@relayplane/proxy 1.8.6 → 1.8.8

package/dist/ollama.d.ts

@@ -0,0 +1,264 @@
+/**
+ * Ollama Local Model Provider
+ *
+ * Routes requests to a local Ollama instance for cost-free inference.
+ * Handles API format translation between OpenAI-compatible and Ollama's
+ * native /api/chat endpoint.
+ *
+ * Features:
+ * - Configurable base URL (default: http://localhost:11434)
+ * - Model availability detection with graceful fallback
+ * - Complexity-based routing (route "simple" tasks to local models)
+ * - OpenAI-compatible response format translation
+ * - Streaming support (SSE)
+ * - Health checking for availability
+ *
+ * Ollama API reference: https://github.com/ollama/ollama/blob/main/docs/api.md
+ *
+ * @packageDocumentation
+ */
+/**
+ * Ollama provider configuration.
+ *
+ * Example in ~/.relayplane/config.json:
+ * ```json
+ * {
+ *   "providers": {
+ *     "ollama": {
+ *       "baseUrl": "http://localhost:11434",
+ *       "models": ["llama3.2", "codestral"],
+ *       "routeWhen": {
+ *         "complexity": ["simple"]
+ *       }
+ *     }
+ *   }
+ * }
+ * ```
+ */
+export interface OllamaProviderConfig {
+    /** Base URL for the Ollama API (default: http://localhost:11434) */
+    baseUrl?: string;
+    /** List of available/preferred model names */
+    models?: string[];
+    /** Routing conditions: when to use Ollama instead of cloud providers */
+    routeWhen?: {
+        /** Complexity levels that should be routed to Ollama (e.g., ["simple"]) */
+        complexity?: string[];
+        /** Task types that should be routed to Ollama (e.g., ["question_answering"]) */
+        taskTypes?: string[];
+    };
+    /** Timeout in milliseconds for Ollama requests (default: 120000 = 2 min) */
+    timeoutMs?: number;
+    /** Default model to use when none specified (default: first in models list) */
+    defaultModel?: string;
+    /** Enable Ollama provider (default: true when configured) */
+    enabled?: boolean;
+}
+/** Default Ollama configuration values */
+export declare const OLLAMA_DEFAULTS: {
+    readonly baseUrl: "http://localhost:11434";
+    readonly timeoutMs: 120000;
+    readonly models: string[];
+    readonly enabled: true;
+};
+/**
+ * Ollama /api/chat request body
+ * @see https://github.com/ollama/ollama/blob/main/docs/api.md#generate-a-chat-completion
+ */
+interface OllamaChatRequest {
+    model: string;
+    messages: Array<{
+        role: string;
+        content: string;
+    }>;
+    stream: boolean;
+    options?: {
+        temperature?: number;
+        num_predict?: number;
+        top_p?: number;
+        top_k?: number;
+    };
+    tools?: unknown[];
+}
+/**
+ * Ollama /api/chat response
+ */
+interface OllamaChatResponse {
+    model: string;
+    message: {
+        role: string;
+        content: string;
+        tool_calls?: Array<{
+            function: {
+                name: string;
+                arguments: Record<string, unknown>;
+            };
+        }>;
+    };
+    done: boolean;
+    total_duration?: number;
+    load_duration?: number;
+    prompt_eval_count?: number;
+    eval_count?: number;
+    eval_duration?: number;
+}
+/**
+ * Ollama streaming chunk (NDJSON)
+ */
+interface OllamaStreamChunk {
+    model: string;
+    message: {
+        role: string;
+        content: string;
+    };
+    done: boolean;
+    total_duration?: number;
+    prompt_eval_count?: number;
+    eval_count?: number;
+}
+/**
+ * Result of an Ollama health/availability check
+ */
+export interface OllamaHealthResult {
+    available: boolean;
+    models: string[];
+    error?: string;
+    responseTimeMs?: number;
+}
+/**
+ * Check if the Ollama server is running and list available models.
+ * Non-blocking, with a short timeout.
+ */
+export declare function checkOllamaHealth(baseUrl?: string, timeoutMs?: number): Promise<OllamaHealthResult>;
+/**
+ * Determine whether a request should be routed to Ollama based on config.
+ *
+ * @param config     - Ollama provider configuration
+ * @param complexity - Classified complexity of the request (simple/moderate/complex)
+ * @param taskType   - Inferred task type
+ * @param model      - Requested model name (may match an Ollama model)
+ *
+ * @returns true if the request should be routed to Ollama
+ */
+export declare function shouldRouteToOllama(config: OllamaProviderConfig, complexity?: string, taskType?: string, model?: string): boolean;
+/**
+ * Resolve the Ollama model name from a requested model string.
+ *
+ * - "ollama/llama3.2" → "llama3.2"
+ * - "llama3.2" (if in models list) → "llama3.2"
+ * - Unknown model → defaultModel or first in models list
+ */
+export declare function resolveOllamaModel(requestedModel: string, config: OllamaProviderConfig): string;
+/**
+ * Convert OpenAI-format messages to Ollama format.
+ *
+ * Ollama messages are simpler:
+ * - role: "system" | "user" | "assistant"
+ * - content: string (no array / multimodal blocks in base API)
+ * - Tool calls are passed through for models that support them
+ */
+export declare function convertMessagesToOllama(messages: Array<{
+    role: string;
+    content: string | unknown;
+    [key: string]: unknown;
+}>): Array<{
+    role: string;
+    content: string;
+}>;
+/**
+ * Build an Ollama /api/chat request body from an OpenAI-compatible request.
+ */
+export declare function buildOllamaRequest(model: string, messages: Array<{
+    role: string;
+    content: string | unknown;
+    [key: string]: unknown;
+}>, stream: boolean, options?: {
+    temperature?: number;
+    max_tokens?: number;
+    tools?: unknown[];
+}): OllamaChatRequest;
+/**
+ * Convert an Ollama /api/chat response to OpenAI chat completion format.
+ */
+export declare function convertOllamaResponse(ollamaData: OllamaChatResponse, requestedModel: string): Record<string, unknown>;
+/**
+ * Convert Ollama NDJSON streaming chunk to OpenAI SSE format.
+ */
+export declare function convertOllamaStreamChunk(chunk: OllamaStreamChunk, messageId: string, isFirst: boolean): string | null;
+/**
+ * Forward a non-streaming request to Ollama and return the response
+ * translated to OpenAI format.
+ */
+export declare function forwardToOllama(model: string, messages: Array<{
+    role: string;
+    content: string | unknown;
+    [key: string]: unknown;
+}>, options?: {
+    temperature?: number;
+    max_tokens?: number;
+    tools?: unknown[];
+    baseUrl?: string;
+    timeoutMs?: number;
+}): Promise<{
+    success: boolean;
+    data?: Record<string, unknown>;
+    usage?: {
+        prompt_tokens: number;
+        completion_tokens: number;
+        total_tokens: number;
+    };
+    error?: {
+        code: string;
+        message: string;
+        status: number;
+        retryable: boolean;
+    };
+    latencyMs: number;
+}>;
+/**
+ * Forward a streaming request to Ollama.
+ * Returns a readable stream of OpenAI-format SSE events.
+ *
+ * Ollama streams NDJSON (newline-delimited JSON), which we convert
+ * to OpenAI SSE (data: {...}\n\n) on the fly.
+ */
+export declare function forwardToOllamaStream(model: string, messages: Array<{
+    role: string;
+    content: string | unknown;
+    [key: string]: unknown;
+}>, options?: {
+    temperature?: number;
+    max_tokens?: number;
+    tools?: unknown[];
+    baseUrl?: string;
+    timeoutMs?: number;
+}): Promise<{
+    success: boolean;
+    stream?: AsyncGenerator<string, void, unknown>;
+    error?: {
+        code: string;
+        message: string;
+        status: number;
+        retryable: boolean;
+    };
+}>;
+/**
+ * Check Ollama health with caching to avoid excessive probing.
+ */
+export declare function checkOllamaHealthCached(baseUrl?: string): Promise<OllamaHealthResult>;
+/**
+ * Clear the health cache (useful for testing or after config changes).
+ */
+export declare function clearOllamaHealthCache(): void;
+/**
+ * Map cloud model names to reasonable Ollama equivalents.
+ * Used when falling back to Ollama from a cloud provider.
+ */
+export declare const CLOUD_TO_OLLAMA_MODEL_MAP: Record<string, string>;
+/**
+ * Map a cloud model name to an Ollama model.
+ * Returns the configured default or a reasonable fallback.
+ */
+export declare function mapCloudModelToOllama(cloudModel: string, config: OllamaProviderConfig): string;
+export {};
+//# sourceMappingURL=ollama.d.ts.map

package/dist/ollama.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ollama.d.ts","sourceRoot":"","sources":["../src/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,WAAW,oBAAoB;IACnC,oEAAoE;IACpE,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,8CAA8C;IAC9C,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,wEAAwE;IACxE,SAAS,CAAC,EAAE;QACV,2EAA2E;QAC3E,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;QACtB,gFAAgF;QAChF,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;KACtB,CAAC;IACF,4EAA4E;IAC5E,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,+EAA+E;IAC/E,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,6DAA6D;IAC7D,OAAO,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,0CAA0C;AAC1C,eAAO,MAAM,eAAe;;;qBAGZ,MAAM,EAAE;;CAEd,CAAC;AAEX;;;GAGG;AACH,UAAU,iBAAiB;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACnD,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE;QACR,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;CACnB;AAED;;GAEG;AACH,UAAU,kBAAkB;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE;QACP,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,CAAC,EAAE,KAAK,CAAC;YACjB,QAAQ,EAAE;gBAAE,IAAI,EAAE,MAAM,CAAC;gBAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;aAAE,CAAC;SAChE,CAAC,CAAC;KACJ,CAAC;IACF,IAAI,EAAE,OAAO,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED;;GAEG;AACH,UAAU,iBAAiB;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC;IAC3C,IAAI,EAAE,OAAO,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAcD;;GAEG;AACH,MAAM,WAAW,kBAAkB;IACjC,SAAS,EAAE,OAAO,CAAC;IACnB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAMD;;;GAGG;AACH,wBAAsB,iBAAiB,CACrC,OAAO,GAAE,MAAgC,EACzC,SAAS,GAAE,MAAa,GACvB,OAAO,CAAC,kBAAkB,CAAC,CAuC7B;AAMD;;;;;;;;;GASG;AACH,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,oBAAoB,EAC5B,UAAU,CAAC,EAAE,MAAM,EACnB,QAAQ,CAAC,EAAE,MAAM,EACjB,KAAK,CAAC,EAAE,MAAM,GACb,OAAO,CA+BT;AAED;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CAChC,cAAc,EAAE,MAAM,EACtB,MAAM,EAAE,oBAAoB,GAC3B,MAAM,CAaR;AAiBD;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CACrC,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CAAE,CAAC,GACnF,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CAkB1C;AAED;;GAEG;AACH,wBAAgB,kBAAkB,CAChC,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CAAE,CAAC,EACpF,MAAM,EAAE,OAAO,EACf,OAAO,CAAC,EAAE;IACR,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;CACnB,GACA,iBAAiB,CA0BnB;AAMD;;GAEG;AACH,wBAAgB,qBAAqB,CACnC,UAAU,EAAE,kBAAkB,EAC9B,cAAc,EAAE,MAAM,GACrB,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAyCzB;AAED;;GAEG;AACH,wBAAgB,wBAAwB,CACtC,KAAK,EAAE,iBAAiB,EACxB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,OAAO,GACf,MAAM,GAAG,IAAI,CAyCf;AAMD;;;GAGG;AACH,wBAAsB,eAAe,CACnC,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CAAE,CAAC,EACpF,OAAO,CAAC,EAAE;IACR,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GACA,OAAO,CAAC;IACT,OAAO,EAAE,OAAO,CAAC;IACjB,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,KAAK,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAC;QAAC,iBAAiB,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAA;KAAE,CAAC;IACnF,KAAK,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,OAAO,CAAA;KAAE,CAAC;IAC9E,SAAS,EAAE,MAAM,CAAC;CACnB,CAAC,CA2DD;AAED;;;;;;GAMG;AACH,wBAAsB,qBAAqB,CACzC,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CAAE,CAAC,EACpF,OAAO,CAAC,EAAE;IACR,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GACA,OAAO,CAAC;IACT,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,cAAc,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;IAC/C,KAAK,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,OAAO,CAAA;KAAE,CAAC;CAC/E,CAAC,CA8GD;AASD;;GAEG;AACH,wBAAsB,uBAAuB,CAC3C,OAAO,GAAE,MAAgC,GACxC,OAAO,CAAC,kBAAkB,CAAC,CAS7B;AAED;;GAEG;AACH,wBAAgB,sBAAsB,IAAI,IAAI,CAE7C;AAMD;;;GAGG;AACH,eAAO,MAAM,yBAAyB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAc5D,CAAC;AAEF;;;GAGG;AACH,wBAAgB,qBAAqB,CACnC,UAAU,EAAE,MAAM,EAClB,MAAM,EAAE,oBAAoB,GAC3B,MAAM,CAUR"}