@revealui/ai 0.6.3 → 0.7.0

package/dist/llm/providers/anthropic.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Anthropic Provider
+ *
+ * Thin wrapper over OpenAICompatProvider using Anthropic's OpenAI-compatible
+ * surface at https://api.anthropic.com/v1. No proprietary Anthropic SDK — the
+ * fleet posture is "No proprietary provider SDKs" (see client.ts). Anthropic
+ * exposes an OpenAI-compatible chat/completions endpoint, so the same base
+ * implementation that serves Groq/Ollama/inference-snaps serves Anthropic too.
+ *
+ * Docs: https://docs.anthropic.com/en/api/openai-sdk
+ */
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
+export interface AnthropicProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
+    apiKey: string;
+    /** Defaults to https://api.anthropic.com/v1 */
+    baseURL?: string;
+    /** Defaults to claude-sonnet-4-6 (conservative documented default; overridable) */
+    model?: string;
+    timeout?: number;
+    maxRetries?: number;
+}
+export declare class AnthropicProvider implements LLMProvider {
+    private inner;
+    constructor(config: AnthropicProviderConfig);
+    capabilities(): ReasonerCapabilities;
+    chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
+    stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
+    embed(_text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
+}
+//# sourceMappingURL=anthropic.d.ts.map

package/dist/llm/providers/anthropic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"anthropic.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/anthropic.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,uBAAwB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,+CAA+C;IAC/C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,mFAAmF;IACnF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,iBAAkB,YAAW,WAAW;IACnD,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,uBAAuB;IAQ3C,YAAY,IAAI,oBAAoB;IAepC,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;IAIhF,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,QAAQ,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG9F"}

package/dist/llm/providers/anthropic.js

@@ -0,0 +1,45 @@
+/**
+ * Anthropic Provider
+ *
+ * Thin wrapper over OpenAICompatProvider using Anthropic's OpenAI-compatible
+ * surface at https://api.anthropic.com/v1. No proprietary Anthropic SDK — the
+ * fleet posture is "No proprietary provider SDKs" (see client.ts). Anthropic
+ * exposes an OpenAI-compatible chat/completions endpoint, so the same base
+ * implementation that serves Groq/Ollama/inference-snaps serves Anthropic too.
+ *
+ * Docs: https://docs.anthropic.com/en/api/openai-sdk
+ */
+import { OpenAICompatProvider } from './openai-compat.js';
+export class AnthropicProvider {
+    inner;
+    constructor(config) {
+        this.inner = new OpenAICompatProvider({
+            ...config,
+            baseURL: config.baseURL ?? 'https://api.anthropic.com/v1',
+            model: config.model ?? 'claude-sonnet-4-6',
+        });
+    }
+    capabilities() {
+        return {
+            providerTag: 'anthropic',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            // Anthropic exposes no embeddings endpoint on its OpenAI-compat surface.
+            embeddings: false,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
+    chat(messages, options) {
+        return this.inner.chat(messages, options);
+    }
+    stream(messages, options) {
+        return this.inner.stream(messages, options);
+    }
+    embed(_text, _options) {
+        throw new Error('Anthropic does not expose an embeddings endpoint. Use OpenAI or Ollama.');
+    }
+}

package/dist/llm/providers/base.d.ts

@@ -1,7 +1,11 @@
 /**
  * LLM Provider Base Interface
  *
- * Abstract interface for all LLM providers (OpenAI, Anthropic, etc.)
+ * The agnostic Reasoner port (ADR 2026-06-25). Providers are negotiated by
+ * `capabilities()`, never by provider-name string. The control vocabulary is neutral
+ * and capability-gated (`effort`, `cache`/`cacheHint`); genuinely unportable knobs ride
+ * the namespaced opaque `providerOptions` channel; cross-provider usage (e.g. cache token
+ * stats) stays neutral in `LLMResponse.usage`.
  */
 /**
  * A plain text content part  -  used in multipart messages.
@@ -36,10 +40,11 @@ export interface Message {
     name?: string;
     toolCalls?: ToolCall[];
     toolCallId?: string;
-    /** Anthropic prompt caching - marks content for caching (5min TTL, 90% cost reduction) */
-    cacheControl?: {
-        type: 'ephemeral';
-    };
+    /**
+     * Neutral cache breakpoint hint (gated by the `promptCache` capability). An adapter
+     * that supports prompt caching maps this to its native cache-control; others ignore it.
+     */
+    cache?: boolean;
 }
 export interface ToolCall {
     id: string;
@@ -58,7 +63,11 @@ export interface LLMResponse {
         promptTokens: number;
         completionTokens: number;
         totalTokens: number;
-        /** Anthropic cache stats */
+        /**
+         * Cross-provider cache usage (neutral; gated by `promptCache`). Populated by any
+         * caching provider  -  Anthropic cache tokens, OpenAI `cached_tokens`, Gemini
+         * `cachedContentTokenCount`. Read by the cost path; stays in neutral `usage`.
+         */
         cacheCreationTokens?: number;
         cacheReadTokens?: number;
     };
@@ -81,22 +90,66 @@ export interface LLMProviderConfig {
     maxTokens?: number;
 }
 /**
- * Base interface for LLM providers
+ * Reasoning-depth hint (neutral, capability-gated by `reasoningEffort`).
+ *
+ * This is a *reasoning-depth* axis, NOT output entropy or length  -  adapters must never
+ * remap it to `temperature`/`maxTokens`. An adapter with a reasoning axis maps it to its
+ * native control (e.g. a thinking-token budget); an adapter without one advertises
+ * `reasoningEffort: false` and treats `effort` as a no-op.
  */
-export interface LLMProvider {
-    /**
-     * Chat completion
-     */
+export type ReasoningEffort = 'off' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh';
+/**
+ * Runtime-queryable capability record. The loop branches on these, never on a
+ * provider-name string. Conformance-tested (transport tier) so a provider cannot lie:
+ * advertise-true-but-not-wired fails the positive assertion; advertise-false-but-emits-a-
+ * native-field fails the inertness assertion.
+ */
+export interface ReasonerCapabilities {
+    /** Stable adapter id; also the only `providerOptions` namespace this adapter may read ('x-<tag>'). */
+    readonly providerTag: string;
+    readonly tools: boolean;
+    /** Implies `tools` (consistency invariant). */
+    readonly parallelToolCalls: boolean;
+    /** Implies the adapter accepts `ContentPart[]` message content. */
+    readonly vision: boolean;
+    readonly streaming: boolean;
+    readonly embeddings: boolean;
+    readonly reasoningEffort: boolean;
+    readonly promptCache: boolean;
+    readonly structuredOutput: boolean;
+    /** Max context window in tokens; undefined = unknown (loop budgets conservatively). */
+    readonly contextWindow?: number;
+}
+/**
+ * Namespaced opaque extension for genuinely unportable, request-shaped knobs.
+ *
+ * Core/loop code MUST pass this through and MUST NOT read any key  -  per-namespace schema
+ * lives in the owning adapter. A known-namespace registry (typed keys) rather than a bare
+ * `x-*` string convention, so a typo'd namespace fails loudly instead of silently no-opping.
+ */
+export interface ProviderOptions {
+    'x-anthropic'?: unknown;
+    'x-openai'?: unknown;
+    'x-google'?: unknown;
+}
+/**
+ * The Reasoner port  -  one swappable reasoning step plus the side capabilities
+ * (embeddings/streaming). The loop negotiates by `capabilities()`, never by name.
+ */
+export interface Reasoner {
+    /** Runtime-queryable, conformance-tested capability record. */
+    capabilities(): ReasonerCapabilities;
+    /** Chat completion. */
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
-    /**
-     * Generate embeddings
-     */
+    /** Generate embeddings. */
     embed(text: string | string[], options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
-    /**
-     * Stream chat completion
-     */
+    /** Stream chat completion. */
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
 }
+/**
+ * @deprecated Use `Reasoner`. Alias retained one release cycle for external callers.
+ */
+export type LLMProvider = Reasoner;
 export interface LLMChatOptions {
     temperature?: number;
     maxTokens?: number;
@@ -107,14 +160,12 @@ export interface LLMChatOptions {
             name: string;
         };
     };
-    /** Enable prompt caching (Anthropic only) - caches system prompts and tools */
-    enableCache?: boolean;
-    /**
-     * Extended thinking token budget (Anthropic only).
-     * Enables Claude's internal reasoning before responding.
-     * Typical values: 512 (minimal) → 31999 (xhigh). 0 or undefined = disabled.
-     */
-    thinkingBudget?: number;
+    /** Neutral reasoning-depth hint (gated by `reasoningEffort`). No-op where unsupported. */
+    effort?: ReasoningEffort;
+    /** Neutral prompt-cache hint for the stable prefix  -  system + tools (gated by `promptCache`). */
+    cacheHint?: boolean;
+    /** Opaque, namespaced provider extension. Core never reads it; the adapter owns the schema. */
+    providerOptions?: ProviderOptions;
 }
 export interface LLMEmbedOptions {
     model?: string;
@@ -123,8 +174,10 @@ export interface LLMStreamOptions {
     temperature?: number;
     maxTokens?: number;
     tools?: ToolDefinition[];
-    /** Enable prompt caching (Anthropic only) - caches system prompts and tools */
-    enableCache?: boolean;
+    /** Neutral reasoning-depth hint (gated by `reasoningEffort`). No-op where unsupported. */
+    effort?: ReasoningEffort;
+    /** Opaque, namespaced provider extension. Core never reads it; the adapter owns the schema. */
+    providerOptions?: ProviderOptions;
 }
 export interface ToolDefinition {
     type: 'function';

package/dist/llm/providers/base.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/base.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE;QACT,sEAAsE;QACtE,GAAG,EAAE,MAAM,CAAC;QACZ,yDAAyD;QACzD,MAAM,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;KAClC,CAAC;CACH;AAED,8DAA8D;AAC9D,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,GAAG,MAAM,CAAC;IAC/C;yFACqF;IACrF,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0FAA0F;IAC1F,YAAY,CAAC,EAAE;QAAE,IAAI,EAAE,WAAW,CAAA;KAAE,CAAC;CACtC;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;QACpB,4BAA4B;QAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;CACH;AAED,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B;;OAEG;IACH,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC,CAAC;IAE1E;;OAEG;IACH,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC,CAAC;IAE5F;;OAEG;IACH,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED,MAAM,WAAW,cAAc;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAChF,+EAA+E;IAC/E,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;OAIG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,+EAA+E;IAC/E,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACrC,CAAC;CACH;AAED,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC"}
+{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/base.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE;QACT,sEAAsE;QACtE,GAAG,EAAE,MAAM,CAAC;QACZ,yDAAyD;QACzD,MAAM,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;KAClC,CAAC;CACH;AAED,8DAA8D;AAC9D,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,GAAG,MAAM,CAAC;IAC/C;yFACqF;IACrF,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;OAGG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;QACpB;;;;WAIG;QACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;CACH;AAED,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAEtF;;;;;GAKG;AACH,MAAM,WAAW,oBAAoB;IACnC,sGAAsG;IACtG,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC;IACxB,+CAA+C;IAC/C,QAAQ,CAAC,iBAAiB,EAAE,OAAO,CAAC;IACpC,mEAAmE;IACnE,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,UAAU,EAAE,OAAO,CAAC;IAC7B,QAAQ,CAAC,eAAe,EAAE,OAAO,CAAC;IAClC,QAAQ,CAAC,WAAW,EAAE,OAAO,CAAC;IAC9B,QAAQ,CAAC,gBAAgB,EAAE,OAAO,CAAC;IACnC,uFAAuF;IACvF,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,CAAC;CACjC;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AAED;;;GAGG;AACH,MAAM,WAAW,QAAQ;IACvB,+DAA+D;IAC/D,YAAY,IAAI,oBAAoB,CAAC;IAErC,uBAAuB;IACvB,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC,CAAC;IAE1E,2BAA2B;IAC3B,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC,CAAC;IAE5F,8BAA8B;IAC9B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,WAAW,GAAG,QAAQ,CAAC;AAEnC,MAAM,WAAW,cAAc;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAChF,0FAA0F;IAC1F,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB,mGAAmG;IACnG,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,+FAA+F;IAC/F,eAAe,CAAC,EAAE,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,0FAA0F;IAC1F,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB,+FAA+F;IAC/F,eAAe,CAAC,EAAE,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACrC,CAAC;CACH;AAED,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC"}

package/dist/llm/providers/base.js

@@ -1,6 +1,10 @@
 /**
  * LLM Provider Base Interface
  *
- * Abstract interface for all LLM providers (OpenAI, Anthropic, etc.)
+ * The agnostic Reasoner port (ADR 2026-06-25). Providers are negotiated by
+ * `capabilities()`, never by provider-name string. The control vocabulary is neutral
+ * and capability-gated (`effort`, `cache`/`cacheHint`); genuinely unportable knobs ride
+ * the namespaced opaque `providerOptions` channel; cross-provider usage (e.g. cache token
+ * stats) stays neutral in `LLMResponse.usage`.
  */
 export {};

package/dist/llm/providers/groq.d.ts

@@ -5,7 +5,7 @@
  * Free tier: 6,000 TPM / 500k TPD.
  * Sign up: console.groq.com
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface GroqProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey: string;
     /** Defaults to https://api.groq.com/openai/v1 */
@@ -16,6 +16,7 @@ export interface GroqProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
 export declare class GroqProvider implements LLMProvider {
     private inner;
     constructor(config: GroqProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(_text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

package/dist/llm/providers/groq.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"groq.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/groq.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;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,kBAAmB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC3E,MAAM,EAAE,MAAM,CAAC;IACf,iDAAiD;IACjD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,qBAAa,YAAa,YAAW,WAAW;IAC9C,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,kBAAkB;IAQtC,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;IAIhF,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,QAAQ,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG9F"}
+{"version":3,"file":"groq.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/groq.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,kBAAmB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC3E,MAAM,EAAE,MAAM,CAAC;IACf,iDAAiD;IACjD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,qBAAa,YAAa,YAAW,WAAW;IAC9C,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,kBAAkB;IAQtC,YAAY,IAAI,oBAAoB;IAepC,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;IAIhF,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,QAAQ,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG9F"}

package/dist/llm/providers/groq.js

@@ -15,6 +15,20 @@ export class GroqProvider {
             model: config.model ?? 'qwen/qwen3-32b',
         });
     }
+    capabilities() {
+        // Profile for the default model (qwen/qwen3-32b). Groq exposes no embeddings endpoint.
+        return {
+            providerTag: 'groq',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: false,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

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

@@ -24,7 +24,7 @@
  *
  * Docs: https://documentation.ubuntu.com/inference-snaps
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface InferenceSnapsProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey?: string;
     /** Base URL of the inference-snaps service, e.g. http://localhost:9090/v1 */
@@ -39,6 +39,7 @@ export declare class InferenceSnapsProvider implements LLMProvider {
     private embedModel;
     private baseURL;
     constructor(config: InferenceSnapsProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

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;;;;;;;;;;;;;;;;;;;;;;;;;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"}
+{"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,EACP,oBAAoB,EACrB,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,YAAY,IAAI,oBAAoB;IAgBpC,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

@@ -41,6 +41,21 @@ export class InferenceSnapsProvider {
             model: config.model ?? 'gemma3',
         });
     }
+    capabilities() {
+        // Profile for the default model (gemma3, text-only). Per-model capability
+        // negotiation (e.g. qwen-vl vision, deepseek-r1 reasoningEffort) is a follow-up.
+        return {
+            providerTag: 'inference-snaps',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

package/dist/llm/providers/ollama.d.ts

@@ -5,7 +5,7 @@
  * No API key required. Zero cost, fully offline.
  * Install: https://ollama.com
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface OllamaProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey?: string;
     /** Defaults to http://localhost:11434/v1 */
@@ -20,6 +20,7 @@ export declare class OllamaProvider implements LLMProvider {
     private embedModel;
     private baseURL;
     constructor(config: OllamaProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

package/dist/llm/providers/ollama.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ollama.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;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,oBAAqB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC7E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gFAAgF;IAChF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iGAAiG;IACjG,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,cAAe,YAAW,WAAW;IAChD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,oBAAoB;IAaxC,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":"ollama.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,oBAAqB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC7E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gFAAgF;IAChF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iGAAiG;IACjG,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,cAAe,YAAW,WAAW;IAChD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,oBAAoB;IAaxC,YAAY,IAAI,oBAAoB;IAepC,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/ollama.js

@@ -22,6 +22,20 @@ export class OllamaProvider {
             model: config.model ?? 'gemma4:e2b',
         });
     }
+    capabilities() {
+        // Profile for the default model (gemma4:e2b, text-only).
+        return {
+            providerTag: 'ollama',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

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

@@ -5,13 +5,30 @@
  * 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';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface OpenAICompatConfig extends LLMProviderConfig {
+    /** Per-request timeout in ms applied to connection setup. Default 60_000. */
+    timeout?: number;
+    /** Retries on connection/timeout errors (not on HTTP error responses). Default 1. */
+    maxRetries?: number;
 }
 export declare class OpenAICompatProvider implements LLMProvider {
     private config;
     private baseURL;
+    private timeoutMs;
+    private maxRetries;
     constructor(config: OpenAICompatConfig);
+    /**
+     * fetch() with a connection-setup timeout and bounded retries. The timeout
+     * aborts a hung connection (the localhost-default failure mode that otherwise
+     * burns the whole serverless duration) and is cleared once headers arrive, so
+     * a legitimately long streaming body is not truncated. Retries fire only on a
+     * thrown fetch error (connection refused, DNS, timeout) — never on an HTTP
+     * error response, which the caller inspects — so a POST is re-sent only when
+     * no response was received.
+     */
+    private fetchWithResilience;
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     embed(text: string | string[], options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;

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

@@ -1 +1 @@
-{"version":3,"file":"openai-compat.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/openai-compat.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EAET,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EAER,MAAM,WAAW,CAAC;AAEnB,MAAM,WAAW,kBAAmB,SAAQ,iBAAiB;CAAG;AAqDhE,qBAAa,oBAAqB,YAAW,WAAW;IACtD,OAAO,CAAC,MAAM,CAAqB;IACnC,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,kBAAkB;IAUhC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IA6EzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IAyC5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAuFvF,OAAO,CAAC,cAAc;CA4BvB"}
+{"version":3,"file":"openai-compat.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/openai-compat.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EAET,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EAErB,MAAM,WAAW,CAAC;AAEnB,MAAM,WAAW,kBAAmB,SAAQ,iBAAiB;IAC3D,6EAA6E;IAC7E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,qFAAqF;IACrF,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAwDD,qBAAa,oBAAqB,YAAW,WAAW;IACtD,OAAO,CAAC,MAAM,CAAqB;IACnC,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,UAAU,CAAS;gBAEf,MAAM,EAAE,kBAAkB;IAYtC;;;;;;;;OAQG;YACW,mBAAmB;IAgBjC,YAAY,IAAI,oBAAoB;IAgB9B,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IA6EzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IAyC5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAuFvF,OAAO,CAAC,cAAc;CA4BvB"}

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

@@ -5,6 +5,8 @@
  * format. Used by: Ollama, Groq, Inference Snaps.
  * NOT for direct OpenAI usage  -  RevealUI uses open-source models only.
  */
+const DEFAULT_TIMEOUT_MS = 60_000;
+const DEFAULT_MAX_RETRIES = 1;
 const authorizationHeader = 'Authorization';
 const maxTokensKey = 'max_tokens';
 const toolChoiceKey = 'tool_choice';
@@ -31,15 +33,60 @@ const isFunctionToolCall = (call) => {
 export class OpenAICompatProvider {
     config;
     baseURL;
+    timeoutMs;
+    maxRetries;
     constructor(config) {
         this.config = config;
         if (!config.baseURL) {
             throw new Error('OpenAICompatProvider requires a baseURL  -  use a specific provider (InferenceSnapsProvider, OllamaProvider, etc.)');
         }
         this.baseURL = config.baseURL;
+        this.timeoutMs = config.timeout ?? DEFAULT_TIMEOUT_MS;
+        this.maxRetries = config.maxRetries ?? DEFAULT_MAX_RETRIES;
+    }
+    /**
+     * fetch() with a connection-setup timeout and bounded retries. The timeout
+     * aborts a hung connection (the localhost-default failure mode that otherwise
+     * burns the whole serverless duration) and is cleared once headers arrive, so
+     * a legitimately long streaming body is not truncated. Retries fire only on a
+     * thrown fetch error (connection refused, DNS, timeout) — never on an HTTP
+     * error response, which the caller inspects — so a POST is re-sent only when
+     * no response was received.
+     */
+    async fetchWithResilience(url, init) {
+        let lastError;
+        for (let attempt = 0; attempt <= this.maxRetries; attempt++) {
+            const controller = new AbortController();
+            const timeoutId = setTimeout(() => controller.abort(), this.timeoutMs);
+            try {
+                return await fetch(url, { ...init, signal: controller.signal });
+            }
+            catch (error) {
+                lastError = error;
+            }
+            finally {
+                clearTimeout(timeoutId);
+            }
+        }
+        throw lastError instanceof Error ? lastError : new Error(String(lastError));
+    }
+    capabilities() {
+        // Base OpenAI-compatible profile. Concrete providers (Groq, Ollama, inference-snaps)
+        // override per their endpoint and default model.
+        return {
+            providerTag: 'openai-compat',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
     }
     async chat(messages, options) {
-        const response = await fetch(`${this.baseURL}/chat/completions`, {
+        const response = await this.fetchWithResilience(`${this.baseURL}/chat/completions`, {
             method: 'POST',
             headers: {
                 'Content-Type': 'application/json',
@@ -107,7 +154,7 @@ export class OpenAICompatProvider {
     async embed(text, options) {
         const texts = Array.isArray(text) ? text : [text];
         const model = options?.model || 'text-embedding-3-small';
-        const response = await fetch(`${this.baseURL}/embeddings`, {
+        const response = await this.fetchWithResilience(`${this.baseURL}/embeddings`, {
             method: 'POST',
             headers: {
                 'Content-Type': 'application/json',
@@ -140,7 +187,7 @@ export class OpenAICompatProvider {
         return Array.isArray(text) ? embeddings : embeddings[0];
     }
     async *stream(messages, options) {
-        const response = await fetch(`${this.baseURL}/chat/completions`, {
+        const response = await this.fetchWithResilience(`${this.baseURL}/chat/completions`, {
             method: 'POST',
             headers: {
                 'Content-Type': 'application/json',

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

@@ -0,0 +1,27 @@
+/**
+ * OpenAI Provider
+ *
+ * Thin wrapper over OpenAICompatProvider targeting OpenAI's native
+ * chat/completions + embeddings API at https://api.openai.com/v1. OpenAI is the
+ * reference OpenAI-compatible surface, so the shared base implementation serves
+ * it directly — no proprietary SDK (fleet posture in client.ts).
+ */
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
+export interface OpenAIProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
+    apiKey: string;
+    /** Defaults to https://api.openai.com/v1 */
+    baseURL?: string;
+    /** Defaults to gpt-4o (conservative documented default; overridable) */
+    model?: string;
+    timeout?: number;
+    maxRetries?: number;
+}
+export declare class OpenAIProvider implements LLMProvider {
+    private inner;
+    constructor(config: OpenAIProviderConfig);
+    capabilities(): ReasonerCapabilities;
+    chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
+    stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
+    embed(text: string | string[], options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
+}
+//# sourceMappingURL=openai.d.ts.map

package/dist/llm/providers/openai.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/openai.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,oBAAqB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC7E,MAAM,EAAE,MAAM,CAAC;IACf,4CAA4C;IAC5C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,wEAAwE;IACxE,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,cAAe,YAAW,WAAW;IAChD,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,oBAAoB;IAQxC,YAAY,IAAI,oBAAoB;IAcpC,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;IAIhF,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG5F"}

package/dist/llm/providers/openai.js

@@ -0,0 +1,41 @@
+/**
+ * OpenAI Provider
+ *
+ * Thin wrapper over OpenAICompatProvider targeting OpenAI's native
+ * chat/completions + embeddings API at https://api.openai.com/v1. OpenAI is the
+ * reference OpenAI-compatible surface, so the shared base implementation serves
+ * it directly — no proprietary SDK (fleet posture in client.ts).
+ */
+import { OpenAICompatProvider } from './openai-compat.js';
+export class OpenAIProvider {
+    inner;
+    constructor(config) {
+        this.inner = new OpenAICompatProvider({
+            ...config,
+            baseURL: config.baseURL ?? 'https://api.openai.com/v1',
+            model: config.model ?? 'gpt-4o',
+        });
+    }
+    capabilities() {
+        return {
+            providerTag: 'openai',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
+    chat(messages, options) {
+        return this.inner.chat(messages, options);
+    }
+    stream(messages, options) {
+        return this.inner.stream(messages, options);
+    }
+    embed(text, options) {
+        return this.inner.embed(text, options);
+    }
+}