,eAAO,MAAM,iBAAiB;;;;CAIpB,CAAC;AAEX;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CA6BxE"}

package/dist/llm/cache-utils.js

@@ -1,4 +1,5 @@
 /* console-allowed */
+import { MODEL_PRICING } from './token-counter.js';
 /**
  * Mark a message for caching
  * Use this for system prompts, tool definitions, or large context that repeats
@@ -6,7 +7,7 @@
 export function withCache(message) {
     return {
         ...message,
-        cacheControl: { type: 'ephemeral' },
+        cache: true,
     };
 }
 /**
@@ -57,7 +58,7 @@ export function estimateCacheSavings(inputTokens, cacheHitRate, cachedTokenPerce
  *
  * @example
  * ```ts
- * const response = await client.chat(messages, { enableCache: true })
+ * const response = await client.chat(messages, { cacheHint: true })
  * const stats = formatCacheStats(response.usage)
  * console.log(stats)
  * // "Cache: 45% read (2,500 tokens), 10% created (500 tokens)"
@@ -102,7 +103,7 @@ export function shouldCache(content, minTokens = 1024) {
  *   ],
  * })
  *
- * const response = await client.chat(conversation, { enableCache: true })
+ * const response = await client.chat(conversation, { cacheHint: true })
  * ```
  */
 export function createCachedConversation(config) {
@@ -118,7 +119,7 @@ export function createCachedConversation(config) {
             result.push({
                 role: 'system',
                 content: doc,
-                ...(isLast ? { cacheControl: { type: 'ephemeral' } } : {}),
+                ...(isLast ? { cache: true } : {}),
             });
         });
     }
@@ -126,32 +127,26 @@ export function createCachedConversation(config) {
     result.push(...config.messages);
     return result;
 }
+/** Look up a model's pricing, asserting presence (used only for known-present keys). */
+function requirePricing(model) {
+    const pricing = MODEL_PRICING[model];
+    if (!pricing) {
+        throw new Error(`No pricing entry for model: ${model}`);
+    }
+    return pricing;
+}
 /**
- * Pricing information for Anthropic Claude models (as of 2024)
- * Prices are per million tokens
+ * @deprecated Use `MODEL_PRICING` from `./token-counter.js`. Retained as a thin derived
+ * view (the Anthropic-with-cache subset) so existing importers keep one source of truth.
  */
 export const ANTHROPIC_PRICING = {
-    'claude-3-5-sonnet-20241022': {
-        input: 3.0,
-        output: 15.0,
-        cacheWrite: 3.75, // 25% markup for cache creation
-        cacheRead: 0.3, // 90% discount for cache hits
-    },
-    'claude-3-5-haiku-20241022': {
-        input: 1.0,
-        output: 5.0,
-        cacheWrite: 1.25,
-        cacheRead: 0.1,
-    },
-    'claude-3-opus-20240229': {
-        input: 15.0,
-        output: 75.0,
-        cacheWrite: 18.75,
-        cacheRead: 1.5,
-    },
+    'claude-3-5-sonnet-20241022': requirePricing('claude-3-5-sonnet-20241022'),
+    'claude-3-5-haiku-20241022': requirePricing('claude-3-5-haiku-20241022'),
+    'claude-3-opus-20240229': requirePricing('claude-3-opus-20240229'),
 };
 /**
- * Calculate actual cost of a request with caching
+ * Calculate actual cost of a request with caching, using the unified MODEL_PRICING table.
+ * Unknown models price at zero (matching `estimateCost`).
  *
  * @example
  * ```ts
@@ -163,12 +158,17 @@ export const ANTHROPIC_PRICING = {
  *   cacheReadTokens: 5000,
  * })
  *
- * console.log(`Request cost: $${cost.toFixed(4)}`)
+ * console.log(`Request cost: $${cost.total.toFixed(4)}`)
  * console.log(`Savings vs no cache: $${cost.savings.toFixed(4)}`)
  * ```
  */
 export function calculateCacheCost(usage) {
-    const pricing = ANTHROPIC_PRICING[usage.model];
+    const pricing = MODEL_PRICING[usage.model] ?? {
+        input: 0,
+        output: 0,
+        cacheWrite: 0,
+        cacheRead: 0,
+    };
     const uncachedTokens = usage.promptTokens - (usage.cacheCreationTokens || 0) - (usage.cacheReadTokens || 0);
     const costs = {
         input: (uncachedTokens / 1_000_000) * pricing.input,

package/dist/llm/client.d.ts

@@ -10,7 +10,15 @@ 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 = 'groq' | 'ollama' | 'huggingface' | 'inference-snaps';
+export type LLMProviderType = 'anthropic' | 'openai' | 'groq' | 'huggingface' | 'ollama' | 'inference-snaps';
+/**
+ * Providers reachable from a hosted (serverless) deployment. Localhost-only
+ * providers are false. Consumed by PR-2/PR-3 (resolver + settings UI) to pick
+ * which providers a hosted account may configure; no runtime consumer yet.
+ */
+export declare const hostedViable: Record<LLMProviderType, boolean>;
+/** True when the provider can serve a hosted (serverless) deployment. */
+export declare function isHostedViable(provider: LLMProviderType): boolean;
 export interface LLMClientConfig {
     provider: LLMProviderType;
     apiKey: string;
@@ -129,20 +137,24 @@ 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. 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.
+ * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA → ANTHROPIC_API_KEY →
+ * OPENAI_API_KEY. 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 localhost default is discoverable in logs.
  *
- * All providers use OpenAI-compatible APIs. No proprietary provider SDKs.
+ * All providers use OpenAI-compatible APIs. No proprietary provider SDKs
+ * (Anthropic + OpenAI ride their OpenAI-compatible endpoints).
  *
  * 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:
- *   inference-snaps → gemma3       (base URL defaults to http://localhost:9090/v1)
+ *   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)
+ *   ollama          → gemma4:e2b        (base URL defaults to http://localhost:11434)
+ *   anthropic       → claude-sonnet-4-6 (base URL defaults to https://api.anthropic.com/v1)
+ *   openai          → gpt-4o            (base URL defaults to https://api.openai.com/v1)
  */
 export declare function createLLMClientFromEnv(): LLMClient;
 /**
@@ -155,8 +167,16 @@ export declare function createLLMClientFromEnv(): LLMClient;
  * Returns `null` if the user has no stored keys (callers should fall back
  * to `createLLMClientFromEnv()` or return a 402/feature-unavailable error).
  *
+ * When `opts.hostedViableOnly` is set, a stored key whose provider is not
+ * hosted-viable (ollama / inference-snaps are localhost-only) resolves to
+ * `null` instead of a client. The resolver passes this on hosted deployments so
+ * a localhost-only BYOK key can never yield a localhost client — the exact
+ * silent-localhost defect GAP-360 closes (spec §6.5, fail-closed).
+ *
  * @param userId - The user's ID from the `users` table
  * @param db - A Drizzle NeonDB client instance
  */
-export declare function createLLMClientForUser(userId: string, db: Database, auditStore?: AuditStore): Promise<LLMClient | null>;
+export declare function createLLMClientForUser(userId: string, db: Database, auditStore?: Pick<AuditStore, 'append'>, opts?: {
+    hostedViableOnly?: boolean;
+}): Promise<LLMClient | null>;
 //# sourceMappingURL=client.d.ts.map

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;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;IAsD5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IA4EvF;;;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"}
+{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/llm/client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAMH,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;AAElE,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,GACvB,WAAW,GACX,QAAQ,GACR,MAAM,GACN,aAAa,GACb,QAAQ,GACR,iBAAiB,CAAC;AAEtB;;;;GAIG;AACH,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,eAAe,EAAE,OAAO,CAOzD,CAAC;AAEF,yEAAyE;AACzE,wBAAgB,cAAc,CAAC,QAAQ,EAAE,eAAe,GAAG,OAAO,CAEjE;AAMD,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;IAgCtB;;;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;IAsD5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IA4EvF;;;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;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,sBAAsB,IAAI,SAAS,CAyFlD;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAsB,sBAAsB,CAC1C,MAAM,EAAE,MAAM,EACd,EAAE,EAAE,QAAQ,EAGZ,UAAU,CAAC,EAAE,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,EACvC,IAAI,CAAC,EAAE;IAAE,gBAAgB,CAAC,EAAE,OAAO,CAAA;CAAE,GACpC,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC,CAsD3B"}

package/dist/llm/client.js

@@ -3,16 +3,42 @@
  *
  * Single interface for all LLM providers with fallback and rate limiting
  */
+// Log redaction lives in @revealui/security — import `redactLogContext`
+// (recursive walker) or `redactLogField` (single key/value).
+import { createLogger } from '@revealui/core/observability/logger';
 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 { AnthropicProvider } from './providers/anthropic.js';
 import { GroqProvider } from './providers/groq.js';
 import { InferenceSnapsProvider, } from './providers/inference-snaps.js';
 import { OllamaProvider } from './providers/ollama.js';
+import { OpenAIProvider } from './providers/openai.js';
+import { OpenAICompatProvider } from './providers/openai-compat.js';
 import { ResponseCache } from './response-cache.js';
 import { SemanticCache, } from './semantic-cache.js';
 import { estimateRequest as _estimateRequestTokens } from './token-counter.js';
+/**
+ * Providers reachable from a hosted (serverless) deployment. Localhost-only
+ * providers are false. Consumed by PR-2/PR-3 (resolver + settings UI) to pick
+ * which providers a hosted account may configure; no runtime consumer yet.
+ */
+export const hostedViable = {
+    anthropic: true,
+    openai: true,
+    groq: true,
+    huggingface: true,
+    ollama: false,
+    'inference-snaps': false,
+};
+/** True when the provider can serve a hosted (serverless) deployment. */
+export function isHostedViable(provider) {
+    return hostedViable[provider];
+}
+/** Emitted once per process when the zero-config localhost default is selected. */
+let warnedLocalhostDefault = false;
+const envFactoryLogger = createLogger({ component: 'createLLMClientFromEnv' });
 export class LLMClient {
     provider;
     fallbackProvider;
@@ -78,12 +104,22 @@ export class LLMClient {
     }
     createProvider(type, config) {
         switch (type) {
+            case 'anthropic':
+                return new AnthropicProvider(config);
+            case 'openai':
+                return new OpenAIProvider(config);
             case 'groq':
                 return new GroqProvider(config);
             case 'ollama':
                 return new OllamaProvider(config);
             case 'inference-snaps':
                 return new InferenceSnapsProvider(config);
+            case 'huggingface':
+                // HuggingFace exposes an OpenAI-compatible inference endpoint; baseURL is
+                // per-model (HF_MODEL_URL), so it has no dedicated wrapper — the compat
+                // base serves it directly. Fixes the latent defect where the env factory
+                // accepted 'huggingface' but createProvider threw 'Unknown provider type'.
+                return new OpenAICompatProvider(config);
             default:
                 throw new Error(`Unknown provider type: ${String(type)}`);
         }
@@ -412,23 +448,29 @@ 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. 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.
+ * in priority order: INFERENCE_SNAPS → GROQ → OLLAMA → ANTHROPIC_API_KEY →
+ * OPENAI_API_KEY. 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 localhost default is discoverable in logs.
  *
- * All providers use OpenAI-compatible APIs. No proprietary provider SDKs.
+ * All providers use OpenAI-compatible APIs. No proprietary provider SDKs
+ * (Anthropic + OpenAI ride their OpenAI-compatible endpoints).
  *
  * 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:
- *   inference-snaps → gemma3       (base URL defaults to http://localhost:9090/v1)
+ *   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)
+ *   ollama          → gemma4:e2b        (base URL defaults to http://localhost:11434)
+ *   anthropic       → claude-sonnet-4-6 (base URL defaults to https://api.anthropic.com/v1)
+ *   openai          → gpt-4o            (base URL defaults to https://api.openai.com/v1)
  */
 export function createLLMClientFromEnv() {
-    // Auto-detect provider when LLM_PROVIDER is not explicitly set
+    // Auto-detect provider when LLM_PROVIDER is not explicitly set. The existing
+    // priority order (INFERENCE_SNAPS → GROQ → OLLAMA) is preserved so existing
+    // deployments resolve identically; the frontier providers are appended after.
     let provider;
     if (process.env.LLM_PROVIDER) {
         provider = process.env.LLM_PROVIDER;
@@ -442,15 +484,38 @@ export function createLLMClientFromEnv() {
     else if (process.env.OLLAMA_BASE_URL) {
         provider = 'ollama';
     }
+    else if (process.env.ANTHROPIC_API_KEY) {
+        provider = 'anthropic';
+    }
+    else if (process.env.OPENAI_API_KEY) {
+        provider = 'openai';
+    }
     else {
-        // Zero-config Ubuntu default: assume Inference Snaps on the standard
-        // local port. Operator sees the warning once at boot.
+        // Zero-config Ubuntu default: assume Inference Snaps on the standard local
+        // port. This localhost default is unreachable inside a hosted serverless
+        // function, so warn once per process to make the implicit choice visible.
         provider = 'inference-snaps';
+        if (!warnedLocalhostDefault) {
+            warnedLocalhostDefault = true;
+            envFactoryLogger.warn('No LLM provider env var set — defaulting to inference-snaps at ' +
+                'http://localhost:9090/v1. This localhost endpoint is unreachable on a ' +
+                'hosted deployment; set LLM_PROVIDER or a provider key.');
+        }
     }
     let apiKey;
     let baseURL;
     let defaultModel;
-    if (provider === 'huggingface') {
+    if (provider === 'anthropic') {
+        apiKey = process.env.ANTHROPIC_API_KEY;
+        baseURL = process.env.ANTHROPIC_BASE_URL ?? 'https://api.anthropic.com/v1';
+        defaultModel = 'claude-sonnet-4-6';
+    }
+    else if (provider === 'openai') {
+        apiKey = process.env.OPENAI_API_KEY;
+        baseURL = process.env.OPENAI_BASE_URL ?? 'https://api.openai.com/v1';
+        defaultModel = 'gpt-4o';
+    }
+    else if (provider === 'huggingface') {
         apiKey = process.env.HF_TOKEN;
         baseURL = process.env.HF_MODEL_URL;
     }
@@ -475,7 +540,8 @@ export function createLLMClientFromEnv() {
     }
     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, or HF_TOKEN).`);
+            `(INFERENCE_SNAPS_BASE_URL, GROQ_API_KEY, OLLAMA_BASE_URL, HF_TOKEN, ` +
+            `ANTHROPIC_API_KEY, or OPENAI_API_KEY).`);
     }
     return new LLMClient({
         provider,
@@ -501,10 +567,19 @@ export function createLLMClientFromEnv() {
  * Returns `null` if the user has no stored keys (callers should fall back
  * to `createLLMClientFromEnv()` or return a 402/feature-unavailable error).
  *
+ * When `opts.hostedViableOnly` is set, a stored key whose provider is not
+ * hosted-viable (ollama / inference-snaps are localhost-only) resolves to
+ * `null` instead of a client. The resolver passes this on hosted deployments so
+ * a localhost-only BYOK key can never yield a localhost client — the exact
+ * silent-localhost defect GAP-360 closes (spec §6.5, fail-closed).
+ *
  * @param userId - The user's ID from the `users` table
  * @param db - A Drizzle NeonDB client instance
  */
-export async function createLLMClientForUser(userId, db, auditStore) {
+export async function createLLMClientForUser(userId, db, 
+// Only `append` is used; the narrow type lets a persistent store whose entry
+// type widens eventType/severity to `string` (e.g. DrizzleAuditStore) fit.
+auditStore, opts) {
     // Find the user's preferred provider config
     const [preferredConfig] = await db
         .select()
@@ -522,8 +597,12 @@ export async function createLLMClientForUser(userId, db, auditStore) {
     const [keyRow] = await keyQuery;
     if (!keyRow)
         return null;
-    const plaintext = decryptApiKey(keyRow.encryptedKey);
     const provider = keyRow.provider;
+    // Fail-closed hosted filter (§6.5): reject a localhost-only provider before
+    // decrypting, so no plaintext is touched for a key we will not use.
+    if (opts?.hostedViableOnly && !isHostedViable(provider))
+        return null;
+    const plaintext = decryptApiKey(keyRow.encryptedKey);
     const model = preferredConfig?.model ?? undefined;
     // Fire-and-forget: record when this key was last used (best-effort, never blocks)
     db.update(userApiKeys)