@warlock.js/ai-ollama 4.1.1

package/cjs/index.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.cjs","names":["InvalidRequestError","AIError","ProviderTimeoutError","ProviderError","ProviderAuthError","ProviderRateLimitError","ContextLengthExceededError","InvalidRequestError","LOG_MODULE","log","log","Ollama"],"sources":["../../../../../@warlock.js/ai-ollama/src/utils/map-done-reason.ts","../../../../../@warlock.js/ai-ollama/src/utils/to-ollama-messages.ts","../../../../../@warlock.js/ai-ollama/src/utils/to-ollama-tools.ts","../../../../../@warlock.js/ai-ollama/src/utils/wrap-ollama-error.ts","../../../../../@warlock.js/ai-ollama/src/embedder.ts","../../../../../@warlock.js/ai-ollama/src/known-vision-models.ts","../../../../../@warlock.js/ai-ollama/src/model.ts","../../../../../@warlock.js/ai-ollama/src/sdk.ts"],"sourcesContent":["import type { FinishReason } from \"@warlock.js/ai\";\n\nconst doneReasonMap: Record<string, FinishReason> = {\n  stop: \"stop\",\n  length: \"length\",\n};\n\n/**\n * Map Ollama's `done_reason` to the normalized `FinishReason` union.\n *\n * `stop` is the natural terminal; `length` means the `num_predict`\n * cap was hit. Anything else — `load` (model load only, no\n * generation), an empty string, or any future value — falls through\n * to `\"error\"`.\n *\n * Note: Ollama has no tool-use done reason — it sets `done_reason:\n * \"stop\"` and populates `message.tool_calls`. `OllamaModel` derives\n * `\"tool_calls\"` from tool-call presence; this map stays purely about\n * the raw signal.\n *\n * @example\n * mapDoneReason(\"stop\");    // \"stop\"\n * mapDoneReason(\"length\");  // \"length\"\n * mapDoneReason(\"load\");    // \"error\"\n * mapDoneReason(undefined); // \"error\"\n */\nexport function mapDoneReason(raw: string | null | undefined): FinishReason {\n  return doneReasonMap[raw ?? \"\"] ?? \"error\";\n}\n","import { InvalidRequestError, type ContentPart, type Message } from \"@warlock.js/ai\";\nimport type { Message as OllamaMessage } from \"ollama\";\n\n/**\n * Convert vendor-neutral `Message[]` into the Ollama chat message\n * shape.\n *\n * Unlike Anthropic / Gemini / Bedrock, Ollama keeps a first-class\n * `system` role inside `messages`, so there is no system-prompt\n * hoisting — roles pass straight through. The Ollama specifics this\n * absorbs:\n *\n * 1. **Tool calls.** An assistant message with `toolCalls` becomes an\n *    `assistant` message whose `tool_calls` is the Ollama\n *    `{ function: { name, arguments } }` shape (Ollama has no tool-call\n *    id — see `OllamaModel`/decisions for the synthesized-id note).\n * 2. **Tool results.** A neutral `tool` message becomes a `tool`\n *    message with `tool_name` set from `toolCallId` (Ollama matches a\n *    result to its call by tool name).\n * 3. **Images.** Multipart user content collapses to a single\n *    `content` string plus an `images` array of base64 strings.\n *\n * @example\n * const messages = toOllamaMessages([\n *   { role: \"system\", content: \"Be concise.\" },\n *   { role: \"user\", content: \"Hi\" },\n * ]);\n */\nexport function toOllamaMessages(messages: Message[]): OllamaMessage[] {\n  return messages.map((message): OllamaMessage => {\n    if (message.role === \"tool\") {\n      return {\n        role: \"tool\",\n        content: stringifyContent(message.content),\n        tool_name: message.toolCallId ?? \"\",\n      };\n    }\n\n    if (message.role === \"assistant\" && message.toolCalls && message.toolCalls.length > 0) {\n      return {\n        role: \"assistant\",\n        content: stringifyContent(message.content),\n        tool_calls: message.toolCalls.map((toolCall) => ({\n          function: {\n            name: toolCall.name,\n            arguments: (toolCall.input ?? {}) as Record<string, unknown>,\n          },\n        })),\n      };\n    }\n\n    if (message.role === \"user\" && Array.isArray(message.content)) {\n      return toMultipartMessage(message.content);\n    }\n\n    return { role: message.role, content: stringifyContent(message.content) };\n  });\n}\n\n/**\n * Collapse a `ContentPart[]` user message into Ollama's\n * single-string-content + base64-`images` shape. Ollama cannot fetch\n * remote URLs, so a `{ url }` image surfaces a typed\n * `InvalidRequestError` upfront (consistent with the Bedrock/Gemini\n * adapters). The agent has already resolved attachments — nothing is\n * fetched here.\n */\nfunction toMultipartMessage(parts: ContentPart[]): OllamaMessage {\n  const textChunks: string[] = [];\n  const images: string[] = [];\n\n  for (const part of parts) {\n    if (part.type === \"text\") {\n      textChunks.push(part.text);\n\n      continue;\n    }\n\n    if (\"url\" in part.source) {\n      throw new InvalidRequestError(\n        \"Ollama does not fetch remote-URL images; supply base64 image bytes instead.\",\n      );\n    }\n\n    images.push(part.source.base64);\n  }\n\n  return {\n    role: \"user\",\n    content: textChunks.join(\"\"),\n    ...(images.length > 0 ? { images } : {}),\n  };\n}\n\n/**\n * Multipart content on a non-user role collapses to concatenated text;\n * plain strings pass through unchanged.\n */\nfunction stringifyContent(content: string | ContentPart[]): string {\n  if (typeof content === \"string\") {\n    return content;\n  }\n\n  return content\n    .filter((part): part is { type: \"text\"; text: string } => part.type === \"text\")\n    .map((part) => part.text)\n    .join(\"\");\n}\n","import { extractJsonSchema, type ToolConfig } from \"@warlock.js/ai\";\nimport type { Tool } from \"ollama\";\n\n/**\n * Convert vendor-neutral `ToolConfig[]` into Ollama's `tools` array.\n * Each tool becomes a `{ type: \"function\", function: { name,\n * description, parameters } }` entry. Non-object extractions degrade\n * to a parameterless object so registration never fails.\n *\n * Returns `undefined` when there are no tools so the caller can omit\n * `tools` from the request.\n *\n * @example\n * const tools = toOllamaTools([weatherTool]);\n * await ollama.chat({ model, messages, tools });\n */\nexport function toOllamaTools(\n  tools: ToolConfig<unknown, unknown>[] | undefined,\n): Tool[] | undefined {\n  if (!tools || tools.length === 0) {\n    return undefined;\n  }\n\n  return tools.map((tool) => ({\n    type: \"function\",\n    function: {\n      name: tool.name,\n      description: tool.description,\n      parameters: toParameters(tool.input),\n    },\n  }));\n}\n\n/**\n * Resolve a tool's input schema to a JSON-Schema object. Ollama wants\n * an object root for function parameters; anything else (or a failed\n * extraction) degrades to a parameterless object.\n */\nfunction toParameters(input: ToolConfig<unknown, unknown>[\"input\"]): Tool[\"function\"][\"parameters\"] {\n  const schema = extractJsonSchema(input);\n\n  if (schema && schema.type === \"object\") {\n    return schema as Tool[\"function\"][\"parameters\"];\n  }\n\n  return { type: \"object\" } as Tool[\"function\"][\"parameters\"];\n}\n","import {\n  AIError,\n  ContextLengthExceededError,\n  InvalidRequestError,\n  ProviderAuthError,\n  ProviderError,\n  ProviderRateLimitError,\n  ProviderTimeoutError,\n} from \"@warlock.js/ai\";\n\n/**\n * Raw-error fields the wrapper reads off an Ollama client error. The\n * `ollama` client throws a `ResponseError` (`name: \"ResponseError\"`,\n * numeric `status_code`, message = the server's `error` text) for HTTP\n * faults; transport failures surface as a `fetch`-layer `TypeError`\n * with an `ECONNREFUSED` / `ETIMEDOUT` cause. We duck-type both —\n * `ResponseError` is internal to the package and not exported.\n */\ntype OllamaErrorShape = {\n  name?: string;\n  message?: string;\n  statusCode?: number;\n  code?: string;\n};\n\n/**\n * Wrap any thrown value caught inside the Ollama adapter into the\n * appropriate `@warlock.js/ai` `AIError` subclass.\n *\n * **Dispatch strategy.** HTTP faults carry `status_code`; the local\n * daemon being down surfaces as a connection error (`ECONNREFUSED` /\n * \"fetch failed\") — mapped to `ProviderError` since it's an\n * operational \"is Ollama running?\" condition, not a request defect.\n * `400` with context-length phrasing maps to\n * `ContextLengthExceededError`.\n *\n * `AIError` instances pass through unchanged so `catch/throw wrap(e)`\n * pipelines never double-wrap.\n *\n * @example\n * try {\n *   return await this.client.chat({ ... });\n * } catch (thrown) {\n *   throw wrapOllamaError(thrown);\n * }\n */\nexport function wrapOllamaError(thrown: unknown): AIError {\n  if (thrown instanceof AIError) {\n    return thrown;\n  }\n\n  const shape = toShape(thrown);\n  const context = buildContext(shape);\n  const message = shape.message ?? (thrown instanceof Error ? thrown.message : String(thrown));\n\n  if (isTimeout(shape)) {\n    return new ProviderTimeoutError(message, { cause: thrown, context });\n  }\n\n  if (isConnectionRefused(shape, message)) {\n    return new ProviderError(message, { cause: thrown, context });\n  }\n\n  if (shape.statusCode === 401 || shape.statusCode === 403) {\n    return new ProviderAuthError(message, { cause: thrown, context });\n  }\n\n  if (shape.statusCode === 429) {\n    return new ProviderRateLimitError(message, { cause: thrown, context });\n  }\n\n  if (isClientStatus(shape.statusCode)) {\n    if (/context length|too long|exceeds|maximum context/i.test(message)) {\n      return new ContextLengthExceededError(message, { cause: thrown, context });\n    }\n\n    return new InvalidRequestError(message, { cause: thrown, context });\n  }\n\n  return new ProviderError(message, { cause: thrown, context });\n}\n\n/**\n * Read the raw error shape. `ResponseError` exposes `status_code`;\n * fetch-layer errors carry a `cause` whose `code` is the OS-level\n * socket error.\n */\nfunction toShape(thrown: unknown): OllamaErrorShape {\n  if (typeof thrown !== \"object\" || thrown === null) {\n    return {};\n  }\n\n  const raw = thrown as Record<string, unknown>;\n  const cause = raw.cause as Record<string, unknown> | undefined;\n\n  return {\n    name: typeof raw.name === \"string\" ? raw.name : undefined,\n    message: typeof raw.message === \"string\" ? raw.message : undefined,\n    statusCode: typeof raw.status_code === \"number\" ? raw.status_code : undefined,\n    code:\n      typeof raw.code === \"string\"\n        ? raw.code\n        : cause && typeof cause.code === \"string\"\n          ? cause.code\n          : undefined,\n  };\n}\n\n/** Transport-level timeout signals. */\nfunction isTimeout(shape: OllamaErrorShape): boolean {\n  if (shape.name === \"AbortError\" || shape.name === \"TimeoutError\") {\n    return true;\n  }\n\n  return shape.code === \"ETIMEDOUT\" || shape.code === \"ECONNABORTED\";\n}\n\n/**\n * The Ollama daemon not being reachable (most common local failure):\n * connection refused at the socket layer, or the `fetch failed`\n * TypeError the client surfaces when the host is down.\n */\nfunction isConnectionRefused(shape: OllamaErrorShape, message: string): boolean {\n  return shape.code === \"ECONNREFUSED\" || /fetch failed|econnrefused/i.test(message);\n}\n\n/** True for HTTP 4xx — a client-side request problem, not a server fault. */\nfunction isClientStatus(status: number | undefined): boolean {\n  return typeof status === \"number\" && status >= 400 && status < 500;\n}\n\n/** Attach the diagnostic fields to `error.context`. */\nfunction buildContext(shape: OllamaErrorShape): Record<string, unknown> {\n  const context: Record<string, unknown> = {};\n\n  if (shape.statusCode !== undefined) {\n    context.status = shape.statusCode;\n  }\n\n  if (shape.code) {\n    context.code = shape.code;\n  }\n\n  return context;\n}\n","import {\n  type EmbeddingBatchResult,\n  type EmbeddingResult,\n  type EmbeddingUsage,\n  type EmbedderContract,\n} from \"@warlock.js/ai\";\nimport { log, type Logger } from \"@warlock.js/logger\";\nimport type { EmbedResponse, Ollama } from \"ollama\";\nimport type { OllamaEmbedderConfig } from \"./config.type\";\nimport { wrapOllamaError } from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/**\n * Ollama-backed implementation of `EmbedderContract`\n * (`nomic-embed-text`, `mxbai-embed-large`, …) via `client.embed`.\n *\n * **Role.** Converts text into floating-point vectors. Standalone\n * primitive — unrelated to chat / tools / the agent loop.\n *\n * **Batch is native.** Ollama's `embed` accepts a string array and\n * returns `embeddings` in input order, so `embedMany` is a single\n * request (like the Gemini adapter, unlike Bedrock/Titan).\n *\n * **Usage.** Ollama returns only `prompt_eval_count` (no separate\n * total); it is reported as both `promptTokens` and `totalTokens`.\n *\n * **Dimensions.** When no `dimensions` override is given,\n * `this.dimensions` starts at `0` and is populated from the first\n * response's vector length, then cached. Passing `dimensions`\n * forwards Ollama's truncation field and sets the initial value.\n *\n * @example\n * const embedder = new OllamaEmbedder(client, { name: \"nomic-embed-text\" });\n * const { vector } = await embedder.embed(\"Hello world\");\n * const { vectors } = await embedder.embedMany([\"doc 1\", \"doc 2\"]);\n */\nexport class OllamaEmbedder implements EmbedderContract {\n  public readonly name: string;\n  public readonly provider: string;\n  public dimensions: number;\n\n  private readonly client: Ollama;\n  private readonly configuredDimensions: number | undefined;\n  private readonly logger: Logger = log;\n\n  public constructor(\n    client: Ollama,\n    config: OllamaEmbedderConfig,\n    provider: string = \"ollama\",\n  ) {\n    this.client = client;\n    this.name = config.name;\n    this.provider = provider;\n    this.configuredDimensions = config.dimensions;\n    this.dimensions = config.dimensions ?? 0;\n  }\n\n  public async embed(input: string): Promise<EmbeddingResult> {\n    const { embeddings, usage } = await this.request([input]);\n\n    return { vector: embeddings[0] ?? [], dimensions: this.dimensions, usage };\n  }\n\n  public async embedMany(inputs: string[]): Promise<EmbeddingBatchResult> {\n    const { embeddings, usage } = await this.request(inputs);\n\n    return { vectors: embeddings, dimensions: this.dimensions, usage };\n  }\n\n  /**\n   * Shared transport: one `embed` call for the whole batch, wrap\n   * provider errors, cache `dimensions` from the first vector, and\n   * return vectors in input order plus a neutral usage object.\n   */\n  private async request(\n    inputs: string[],\n  ): Promise<{ embeddings: number[][]; usage: EmbeddingUsage }> {\n    this.logger.debug(LOG_MODULE, \"embedder.request\", \"embed\", {\n      model: this.name,\n      count: inputs.length,\n    });\n\n    let response: EmbedResponse;\n\n    try {\n      response = await this.client.embed({\n        model: this.name,\n        input: inputs,\n        ...(this.configuredDimensions !== undefined\n          ? { dimensions: this.configuredDimensions }\n          : {}),\n      });\n    } catch (thrown) {\n      const wrapped = wrapOllamaError(thrown);\n\n      this.logger.error(LOG_MODULE, \"embedder.error\", wrapped.message, {\n        code: wrapped.code,\n        context: wrapped.context,\n      });\n\n      throw wrapped;\n    }\n\n    const embeddings = response.embeddings ?? [];\n\n    if (this.dimensions === 0 && embeddings[0]) {\n      this.dimensions = embeddings[0].length;\n    }\n\n    const tokens = response.prompt_eval_count ?? 0;\n    const usage: EmbeddingUsage = { promptTokens: tokens, totalTokens: tokens };\n\n    this.logger.debug(LOG_MODULE, \"embedder.response\", \"embed returned\", {\n      count: embeddings.length,\n      dimensions: this.dimensions,\n    });\n\n    return { embeddings, usage };\n  }\n}\n","/**\n * Substrings identifying Ollama model tags whose family accepts image\n * input (vision).\n *\n * Ollama tags are family-named with optional size/quant suffixes\n * (`llama3.2-vision:11b`, `llava:13b-v1.6`, `qwen2.5-vl:7b`). A\n * substring match tolerates those suffixes. Covers the common\n * multimodal families on the Ollama registry; text-only models\n * (`llama3.1`, `mistral`, `phi3`, `nomic-embed-text`) are excluded.\n * Override per-model via `ollama.model({ name, vision: true | false })`.\n */\nconst VISION_CAPABLE_SUBSTRINGS = [\n  \"llava\",\n  \"vision\",\n  \"bakllava\",\n  \"moondream\",\n  \"minicpm-v\",\n  \"qwen2-vl\",\n  \"qwen2.5-vl\",\n  \"llama4\",\n  \"gemma3\",\n];\n\n/**\n * Infer whether an Ollama model tag supports vision based on the known\n * multimodal-family substrings. Unknown tags default to `false` so\n * passing an image to a text-only local model surfaces a clear,\n * agent-side capability error instead of the image being silently\n * ignored by the model.\n *\n * @example\n * inferVisionCapability(\"llama3.2-vision:11b\"); // → true\n * inferVisionCapability(\"llava:13b\");           // → true\n * inferVisionCapability(\"llama3.1\");            // → false\n * inferVisionCapability(\"nomic-embed-text\");    // → false\n */\nexport function inferVisionCapability(modelName: string): boolean {\n  const normalized = modelName.toLowerCase();\n\n  return VISION_CAPABLE_SUBSTRINGS.some((fragment) => normalized.includes(fragment));\n}\n","import {\n  type Message,\n  type ModelCallOptions,\n  type ModelCapabilities,\n  type ModelContract,\n  type ModelPricing,\n  type ModelResponse,\n  type ModelStreamChunk,\n  type ModelToolCallRequest,\n  type Usage,\n} from \"@warlock.js/ai\";\nimport { log, type Logger } from \"@warlock.js/logger\";\nimport type {\n  AbortableAsyncIterator,\n  ChatRequest,\n  ChatResponse,\n  Ollama,\n  Options,\n} from \"ollama\";\nimport type { OllamaModelConfig } from \"./config.type\";\nimport { inferVisionCapability } from \"./known-vision-models\";\nimport { mapDoneReason, toOllamaMessages, toOllamaTools, wrapOllamaError } from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/**\n * Ollama-backed implementation of `ModelContract`.\n *\n * **Role.** The provider-facing bridge between the vendor-neutral\n * `@warlock.js/ai` agent runtime and a local (or self-hosted) Ollama\n * server via the official `ollama` client.\n *\n * **Responsibility.**\n * - Owns: a long-lived `Ollama` client + frozen `ModelConfig` (model\n *   tag, temperature, maxTokens) used as per-call defaults.\n * - Owns: translating vendor-neutral `Message[]` / `ToolConfig[]` into\n *   Ollama's chat shapes (system stays a real role, `tool_calls` /\n *   `tool_name`, base64 `images`) and Ollama's response (content, tool\n *   calls, done reason, eval-count usage) back into neutral shapes.\n * - Does NOT own: tool dispatch, looping, history, retries — agent\n *   concerns. The model is a per-call protocol adapter.\n *\n * **Tool-call ids.** Ollama has no tool-call id concept — a `tool_call`\n * is `{ function: { name, arguments } }`. The adapter synthesizes the\n * neutral `id` from the tool name so the agent's tool-result round-trip\n * (which keys on `toolCallId`) maps back to Ollama's name-based\n * matching. Parallel calls to the *same* tool in one turn therefore\n * share an id — a documented v1 limitation inherent to Ollama's wire\n * format, not this adapter.\n *\n * Modeled as a class (see §4.2 of code-style.md — \"long-lived state\n * across calls\").\n *\n * @example\n * import { Ollama } from \"ollama\";\n * const client = new Ollama({ host: \"http://127.0.0.1:11434\" });\n * const model = new OllamaModel(client, { name: \"llama3.1\" });\n *\n * const myAgent = agent({ model, tools: [searchTool] });\n * const result = await myAgent.execute(\"Summarize today's news.\");\n */\nexport class OllamaModel implements ModelContract {\n  public readonly name: string;\n  public readonly provider: string;\n  public readonly capabilities: ModelCapabilities;\n  public readonly pricing?: ModelPricing;\n\n  private readonly client: Ollama;\n  private readonly config: OllamaModelConfig;\n  private readonly logger: Logger = log;\n\n  public constructor(client: Ollama, config: OllamaModelConfig, provider: string = \"ollama\") {\n    this.client = client;\n    this.config = config;\n    this.name = config.name;\n    this.provider = provider;\n    this.pricing = config.pricing;\n    this.capabilities = {\n      structuredOutput: config.structuredOutput ?? true,\n      vision: config.vision ?? inferVisionCapability(config.name),\n    };\n  }\n\n  /**\n   * Single-shot completion. Sends the full message list to\n   * `client.chat`, waits for the terminal response, and reshapes it\n   * into a vendor-neutral `ModelResponse`. Per-call `options` override\n   * the instance defaults for this call only.\n   */\n  public async complete(messages: Message[], options?: ModelCallOptions): Promise<ModelResponse> {\n    this.logger.debug(LOG_MODULE, \"request\", \"Starting chat call\", {\n      model: this.name,\n      messageCount: messages.length,\n      streaming: false,\n      toolCount: options?.tools?.length ?? 0,\n    });\n\n    let response: ChatResponse;\n\n    try {\n      response = await this.client.chat({ ...this.buildRequest(messages, options), stream: false });\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    const toolCalls = this.extractToolCalls(response.message);\n    const finishReason = toolCalls ? \"tool_calls\" : mapDoneReason(response.done_reason);\n    const usage = this.extractUsage(response);\n\n    this.logger.debug(LOG_MODULE, \"response\", \"chat call succeeded\", { finishReason, usage });\n\n    return {\n      content: response.message?.content ?? \"\",\n      finishReason,\n      usage,\n      toolCalls,\n    };\n  }\n\n  /**\n   * Incremental streaming completion. Yields neutral\n   * `ModelStreamChunk`s — `delta` for content, `tool-call` per\n   * function call (Ollama streams a fully-formed call, not partial\n   * JSON), and a terminal `done` with the final finish reason + usage.\n   * Honors `options.signal` by aborting the underlying stream.\n   */\n  public async *stream(\n    messages: Message[],\n    options?: ModelCallOptions,\n  ): AsyncIterable<ModelStreamChunk> {\n    this.logger.debug(LOG_MODULE, \"request\", \"Starting streaming chat call\", {\n      model: this.name,\n      messageCount: messages.length,\n      streaming: true,\n      toolCount: options?.tools?.length ?? 0,\n    });\n\n    let stream: AbortableAsyncIterator<ChatResponse>;\n\n    try {\n      stream = await this.client.chat({ ...this.buildRequest(messages, options), stream: true });\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    if (options?.signal) {\n      if (options.signal.aborted) {\n        stream.abort();\n      } else {\n        options.signal.addEventListener(\"abort\", () => stream.abort(), { once: true });\n      }\n    }\n\n    let rawDoneReason: string | undefined;\n    let sawToolCall = false;\n    const usage: Usage = { input: 0, output: 0, total: 0 };\n\n    try {\n      for await (const chunk of stream) {\n        const content = chunk.message?.content;\n\n        if (content) {\n          yield { type: \"delta\", content };\n        }\n\n        for (const call of chunk.message?.tool_calls ?? []) {\n          sawToolCall = true;\n\n          yield {\n            type: \"tool-call\",\n            id: call.function.name,\n            name: call.function.name,\n            input: (call.function.arguments ?? {}) as Record<string, unknown>,\n          };\n        }\n\n        if (chunk.done_reason) {\n          rawDoneReason = chunk.done_reason;\n        }\n\n        if (chunk.done) {\n          usage.input = chunk.prompt_eval_count ?? usage.input;\n          usage.output = chunk.eval_count ?? usage.output;\n          usage.total = usage.input + usage.output;\n        }\n      }\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    const finishReason = sawToolCall ? \"tool_calls\" : mapDoneReason(rawDoneReason);\n\n    this.logger.debug(LOG_MODULE, \"response\", \"streaming chat call succeeded\", {\n      finishReason,\n      usage,\n    });\n\n    yield { type: \"done\", finishReason, usage };\n  }\n\n  /**\n   * Assemble the Ollama chat request shared by `complete()` and\n   * `stream()` (each adds its own `stream` literal so the client's\n   * overload resolves). Maps inference params into Ollama `options`\n   * and conditionally attaches tools + native structured output.\n   */\n  private buildRequest(\n    messages: Message[],\n    options: ModelCallOptions | undefined,\n  ): Omit<ChatRequest, \"stream\"> {\n    const temperature = options?.temperature ?? this.config.temperature;\n    const maxTokens = options?.maxTokens ?? this.config.maxTokens;\n\n    const ollamaOptions: Partial<Options> = {\n      ...(temperature !== undefined ? { temperature } : {}),\n      ...(maxTokens !== undefined ? { num_predict: maxTokens } : {}),\n    };\n\n    return {\n      model: this.name,\n      messages: toOllamaMessages(messages),\n      ...(Object.keys(ollamaOptions).length > 0 ? { options: ollamaOptions } : {}),\n      ...this.buildTools(options?.tools),\n      ...this.buildFormat(options?.responseSchema),\n    };\n  }\n\n  /**\n   * Spread-friendly tools fragment. Empty object when no tools were\n   * supplied so the caller can unconditionally spread it.\n   */\n  private buildTools(tools: ModelCallOptions[\"tools\"]): Pick<ChatRequest, \"tools\"> {\n    const mapped = toOllamaTools(tools);\n\n    return mapped ? { tools: mapped } : {};\n  }\n\n  /**\n   * Translate the neutral `responseSchema` into Ollama's native\n   * structured output (`format` accepts a JSON Schema object).\n   * Emitted only when the model is `structuredOutput`-capable and the\n   * schema is an object root — otherwise the agent's soft prompt hint\n   * + client-side `validate()` carry shape.\n   */\n  private buildFormat(\n    responseSchema: Record<string, unknown> | undefined,\n  ): Pick<ChatRequest, \"format\"> {\n    if (!responseSchema || !this.capabilities.structuredOutput) {\n      return {};\n    }\n\n    if (responseSchema.type !== \"object\" || typeof responseSchema.properties !== \"object\") {\n      return {};\n    }\n\n    return { format: responseSchema };\n  }\n\n  /**\n   * Reshape Ollama's `message.tool_calls` into the neutral\n   * `ModelToolCallRequest[]`. Ollama has no tool-call id, so the\n   * neutral `id` is synthesized from the tool name (see the class\n   * doc). Returns `undefined` when no tools were requested.\n   */\n  private extractToolCalls(\n    message: ChatResponse[\"message\"] | undefined,\n  ): ModelToolCallRequest[] | undefined {\n    const calls = message?.tool_calls;\n\n    if (!calls || calls.length === 0) {\n      return undefined;\n    }\n\n    return calls.map((call) => ({\n      id: call.function.name,\n      name: call.function.name,\n      input: (call.function.arguments ?? {}) as Record<string, unknown>,\n    }));\n  }\n\n  /**\n   * Normalize Ollama's eval counts into the neutral `Usage` shape.\n   * Ollama runs locally with no prompt cache, so there is no\n   * `cachedTokens`; `total` is computed from input + output.\n   */\n  private extractUsage(response: ChatResponse): Usage {\n    const input = response.prompt_eval_count ?? 0;\n    const output = response.eval_count ?? 0;\n\n    return { input, output, total: input + output };\n  }\n\n  /**\n   * Wrap a thrown provider error into the typed `AIError` hierarchy\n   * and emit the standard error log line before it propagates.\n   */\n  private logAndWrap(thrown: unknown) {\n    const wrapped = wrapOllamaError(thrown);\n\n    this.logger.error(LOG_MODULE, \"error\", wrapped.message, {\n      code: wrapped.code,\n      context: wrapped.context,\n    });\n\n    return wrapped;\n  }\n}\n","import { Ollama } from \"ollama\";\nimport type {\n  EmbedderContract,\n  ModelContract,\n  ModelPricing,\n  SDKAdapterContract,\n} from \"@warlock.js/ai\";\nimport { approximateTokenCount } from \"@warlock.js/ai\";\nimport type {\n  OllamaEmbedderConfig,\n  OllamaModelConfig,\n  OllamaSDKConfig,\n} from \"./config.type\";\nimport { OllamaEmbedder } from \"./embedder\";\nimport { OllamaModel } from \"./model\";\n\n/**\n * Ollama-backed implementation of `SDKAdapterContract`.\n *\n * **Role.** The package entry point for local / self-hosted models\n * served by an Ollama daemon via the official `ollama` client. One\n * `OllamaSDK` holds one live `Ollama` client, shared by every\n * `ModelContract` / `EmbedderContract` it produces.\n *\n * **Responsibility.**\n * - Owns: a long-lived `Ollama` client (host, headers) and its\n *   lifetime. Factory for `OllamaModel` / `OllamaEmbedder` instances\n *   sharing that client.\n * - Does NOT own: anything per-call — those live in `OllamaModel` /\n *   `OllamaEmbedder` and the agent runtime.\n *\n * Modeled as a class (see §4.2 of code-style.md — \"long-lived state\n * across many calls\"), fronted by FP usage like the other adapters.\n *\n * @example\n * const ollama = new OllamaSDK({});                  // local default host\n * const model = ollama.model({ name: \"llama3.1\", temperature: 0.7 });\n * const embedder = ollama.embedder({ name: \"nomic-embed-text\" });\n */\nexport class OllamaSDK implements SDKAdapterContract {\n  private readonly client: Ollama;\n  private readonly provider: string;\n  private readonly pricing?: Record<string, ModelPricing>;\n\n  public constructor(config: OllamaSDKConfig = {}) {\n    const { provider, pricing, ...clientConfig } = config;\n\n    this.client = new Ollama(clientConfig);\n    this.provider = provider ?? \"ollama\";\n    this.pricing = pricing;\n  }\n\n  /**\n   * Build an `OllamaModel` bound to this SDK's client. Each call\n   * returns a fresh instance; all instances share the underlying\n   * `Ollama` client. The SDK's `provider` label is forwarded.\n   *\n   * Pricing resolution: per-model `config.pricing` wins; otherwise the\n   * SDK-level registry entry keyed by `config.name`; otherwise\n   * `undefined` (local Ollama is free, so usually undefined).\n   */\n  public model(config: OllamaModelConfig): ModelContract {\n    const resolvedPricing = config.pricing ?? this.pricing?.[config.name];\n    const resolvedConfig: OllamaModelConfig =\n      resolvedPricing === config.pricing ? config : { ...config, pricing: resolvedPricing };\n\n    return new OllamaModel(this.client, resolvedConfig, this.provider);\n  }\n\n  /**\n   * Rough token-count estimate. Uses the character-heuristic\n   * (`approximateTokenCount`) from the core package — good enough for\n   * budgeting / context guards, not billing (and Ollama is free\n   * anyway). The optional model id is reserved for future per-model\n   * tokenizer dispatch; currently ignored.\n   */\n  public async count(text: string, _model?: string): Promise<number> {\n    return approximateTokenCount(text);\n  }\n\n  /**\n   * Build an `OllamaEmbedder` bound to this SDK's client.\n   *\n   * @example\n   * const embedder = ollama.embedder({ name: \"nomic-embed-text\" });\n   * const { vector } = await embedder.embed(\"Hello world\");\n   */\n  public embedder(config: OllamaEmbedderConfig): EmbedderContract {\n    return new OllamaEmbedder(this.client, config, this.provider);\n  }\n}\n"],"mappings":";;;;;;AAEA,MAAM,gBAA8C;CAClD,MAAM;CACN,QAAQ;AACV;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,cAAc,KAA8C;CAC1E,OAAO,cAAc,OAAO,OAAO;AACrC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACAA,SAAgB,iBAAiB,UAAsC;CACrE,OAAO,SAAS,KAAK,YAA2B;EAC9C,IAAI,QAAQ,SAAS,QACnB,OAAO;GACL,MAAM;GACN,SAAS,iBAAiB,QAAQ,OAAO;GACzC,WAAW,QAAQ,cAAc;EACnC;EAGF,IAAI,QAAQ,SAAS,eAAe,QAAQ,aAAa,QAAQ,UAAU,SAAS,GAClF,OAAO;GACL,MAAM;GACN,SAAS,iBAAiB,QAAQ,OAAO;GACzC,YAAY,QAAQ,UAAU,KAAK,cAAc,EAC/C,UAAU;IACR,MAAM,SAAS;IACf,WAAY,SAAS,SAAS,CAAC;GACjC,EACF,EAAE;EACJ;EAGF,IAAI,QAAQ,SAAS,UAAU,MAAM,QAAQ,QAAQ,OAAO,GAC1D,OAAO,mBAAmB,QAAQ,OAAO;EAG3C,OAAO;GAAE,MAAM,QAAQ;GAAM,SAAS,iBAAiB,QAAQ,OAAO;EAAE;CAC1E,CAAC;AACH;;;;;;;;;AAUA,SAAS,mBAAmB,OAAqC;CAC/D,MAAM,aAAuB,CAAC;CAC9B,MAAM,SAAmB,CAAC;CAE1B,KAAK,MAAM,QAAQ,OAAO;EACxB,IAAI,KAAK,SAAS,QAAQ;GACxB,WAAW,KAAK,KAAK,IAAI;GAEzB;EACF;EAEA,IAAI,SAAS,KAAK,QAChB,MAAM,IAAIA,mCACR,6EACF;EAGF,OAAO,KAAK,KAAK,OAAO,MAAM;CAChC;CAEA,OAAO;EACL,MAAM;EACN,SAAS,WAAW,KAAK,EAAE;EAC3B,GAAI,OAAO,SAAS,IAAI,EAAE,OAAO,IAAI,CAAC;CACxC;AACF;;;;;AAMA,SAAS,iBAAiB,SAAyC;CACjE,IAAI,OAAO,YAAY,UACrB,OAAO;CAGT,OAAO,QACJ,QAAQ,SAAiD,KAAK,SAAS,MAAM,EAC7E,KAAK,SAAS,KAAK,IAAI,EACvB,KAAK,EAAE;AACZ;;;;;;;;;;;;;;;;;AC3FA,SAAgB,cACd,OACoB;CACpB,IAAI,CAAC,SAAS,MAAM,WAAW,GAC7B;CAGF,OAAO,MAAM,KAAK,UAAU;EAC1B,MAAM;EACN,UAAU;GACR,MAAM,KAAK;GACX,aAAa,KAAK;GAClB,YAAY,aAAa,KAAK,KAAK;EACrC;CACF,EAAE;AACJ;;;;;;AAOA,SAAS,aAAa,OAA8E;CAClG,MAAM,+CAA2B,KAAK;CAEtC,IAAI,UAAU,OAAO,SAAS,UAC5B,OAAO;CAGT,OAAO,EAAE,MAAM,SAAS;AAC1B;;;;;;;;;;;;;;;;;;;;;;;;;ACAA,SAAgB,gBAAgB,QAA0B;CACxD,IAAI,kBAAkBC,wBACpB,OAAO;CAGT,MAAM,QAAQ,QAAQ,MAAM;CAC5B,MAAM,UAAU,aAAa,KAAK;CAClC,MAAM,UAAU,MAAM,YAAY,kBAAkB,QAAQ,OAAO,UAAU,OAAO,MAAM;CAE1F,IAAI,UAAU,KAAK,GACjB,OAAO,IAAIC,oCAAqB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGrE,IAAI,oBAAoB,OAAO,OAAO,GACpC,OAAO,IAAIC,6BAAc,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAG9D,IAAI,MAAM,eAAe,OAAO,MAAM,eAAe,KACnD,OAAO,IAAIC,iCAAkB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGlE,IAAI,MAAM,eAAe,KACvB,OAAO,IAAIC,sCAAuB,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;CAGvE,IAAI,eAAe,MAAM,UAAU,GAAG;EACpC,IAAI,mDAAmD,KAAK,OAAO,GACjE,OAAO,IAAIC,0CAA2B,SAAS;GAAE,OAAO;GAAQ;EAAQ,CAAC;EAG3E,OAAO,IAAIC,mCAAoB,SAAS;GAAE,OAAO;GAAQ;EAAQ,CAAC;CACpE;CAEA,OAAO,IAAIJ,6BAAc,SAAS;EAAE,OAAO;EAAQ;CAAQ,CAAC;AAC9D;;;;;;AAOA,SAAS,QAAQ,QAAmC;CAClD,IAAI,OAAO,WAAW,YAAY,WAAW,MAC3C,OAAO,CAAC;CAGV,MAAM,MAAM;CACZ,MAAM,QAAQ,IAAI;CAElB,OAAO;EACL,MAAM,OAAO,IAAI,SAAS,WAAW,IAAI,OAAO;EAChD,SAAS,OAAO,IAAI,YAAY,WAAW,IAAI,UAAU;EACzD,YAAY,OAAO,IAAI,gBAAgB,WAAW,IAAI,cAAc;EACpE,MACE,OAAO,IAAI,SAAS,WAChB,IAAI,OACJ,SAAS,OAAO,MAAM,SAAS,WAC7B,MAAM,OACN;CACV;AACF;;AAGA,SAAS,UAAU,OAAkC;CACnD,IAAI,MAAM,SAAS,gBAAgB,MAAM,SAAS,gBAChD,OAAO;CAGT,OAAO,MAAM,SAAS,eAAe,MAAM,SAAS;AACtD;;;;;;AAOA,SAAS,oBAAoB,OAAyB,SAA0B;CAC9E,OAAO,MAAM,SAAS,kBAAkB,6BAA6B,KAAK,OAAO;AACnF;;AAGA,SAAS,eAAe,QAAqC;CAC3D,OAAO,OAAO,WAAW,YAAY,UAAU,OAAO,SAAS;AACjE;;AAGA,SAAS,aAAa,OAAkD;CACtE,MAAM,UAAmC,CAAC;CAE1C,IAAI,MAAM,eAAe,QACvB,QAAQ,SAAS,MAAM;CAGzB,IAAI,MAAM,MACR,QAAQ,OAAO,MAAM;CAGvB,OAAO;AACT;;;;ACrIA,MAAMK,eAAa;;;;;;;;;;;;;;;;;;;;;;;;;AA0BnB,IAAa,iBAAb,MAAwD;CAStD,AAAO,YACL,QACA,QACA,WAAmB,UACnB;gBANgCC;EAOhC,KAAK,SAAS;EACd,KAAK,OAAO,OAAO;EACnB,KAAK,WAAW;EAChB,KAAK,uBAAuB,OAAO;EACnC,KAAK,aAAa,OAAO,cAAc;CACzC;CAEA,MAAa,MAAM,OAAyC;EAC1D,MAAM,EAAE,YAAY,UAAU,MAAM,KAAK,QAAQ,CAAC,KAAK,CAAC;EAExD,OAAO;GAAE,QAAQ,WAAW,MAAM,CAAC;GAAG,YAAY,KAAK;GAAY;EAAM;CAC3E;CAEA,MAAa,UAAU,QAAiD;EACtE,MAAM,EAAE,YAAY,UAAU,MAAM,KAAK,QAAQ,MAAM;EAEvD,OAAO;GAAE,SAAS;GAAY,YAAY,KAAK;GAAY;EAAM;CACnE;;;;;;CAOA,MAAc,QACZ,QAC4D;EAC5D,KAAK,OAAO,MAAMD,cAAY,oBAAoB,SAAS;GACzD,OAAO,KAAK;GACZ,OAAO,OAAO;EAChB,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,WAAW,MAAM,KAAK,OAAO,MAAM;IACjC,OAAO,KAAK;IACZ,OAAO;IACP,GAAI,KAAK,yBAAyB,SAC9B,EAAE,YAAY,KAAK,qBAAqB,IACxC,CAAC;GACP,CAAC;EACH,SAAS,QAAQ;GACf,MAAM,UAAU,gBAAgB,MAAM;GAEtC,KAAK,OAAO,MAAMA,cAAY,kBAAkB,QAAQ,SAAS;IAC/D,MAAM,QAAQ;IACd,SAAS,QAAQ;GACnB,CAAC;GAED,MAAM;EACR;EAEA,MAAM,aAAa,SAAS,cAAc,CAAC;EAE3C,IAAI,KAAK,eAAe,KAAK,WAAW,IACtC,KAAK,aAAa,WAAW,GAAG;EAGlC,MAAM,SAAS,SAAS,qBAAqB;EAC7C,MAAM,QAAwB;GAAE,cAAc;GAAQ,aAAa;EAAO;EAE1E,KAAK,OAAO,MAAMA,cAAY,qBAAqB,kBAAkB;GACnE,OAAO,WAAW;GAClB,YAAY,KAAK;EACnB,CAAC;EAED,OAAO;GAAE;GAAY;EAAM;CAC7B;AACF;;;;;;;;;;;;;;;AC7GA,MAAM,4BAA4B;CAChC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;;;;;;AAeA,SAAgB,sBAAsB,WAA4B;CAChE,MAAM,aAAa,UAAU,YAAY;CAEzC,OAAO,0BAA0B,MAAM,aAAa,WAAW,SAAS,QAAQ,CAAC;AACnF;;;;ACjBA,MAAM,aAAa;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsCnB,IAAa,cAAb,MAAkD;CAUhD,AAAO,YAAY,QAAgB,QAA2B,WAAmB,UAAU;gBAFzDE;EAGhC,KAAK,SAAS;EACd,KAAK,SAAS;EACd,KAAK,OAAO,OAAO;EACnB,KAAK,WAAW;EAChB,KAAK,UAAU,OAAO;EACtB,KAAK,eAAe;GAClB,kBAAkB,OAAO,oBAAoB;GAC7C,QAAQ,OAAO,UAAU,sBAAsB,OAAO,IAAI;EAC5D;CACF;;;;;;;CAQA,MAAa,SAAS,UAAqB,SAAoD;EAC7F,KAAK,OAAO,MAAM,YAAY,WAAW,sBAAsB;GAC7D,OAAO,KAAK;GACZ,cAAc,SAAS;GACvB,WAAW;GACX,WAAW,SAAS,OAAO,UAAU;EACvC,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,WAAW,MAAM,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAM,CAAC;EAC9F,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,MAAM,YAAY,KAAK,iBAAiB,SAAS,OAAO;EACxD,MAAM,eAAe,YAAY,eAAe,cAAc,SAAS,WAAW;EAClF,MAAM,QAAQ,KAAK,aAAa,QAAQ;EAExC,KAAK,OAAO,MAAM,YAAY,YAAY,uBAAuB;GAAE;GAAc;EAAM,CAAC;EAExF,OAAO;GACL,SAAS,SAAS,SAAS,WAAW;GACtC;GACA;GACA;EACF;CACF;;;;;;;;CASA,OAAc,OACZ,UACA,SACiC;EACjC,KAAK,OAAO,MAAM,YAAY,WAAW,gCAAgC;GACvE,OAAO,KAAK;GACZ,cAAc,SAAS;GACvB,WAAW;GACX,WAAW,SAAS,OAAO,UAAU;EACvC,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,SAAS,MAAM,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAK,CAAC;EAC3F,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,IAAI,SAAS,QACX,IAAI,QAAQ,OAAO,SACjB,OAAO,MAAM;OAEb,QAAQ,OAAO,iBAAiB,eAAe,OAAO,MAAM,GAAG,EAAE,MAAM,KAAK,CAAC;EAIjF,IAAI;EACJ,IAAI,cAAc;EAClB,MAAM,QAAe;GAAE,OAAO;GAAG,QAAQ;GAAG,OAAO;EAAE;EAErD,IAAI;GACF,WAAW,MAAM,SAAS,QAAQ;IAChC,MAAM,UAAU,MAAM,SAAS;IAE/B,IAAI,SACF,MAAM;KAAE,MAAM;KAAS;IAAQ;IAGjC,KAAK,MAAM,QAAQ,MAAM,SAAS,cAAc,CAAC,GAAG;KAClD,cAAc;KAEd,MAAM;MACJ,MAAM;MACN,IAAI,KAAK,SAAS;MAClB,MAAM,KAAK,SAAS;MACpB,OAAQ,KAAK,SAAS,aAAa,CAAC;KACtC;IACF;IAEA,IAAI,MAAM,aACR,gBAAgB,MAAM;IAGxB,IAAI,MAAM,MAAM;KACd,MAAM,QAAQ,MAAM,qBAAqB,MAAM;KAC/C,MAAM,SAAS,MAAM,cAAc,MAAM;KACzC,MAAM,QAAQ,MAAM,QAAQ,MAAM;IACpC;GACF;EACF,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,MAAM,eAAe,cAAc,eAAe,cAAc,aAAa;EAE7E,KAAK,OAAO,MAAM,YAAY,YAAY,iCAAiC;GACzE;GACA;EACF,CAAC;EAED,MAAM;GAAE,MAAM;GAAQ;GAAc;EAAM;CAC5C;;;;;;;CAQA,AAAQ,aACN,UACA,SAC6B;EAC7B,MAAM,cAAc,SAAS,eAAe,KAAK,OAAO;EACxD,MAAM,YAAY,SAAS,aAAa,KAAK,OAAO;EAEpD,MAAM,gBAAkC;GACtC,GAAI,gBAAgB,SAAY,EAAE,YAAY,IAAI,CAAC;GACnD,GAAI,cAAc,SAAY,EAAE,aAAa,UAAU,IAAI,CAAC;EAC9D;EAEA,OAAO;GACL,OAAO,KAAK;GACZ,UAAU,iBAAiB,QAAQ;GACnC,GAAI,OAAO,KAAK,aAAa,EAAE,SAAS,IAAI,EAAE,SAAS,cAAc,IAAI,CAAC;GAC1E,GAAG,KAAK,WAAW,SAAS,KAAK;GACjC,GAAG,KAAK,YAAY,SAAS,cAAc;EAC7C;CACF;;;;;CAMA,AAAQ,WAAW,OAA8D;EAC/E,MAAM,SAAS,cAAc,KAAK;EAElC,OAAO,SAAS,EAAE,OAAO,OAAO,IAAI,CAAC;CACvC;;;;;;;;CASA,AAAQ,YACN,gBAC6B;EAC7B,IAAI,CAAC,kBAAkB,CAAC,KAAK,aAAa,kBACxC,OAAO,CAAC;EAGV,IAAI,eAAe,SAAS,YAAY,OAAO,eAAe,eAAe,UAC3E,OAAO,CAAC;EAGV,OAAO,EAAE,QAAQ,eAAe;CAClC;;;;;;;CAQA,AAAQ,iBACN,SACoC;EACpC,MAAM,QAAQ,SAAS;EAEvB,IAAI,CAAC,SAAS,MAAM,WAAW,GAC7B;EAGF,OAAO,MAAM,KAAK,UAAU;GAC1B,IAAI,KAAK,SAAS;GAClB,MAAM,KAAK,SAAS;GACpB,OAAQ,KAAK,SAAS,aAAa,CAAC;EACtC,EAAE;CACJ;;;;;;CAOA,AAAQ,aAAa,UAA+B;EAClD,MAAM,QAAQ,SAAS,qBAAqB;EAC5C,MAAM,SAAS,SAAS,cAAc;EAEtC,OAAO;GAAE;GAAO;GAAQ,OAAO,QAAQ;EAAO;CAChD;;;;;CAMA,AAAQ,WAAW,QAAiB;EAClC,MAAM,UAAU,gBAAgB,MAAM;EAEtC,KAAK,OAAO,MAAM,YAAY,SAAS,QAAQ,SAAS;GACtD,MAAM,QAAQ;GACd,SAAS,QAAQ;EACnB,CAAC;EAED,OAAO;CACT;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;AC3QA,IAAa,YAAb,MAAqD;CAKnD,AAAO,YAAY,SAA0B,CAAC,GAAG;EAC/C,MAAM,EAAE,UAAU,SAAS,GAAG,iBAAiB;EAE/C,KAAK,SAAS,IAAIC,cAAO,YAAY;EACrC,KAAK,WAAW,YAAY;EAC5B,KAAK,UAAU;CACjB;;;;;;;;;;CAWA,AAAO,MAAM,QAA0C;EACrD,MAAM,kBAAkB,OAAO,WAAW,KAAK,UAAU,OAAO;EAChE,MAAM,iBACJ,oBAAoB,OAAO,UAAU,SAAS;GAAE,GAAG;GAAQ,SAAS;EAAgB;EAEtF,OAAO,IAAI,YAAY,KAAK,QAAQ,gBAAgB,KAAK,QAAQ;CACnE;;;;;;;;CASA,MAAa,MAAM,MAAc,QAAkC;EACjE,iDAA6B,IAAI;CACnC;;;;;;;;CASA,AAAO,SAAS,QAAgD;EAC9D,OAAO,IAAI,eAAe,KAAK,QAAQ,QAAQ,KAAK,QAAQ;CAC9D;AACF"}

package/esm/config.type.d.mts

@@ -0,0 +1,80 @@
+import { Config } from "ollama";
+import { EmbedderConfig, ModelConfig, ModelPricing } from "@warlock.js/ai";
+
+//#region ../../@warlock.js/ai-ollama/src/config.type.d.ts
+/**
+ * Configuration for the Ollama SDK adapter.
+ *
+ * Wraps the official `ollama` client `Config`. `host` is optional and
+ * defaults to the client's own default (`http://127.0.0.1:11434`) —
+ * point it at a remote/self-hosted Ollama server when needed. `headers`
+ * is handy for gateways that require auth in front of Ollama. The
+ * whole object is forwarded to `new Ollama(...)`.
+ *
+ * `provider` labels the SDK upstream — flows through to
+ * `ModelContract.provider`, `AgentReport.model`, logs, and
+ * provider-aware middleware. Defaults to `"ollama"`.
+ *
+ * `pricing` is an optional SDK-level registry keyed by model name.
+ * Local Ollama is free so this is usually unset; it exists for parity
+ * (hosted Ollama, internal chargeback). Resolution at `model()` call
+ * time: per-model `pricing` > this SDK registry > `undefined`.
+ *
+ * @example
+ * new OllamaSDK({});                                  // local default host
+ * new OllamaSDK({ host: "http://gpu-box.internal:11434" });
+ *
+ * @example
+ * new OllamaSDK({
+ *   host: "https://ollama.internal",
+ *   headers: { Authorization: `Bearer ${process.env.OLLAMA_TOKEN}` },
+ * });
+ */
+type OllamaSDKConfig = Partial<Config> & {
+  provider?: string;
+  /**
+   * Per-model USD pricing registry, keyed by model name. Surfaced onto
+   * every `OllamaModel` produced by `model()`; per-model
+   * `OllamaModelConfig.pricing` still wins when both are set.
+   */
+  pricing?: Record<string, ModelPricing>;
+};
+/**
+ * Per-model configuration for `OllamaSDK.model()`. `name` is the
+ * Ollama model tag (e.g. `"llama3.1"`, `"qwen2.5:14b"`,
+ * `"llama3.2-vision"`).
+ *
+ * @example
+ * ollama.model({ name: "llama3.1" });
+ * ollama.model({ name: "llama3.2-vision", vision: true });
+ */
+type OllamaModelConfig = ModelConfig & {
+  /**
+   * Override the auto-inferred vision capability. When omitted, the
+   * adapter checks the model tag against the known multimodal Ollama
+   * families (see `known-vision-models.ts`). Explicit `true`/`false`
+   * always wins over inference.
+   */
+  vision?: boolean;
+  /**
+   * Override the inferred `structuredOutput` capability. When omitted,
+   * the adapter treats the model as capable and forwards
+   * `responseSchema` via Ollama's native `format` JSON-schema field.
+   * Set `false` for models that handle it poorly — the agent then
+   * re-injects a soft schema hint into the system prompt instead.
+   */
+  structuredOutput?: boolean;
+};
+/**
+ * Per-embedder configuration for `OllamaSDK.embedder()`. `name` is the
+ * embeddings model tag (e.g. `"nomic-embed-text"`,
+ * `"mxbai-embed-large"`). `dimensions` is forwarded to Ollama's
+ * `dimensions` truncation field (supported by newer embedding models).
+ *
+ * @example
+ * ollama.embedder({ name: "nomic-embed-text" });
+ */
+type OllamaEmbedderConfig = EmbedderConfig;
+//#endregion
+export { OllamaEmbedderConfig, OllamaModelConfig, OllamaSDKConfig };
+//# sourceMappingURL=config.type.d.mts.map

package/esm/config.type.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"config.type.d.mts","names":[],"sources":["../../../../../@warlock.js/ai-ollama/src/config.type.ts"],"mappings":";;;;;;AA+BA;;;;;;;;;;;;;;;;AAOuC;AAYvC;;;;;;;;AAekB;KAlCN,eAAA,GAAkB,OAAA,CAAQ,MAAA;EACpC,QAAA;;;AA6C+C;;;EAvC/C,OAAA,GAAU,MAAA,SAAe,YAAA;AAAA;;;;;;;;;;KAYf,iBAAA,GAAoB,WAAW;;;;;;;EAOzC,MAAA;;;;;;;;EAQA,gBAAA;AAAA;;;;;;;;;;KAYU,oBAAA,GAAuB,cAAc"}

package/esm/embedder.mjs

@@ -0,0 +1,101 @@
+import { wrapOllamaError } from "./utils/wrap-ollama-error.mjs";
+import "./utils/index.mjs";
+import { log } from "@warlock.js/logger";
+
+//#region ../../@warlock.js/ai-ollama/src/embedder.ts
+const LOG_MODULE = "ai.ollama";
+/**
+* Ollama-backed implementation of `EmbedderContract`
+* (`nomic-embed-text`, `mxbai-embed-large`, …) via `client.embed`.
+*
+* **Role.** Converts text into floating-point vectors. Standalone
+* primitive — unrelated to chat / tools / the agent loop.
+*
+* **Batch is native.** Ollama's `embed` accepts a string array and
+* returns `embeddings` in input order, so `embedMany` is a single
+* request (like the Gemini adapter, unlike Bedrock/Titan).
+*
+* **Usage.** Ollama returns only `prompt_eval_count` (no separate
+* total); it is reported as both `promptTokens` and `totalTokens`.
+*
+* **Dimensions.** When no `dimensions` override is given,
+* `this.dimensions` starts at `0` and is populated from the first
+* response's vector length, then cached. Passing `dimensions`
+* forwards Ollama's truncation field and sets the initial value.
+*
+* @example
+* const embedder = new OllamaEmbedder(client, { name: "nomic-embed-text" });
+* const { vector } = await embedder.embed("Hello world");
+* const { vectors } = await embedder.embedMany(["doc 1", "doc 2"]);
+*/
+var OllamaEmbedder = class {
+	constructor(client, config, provider = "ollama") {
+		this.logger = log;
+		this.client = client;
+		this.name = config.name;
+		this.provider = provider;
+		this.configuredDimensions = config.dimensions;
+		this.dimensions = config.dimensions ?? 0;
+	}
+	async embed(input) {
+		const { embeddings, usage } = await this.request([input]);
+		return {
+			vector: embeddings[0] ?? [],
+			dimensions: this.dimensions,
+			usage
+		};
+	}
+	async embedMany(inputs) {
+		const { embeddings, usage } = await this.request(inputs);
+		return {
+			vectors: embeddings,
+			dimensions: this.dimensions,
+			usage
+		};
+	}
+	/**
+	* Shared transport: one `embed` call for the whole batch, wrap
+	* provider errors, cache `dimensions` from the first vector, and
+	* return vectors in input order plus a neutral usage object.
+	*/
+	async request(inputs) {
+		this.logger.debug(LOG_MODULE, "embedder.request", "embed", {
+			model: this.name,
+			count: inputs.length
+		});
+		let response;
+		try {
+			response = await this.client.embed({
+				model: this.name,
+				input: inputs,
+				...this.configuredDimensions !== void 0 ? { dimensions: this.configuredDimensions } : {}
+			});
+		} catch (thrown) {
+			const wrapped = wrapOllamaError(thrown);
+			this.logger.error(LOG_MODULE, "embedder.error", wrapped.message, {
+				code: wrapped.code,
+				context: wrapped.context
+			});
+			throw wrapped;
+		}
+		const embeddings = response.embeddings ?? [];
+		if (this.dimensions === 0 && embeddings[0]) this.dimensions = embeddings[0].length;
+		const tokens = response.prompt_eval_count ?? 0;
+		const usage = {
+			promptTokens: tokens,
+			totalTokens: tokens
+		};
+		this.logger.debug(LOG_MODULE, "embedder.response", "embed returned", {
+			count: embeddings.length,
+			dimensions: this.dimensions
+		});
+		return {
+			embeddings,
+			usage
+		};
+	}
+};
+
+//#endregion
+export { OllamaEmbedder };
+//# sourceMappingURL=embedder.mjs.map

package/esm/embedder.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"embedder.mjs","names":[],"sources":["../../../../../@warlock.js/ai-ollama/src/embedder.ts"],"sourcesContent":["import {\n  type EmbeddingBatchResult,\n  type EmbeddingResult,\n  type EmbeddingUsage,\n  type EmbedderContract,\n} from \"@warlock.js/ai\";\nimport { log, type Logger } from \"@warlock.js/logger\";\nimport type { EmbedResponse, Ollama } from \"ollama\";\nimport type { OllamaEmbedderConfig } from \"./config.type\";\nimport { wrapOllamaError } from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/**\n * Ollama-backed implementation of `EmbedderContract`\n * (`nomic-embed-text`, `mxbai-embed-large`, …) via `client.embed`.\n *\n * **Role.** Converts text into floating-point vectors. Standalone\n * primitive — unrelated to chat / tools / the agent loop.\n *\n * **Batch is native.** Ollama's `embed` accepts a string array and\n * returns `embeddings` in input order, so `embedMany` is a single\n * request (like the Gemini adapter, unlike Bedrock/Titan).\n *\n * **Usage.** Ollama returns only `prompt_eval_count` (no separate\n * total); it is reported as both `promptTokens` and `totalTokens`.\n *\n * **Dimensions.** When no `dimensions` override is given,\n * `this.dimensions` starts at `0` and is populated from the first\n * response's vector length, then cached. Passing `dimensions`\n * forwards Ollama's truncation field and sets the initial value.\n *\n * @example\n * const embedder = new OllamaEmbedder(client, { name: \"nomic-embed-text\" });\n * const { vector } = await embedder.embed(\"Hello world\");\n * const { vectors } = await embedder.embedMany([\"doc 1\", \"doc 2\"]);\n */\nexport class OllamaEmbedder implements EmbedderContract {\n  public readonly name: string;\n  public readonly provider: string;\n  public dimensions: number;\n\n  private readonly client: Ollama;\n  private readonly configuredDimensions: number | undefined;\n  private readonly logger: Logger = log;\n\n  public constructor(\n    client: Ollama,\n    config: OllamaEmbedderConfig,\n    provider: string = \"ollama\",\n  ) {\n    this.client = client;\n    this.name = config.name;\n    this.provider = provider;\n    this.configuredDimensions = config.dimensions;\n    this.dimensions = config.dimensions ?? 0;\n  }\n\n  public async embed(input: string): Promise<EmbeddingResult> {\n    const { embeddings, usage } = await this.request([input]);\n\n    return { vector: embeddings[0] ?? [], dimensions: this.dimensions, usage };\n  }\n\n  public async embedMany(inputs: string[]): Promise<EmbeddingBatchResult> {\n    const { embeddings, usage } = await this.request(inputs);\n\n    return { vectors: embeddings, dimensions: this.dimensions, usage };\n  }\n\n  /**\n   * Shared transport: one `embed` call for the whole batch, wrap\n   * provider errors, cache `dimensions` from the first vector, and\n   * return vectors in input order plus a neutral usage object.\n   */\n  private async request(\n    inputs: string[],\n  ): Promise<{ embeddings: number[][]; usage: EmbeddingUsage }> {\n    this.logger.debug(LOG_MODULE, \"embedder.request\", \"embed\", {\n      model: this.name,\n      count: inputs.length,\n    });\n\n    let response: EmbedResponse;\n\n    try {\n      response = await this.client.embed({\n        model: this.name,\n        input: inputs,\n        ...(this.configuredDimensions !== undefined\n          ? { dimensions: this.configuredDimensions }\n          : {}),\n      });\n    } catch (thrown) {\n      const wrapped = wrapOllamaError(thrown);\n\n      this.logger.error(LOG_MODULE, \"embedder.error\", wrapped.message, {\n        code: wrapped.code,\n        context: wrapped.context,\n      });\n\n      throw wrapped;\n    }\n\n    const embeddings = response.embeddings ?? [];\n\n    if (this.dimensions === 0 && embeddings[0]) {\n      this.dimensions = embeddings[0].length;\n    }\n\n    const tokens = response.prompt_eval_count ?? 0;\n    const usage: EmbeddingUsage = { promptTokens: tokens, totalTokens: tokens };\n\n    this.logger.debug(LOG_MODULE, \"embedder.response\", \"embed returned\", {\n      count: embeddings.length,\n      dimensions: this.dimensions,\n    });\n\n    return { embeddings, usage };\n  }\n}\n"],"mappings":";;;;;AAWA,MAAM,aAAa;;;;;;;;;;;;;;;;;;;;;;;;;AA0BnB,IAAa,iBAAb,MAAwD;CAStD,AAAO,YACL,QACA,QACA,WAAmB,UACnB;gBANgC;EAOhC,KAAK,SAAS;EACd,KAAK,OAAO,OAAO;EACnB,KAAK,WAAW;EAChB,KAAK,uBAAuB,OAAO;EACnC,KAAK,aAAa,OAAO,cAAc;CACzC;CAEA,MAAa,MAAM,OAAyC;EAC1D,MAAM,EAAE,YAAY,UAAU,MAAM,KAAK,QAAQ,CAAC,KAAK,CAAC;EAExD,OAAO;GAAE,QAAQ,WAAW,MAAM,CAAC;GAAG,YAAY,KAAK;GAAY;EAAM;CAC3E;CAEA,MAAa,UAAU,QAAiD;EACtE,MAAM,EAAE,YAAY,UAAU,MAAM,KAAK,QAAQ,MAAM;EAEvD,OAAO;GAAE,SAAS;GAAY,YAAY,KAAK;GAAY;EAAM;CACnE;;;;;;CAOA,MAAc,QACZ,QAC4D;EAC5D,KAAK,OAAO,MAAM,YAAY,oBAAoB,SAAS;GACzD,OAAO,KAAK;GACZ,OAAO,OAAO;EAChB,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,WAAW,MAAM,KAAK,OAAO,MAAM;IACjC,OAAO,KAAK;IACZ,OAAO;IACP,GAAI,KAAK,yBAAyB,SAC9B,EAAE,YAAY,KAAK,qBAAqB,IACxC,CAAC;GACP,CAAC;EACH,SAAS,QAAQ;GACf,MAAM,UAAU,gBAAgB,MAAM;GAEtC,KAAK,OAAO,MAAM,YAAY,kBAAkB,QAAQ,SAAS;IAC/D,MAAM,QAAQ;IACd,SAAS,QAAQ;GACnB,CAAC;GAED,MAAM;EACR;EAEA,MAAM,aAAa,SAAS,cAAc,CAAC;EAE3C,IAAI,KAAK,eAAe,KAAK,WAAW,IACtC,KAAK,aAAa,WAAW,GAAG;EAGlC,MAAM,SAAS,SAAS,qBAAqB;EAC7C,MAAM,QAAwB;GAAE,cAAc;GAAQ,aAAa;EAAO;EAE1E,KAAK,OAAO,MAAM,YAAY,qBAAqB,kBAAkB;GACnE,OAAO,WAAW;GAClB,YAAY,KAAK;EACnB,CAAC;EAED,OAAO;GAAE;GAAY;EAAM;CAC7B;AACF"}

package/esm/index.d.mts

@@ -0,0 +1,3 @@
+import { OllamaEmbedderConfig, OllamaModelConfig, OllamaSDKConfig } from "./config.type.mjs";
+import { OllamaSDK } from "./sdk.mjs";
+export { type OllamaEmbedderConfig, type OllamaModelConfig, OllamaSDK, type OllamaSDKConfig };

package/esm/index.mjs

@@ -0,0 +1,3 @@
+import { OllamaSDK } from "./sdk.mjs";
+
+export { OllamaSDK };

package/esm/known-vision-models.mjs

@@ -0,0 +1,44 @@
+//#region ../../@warlock.js/ai-ollama/src/known-vision-models.ts
+/**
+* Substrings identifying Ollama model tags whose family accepts image
+* input (vision).
+*
+* Ollama tags are family-named with optional size/quant suffixes
+* (`llama3.2-vision:11b`, `llava:13b-v1.6`, `qwen2.5-vl:7b`). A
+* substring match tolerates those suffixes. Covers the common
+* multimodal families on the Ollama registry; text-only models
+* (`llama3.1`, `mistral`, `phi3`, `nomic-embed-text`) are excluded.
+* Override per-model via `ollama.model({ name, vision: true | false })`.
+*/
+const VISION_CAPABLE_SUBSTRINGS = [
+	"llava",
+	"vision",
+	"bakllava",
+	"moondream",
+	"minicpm-v",
+	"qwen2-vl",
+	"qwen2.5-vl",
+	"llama4",
+	"gemma3"
+];
+/**
+* Infer whether an Ollama model tag supports vision based on the known
+* multimodal-family substrings. Unknown tags default to `false` so
+* passing an image to a text-only local model surfaces a clear,
+* agent-side capability error instead of the image being silently
+* ignored by the model.
+*
+* @example
+* inferVisionCapability("llama3.2-vision:11b"); // → true
+* inferVisionCapability("llava:13b");           // → true
+* inferVisionCapability("llama3.1");            // → false
+* inferVisionCapability("nomic-embed-text");    // → false
+*/
+function inferVisionCapability(modelName) {
+	const normalized = modelName.toLowerCase();
+	return VISION_CAPABLE_SUBSTRINGS.some((fragment) => normalized.includes(fragment));
+}
+
+//#endregion
+export { inferVisionCapability };
+//# sourceMappingURL=known-vision-models.mjs.map

package/esm/known-vision-models.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"known-vision-models.mjs","names":[],"sources":["../../../../../@warlock.js/ai-ollama/src/known-vision-models.ts"],"sourcesContent":["/**\n * Substrings identifying Ollama model tags whose family accepts image\n * input (vision).\n *\n * Ollama tags are family-named with optional size/quant suffixes\n * (`llama3.2-vision:11b`, `llava:13b-v1.6`, `qwen2.5-vl:7b`). A\n * substring match tolerates those suffixes. Covers the common\n * multimodal families on the Ollama registry; text-only models\n * (`llama3.1`, `mistral`, `phi3`, `nomic-embed-text`) are excluded.\n * Override per-model via `ollama.model({ name, vision: true | false })`.\n */\nconst VISION_CAPABLE_SUBSTRINGS = [\n  \"llava\",\n  \"vision\",\n  \"bakllava\",\n  \"moondream\",\n  \"minicpm-v\",\n  \"qwen2-vl\",\n  \"qwen2.5-vl\",\n  \"llama4\",\n  \"gemma3\",\n];\n\n/**\n * Infer whether an Ollama model tag supports vision based on the known\n * multimodal-family substrings. Unknown tags default to `false` so\n * passing an image to a text-only local model surfaces a clear,\n * agent-side capability error instead of the image being silently\n * ignored by the model.\n *\n * @example\n * inferVisionCapability(\"llama3.2-vision:11b\"); // → true\n * inferVisionCapability(\"llava:13b\");           // → true\n * inferVisionCapability(\"llama3.1\");            // → false\n * inferVisionCapability(\"nomic-embed-text\");    // → false\n */\nexport function inferVisionCapability(modelName: string): boolean {\n  const normalized = modelName.toLowerCase();\n\n  return VISION_CAPABLE_SUBSTRINGS.some((fragment) => normalized.includes(fragment));\n}\n"],"mappings":";;;;;;;;;;;;AAWA,MAAM,4BAA4B;CAChC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;;;;;;AAeA,SAAgB,sBAAsB,WAA4B;CAChE,MAAM,aAAa,UAAU,YAAY;CAEzC,OAAO,0BAA0B,MAAM,aAAa,WAAW,SAAS,QAAQ,CAAC;AACnF"}

package/esm/model.mjs

@@ -0,0 +1,251 @@
+import { mapDoneReason } from "./utils/map-done-reason.mjs";
+import { toOllamaMessages } from "./utils/to-ollama-messages.mjs";
+import { toOllamaTools } from "./utils/to-ollama-tools.mjs";
+import { wrapOllamaError } from "./utils/wrap-ollama-error.mjs";
+import "./utils/index.mjs";
+import { inferVisionCapability } from "./known-vision-models.mjs";
+import { log } from "@warlock.js/logger";
+
+//#region ../../@warlock.js/ai-ollama/src/model.ts
+const LOG_MODULE = "ai.ollama";
+/**
+* Ollama-backed implementation of `ModelContract`.
+*
+* **Role.** The provider-facing bridge between the vendor-neutral
+* `@warlock.js/ai` agent runtime and a local (or self-hosted) Ollama
+* server via the official `ollama` client.
+*
+* **Responsibility.**
+* - Owns: a long-lived `Ollama` client + frozen `ModelConfig` (model
+*   tag, temperature, maxTokens) used as per-call defaults.
+* - Owns: translating vendor-neutral `Message[]` / `ToolConfig[]` into
+*   Ollama's chat shapes (system stays a real role, `tool_calls` /
+*   `tool_name`, base64 `images`) and Ollama's response (content, tool
+*   calls, done reason, eval-count usage) back into neutral shapes.
+* - Does NOT own: tool dispatch, looping, history, retries — agent
+*   concerns. The model is a per-call protocol adapter.
+*
+* **Tool-call ids.** Ollama has no tool-call id concept — a `tool_call`
+* is `{ function: { name, arguments } }`. The adapter synthesizes the
+* neutral `id` from the tool name so the agent's tool-result round-trip
+* (which keys on `toolCallId`) maps back to Ollama's name-based
+* matching. Parallel calls to the *same* tool in one turn therefore
+* share an id — a documented v1 limitation inherent to Ollama's wire
+* format, not this adapter.
+*
+* Modeled as a class (see §4.2 of code-style.md — "long-lived state
+* across calls").
+*
+* @example
+* import { Ollama } from "ollama";
+* const client = new Ollama({ host: "http://127.0.0.1:11434" });
+* const model = new OllamaModel(client, { name: "llama3.1" });
+*
+* const myAgent = agent({ model, tools: [searchTool] });
+* const result = await myAgent.execute("Summarize today's news.");
+*/
+var OllamaModel = class {
+	constructor(client, config, provider = "ollama") {
+		this.logger = log;
+		this.client = client;
+		this.config = config;
+		this.name = config.name;
+		this.provider = provider;
+		this.pricing = config.pricing;
+		this.capabilities = {
+			structuredOutput: config.structuredOutput ?? true,
+			vision: config.vision ?? inferVisionCapability(config.name)
+		};
+	}
+	/**
+	* Single-shot completion. Sends the full message list to
+	* `client.chat`, waits for the terminal response, and reshapes it
+	* into a vendor-neutral `ModelResponse`. Per-call `options` override
+	* the instance defaults for this call only.
+	*/
+	async complete(messages, options) {
+		this.logger.debug(LOG_MODULE, "request", "Starting chat call", {
+			model: this.name,
+			messageCount: messages.length,
+			streaming: false,
+			toolCount: options?.tools?.length ?? 0
+		});
+		let response;
+		try {
+			response = await this.client.chat({
+				...this.buildRequest(messages, options),
+				stream: false
+			});
+		} catch (thrown) {
+			throw this.logAndWrap(thrown);
+		}
+		const toolCalls = this.extractToolCalls(response.message);
+		const finishReason = toolCalls ? "tool_calls" : mapDoneReason(response.done_reason);
+		const usage = this.extractUsage(response);
+		this.logger.debug(LOG_MODULE, "response", "chat call succeeded", {
+			finishReason,
+			usage
+		});
+		return {
+			content: response.message?.content ?? "",
+			finishReason,
+			usage,
+			toolCalls
+		};
+	}
+	/**
+	* Incremental streaming completion. Yields neutral
+	* `ModelStreamChunk`s — `delta` for content, `tool-call` per
+	* function call (Ollama streams a fully-formed call, not partial
+	* JSON), and a terminal `done` with the final finish reason + usage.
+	* Honors `options.signal` by aborting the underlying stream.
+	*/
+	async *stream(messages, options) {
+		this.logger.debug(LOG_MODULE, "request", "Starting streaming chat call", {
+			model: this.name,
+			messageCount: messages.length,
+			streaming: true,
+			toolCount: options?.tools?.length ?? 0
+		});
+		let stream;
+		try {
+			stream = await this.client.chat({
+				...this.buildRequest(messages, options),
+				stream: true
+			});
+		} catch (thrown) {
+			throw this.logAndWrap(thrown);
+		}
+		if (options?.signal) if (options.signal.aborted) stream.abort();
+		else options.signal.addEventListener("abort", () => stream.abort(), { once: true });
+		let rawDoneReason;
+		let sawToolCall = false;
+		const usage = {
+			input: 0,
+			output: 0,
+			total: 0
+		};
+		try {
+			for await (const chunk of stream) {
+				const content = chunk.message?.content;
+				if (content) yield {
+					type: "delta",
+					content
+				};
+				for (const call of chunk.message?.tool_calls ?? []) {
+					sawToolCall = true;
+					yield {
+						type: "tool-call",
+						id: call.function.name,
+						name: call.function.name,
+						input: call.function.arguments ?? {}
+					};
+				}
+				if (chunk.done_reason) rawDoneReason = chunk.done_reason;
+				if (chunk.done) {
+					usage.input = chunk.prompt_eval_count ?? usage.input;
+					usage.output = chunk.eval_count ?? usage.output;
+					usage.total = usage.input + usage.output;
+				}
+			}
+		} catch (thrown) {
+			throw this.logAndWrap(thrown);
+		}
+		const finishReason = sawToolCall ? "tool_calls" : mapDoneReason(rawDoneReason);
+		this.logger.debug(LOG_MODULE, "response", "streaming chat call succeeded", {
+			finishReason,
+			usage
+		});
+		yield {
+			type: "done",
+			finishReason,
+			usage
+		};
+	}
+	/**
+	* Assemble the Ollama chat request shared by `complete()` and
+	* `stream()` (each adds its own `stream` literal so the client's
+	* overload resolves). Maps inference params into Ollama `options`
+	* and conditionally attaches tools + native structured output.
+	*/
+	buildRequest(messages, options) {
+		const temperature = options?.temperature ?? this.config.temperature;
+		const maxTokens = options?.maxTokens ?? this.config.maxTokens;
+		const ollamaOptions = {
+			...temperature !== void 0 ? { temperature } : {},
+			...maxTokens !== void 0 ? { num_predict: maxTokens } : {}
+		};
+		return {
+			model: this.name,
+			messages: toOllamaMessages(messages),
+			...Object.keys(ollamaOptions).length > 0 ? { options: ollamaOptions } : {},
+			...this.buildTools(options?.tools),
+			...this.buildFormat(options?.responseSchema)
+		};
+	}
+	/**
+	* Spread-friendly tools fragment. Empty object when no tools were
+	* supplied so the caller can unconditionally spread it.
+	*/
+	buildTools(tools) {
+		const mapped = toOllamaTools(tools);
+		return mapped ? { tools: mapped } : {};
+	}
+	/**
+	* Translate the neutral `responseSchema` into Ollama's native
+	* structured output (`format` accepts a JSON Schema object).
+	* Emitted only when the model is `structuredOutput`-capable and the
+	* schema is an object root — otherwise the agent's soft prompt hint
+	* + client-side `validate()` carry shape.
+	*/
+	buildFormat(responseSchema) {
+		if (!responseSchema || !this.capabilities.structuredOutput) return {};
+		if (responseSchema.type !== "object" || typeof responseSchema.properties !== "object") return {};
+		return { format: responseSchema };
+	}
+	/**
+	* Reshape Ollama's `message.tool_calls` into the neutral
+	* `ModelToolCallRequest[]`. Ollama has no tool-call id, so the
+	* neutral `id` is synthesized from the tool name (see the class
+	* doc). Returns `undefined` when no tools were requested.
+	*/
+	extractToolCalls(message) {
+		const calls = message?.tool_calls;
+		if (!calls || calls.length === 0) return;
+		return calls.map((call) => ({
+			id: call.function.name,
+			name: call.function.name,
+			input: call.function.arguments ?? {}
+		}));
+	}
+	/**
+	* Normalize Ollama's eval counts into the neutral `Usage` shape.
+	* Ollama runs locally with no prompt cache, so there is no
+	* `cachedTokens`; `total` is computed from input + output.
+	*/
+	extractUsage(response) {
+		const input = response.prompt_eval_count ?? 0;
+		const output = response.eval_count ?? 0;
+		return {
+			input,
+			output,
+			total: input + output
+		};
+	}
+	/**
+	* Wrap a thrown provider error into the typed `AIError` hierarchy
+	* and emit the standard error log line before it propagates.
+	*/
+	logAndWrap(thrown) {
+		const wrapped = wrapOllamaError(thrown);
+		this.logger.error(LOG_MODULE, "error", wrapped.message, {
+			code: wrapped.code,
+			context: wrapped.context
+		});
+		return wrapped;
+	}
+};
+
+//#endregion
+export { OllamaModel };
+//# sourceMappingURL=model.mjs.map

package/esm/model.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"model.mjs","names":[],"sources":["../../../../../@warlock.js/ai-ollama/src/model.ts"],"sourcesContent":["import {\n  type Message,\n  type ModelCallOptions,\n  type ModelCapabilities,\n  type ModelContract,\n  type ModelPricing,\n  type ModelResponse,\n  type ModelStreamChunk,\n  type ModelToolCallRequest,\n  type Usage,\n} from \"@warlock.js/ai\";\nimport { log, type Logger } from \"@warlock.js/logger\";\nimport type {\n  AbortableAsyncIterator,\n  ChatRequest,\n  ChatResponse,\n  Ollama,\n  Options,\n} from \"ollama\";\nimport type { OllamaModelConfig } from \"./config.type\";\nimport { inferVisionCapability } from \"./known-vision-models\";\nimport { mapDoneReason, toOllamaMessages, toOllamaTools, wrapOllamaError } from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/**\n * Ollama-backed implementation of `ModelContract`.\n *\n * **Role.** The provider-facing bridge between the vendor-neutral\n * `@warlock.js/ai` agent runtime and a local (or self-hosted) Ollama\n * server via the official `ollama` client.\n *\n * **Responsibility.**\n * - Owns: a long-lived `Ollama` client + frozen `ModelConfig` (model\n *   tag, temperature, maxTokens) used as per-call defaults.\n * - Owns: translating vendor-neutral `Message[]` / `ToolConfig[]` into\n *   Ollama's chat shapes (system stays a real role, `tool_calls` /\n *   `tool_name`, base64 `images`) and Ollama's response (content, tool\n *   calls, done reason, eval-count usage) back into neutral shapes.\n * - Does NOT own: tool dispatch, looping, history, retries — agent\n *   concerns. The model is a per-call protocol adapter.\n *\n * **Tool-call ids.** Ollama has no tool-call id concept — a `tool_call`\n * is `{ function: { name, arguments } }`. The adapter synthesizes the\n * neutral `id` from the tool name so the agent's tool-result round-trip\n * (which keys on `toolCallId`) maps back to Ollama's name-based\n * matching. Parallel calls to the *same* tool in one turn therefore\n * share an id — a documented v1 limitation inherent to Ollama's wire\n * format, not this adapter.\n *\n * Modeled as a class (see §4.2 of code-style.md — \"long-lived state\n * across calls\").\n *\n * @example\n * import { Ollama } from \"ollama\";\n * const client = new Ollama({ host: \"http://127.0.0.1:11434\" });\n * const model = new OllamaModel(client, { name: \"llama3.1\" });\n *\n * const myAgent = agent({ model, tools: [searchTool] });\n * const result = await myAgent.execute(\"Summarize today's news.\");\n */\nexport class OllamaModel implements ModelContract {\n  public readonly name: string;\n  public readonly provider: string;\n  public readonly capabilities: ModelCapabilities;\n  public readonly pricing?: ModelPricing;\n\n  private readonly client: Ollama;\n  private readonly config: OllamaModelConfig;\n  private readonly logger: Logger = log;\n\n  public constructor(client: Ollama, config: OllamaModelConfig, provider: string = \"ollama\") {\n    this.client = client;\n    this.config = config;\n    this.name = config.name;\n    this.provider = provider;\n    this.pricing = config.pricing;\n    this.capabilities = {\n      structuredOutput: config.structuredOutput ?? true,\n      vision: config.vision ?? inferVisionCapability(config.name),\n    };\n  }\n\n  /**\n   * Single-shot completion. Sends the full message list to\n   * `client.chat`, waits for the terminal response, and reshapes it\n   * into a vendor-neutral `ModelResponse`. Per-call `options` override\n   * the instance defaults for this call only.\n   */\n  public async complete(messages: Message[], options?: ModelCallOptions): Promise<ModelResponse> {\n    this.logger.debug(LOG_MODULE, \"request\", \"Starting chat call\", {\n      model: this.name,\n      messageCount: messages.length,\n      streaming: false,\n      toolCount: options?.tools?.length ?? 0,\n    });\n\n    let response: ChatResponse;\n\n    try {\n      response = await this.client.chat({ ...this.buildRequest(messages, options), stream: false });\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    const toolCalls = this.extractToolCalls(response.message);\n    const finishReason = toolCalls ? \"tool_calls\" : mapDoneReason(response.done_reason);\n    const usage = this.extractUsage(response);\n\n    this.logger.debug(LOG_MODULE, \"response\", \"chat call succeeded\", { finishReason, usage });\n\n    return {\n      content: response.message?.content ?? \"\",\n      finishReason,\n      usage,\n      toolCalls,\n    };\n  }\n\n  /**\n   * Incremental streaming completion. Yields neutral\n   * `ModelStreamChunk`s — `delta` for content, `tool-call` per\n   * function call (Ollama streams a fully-formed call, not partial\n   * JSON), and a terminal `done` with the final finish reason + usage.\n   * Honors `options.signal` by aborting the underlying stream.\n   */\n  public async *stream(\n    messages: Message[],\n    options?: ModelCallOptions,\n  ): AsyncIterable<ModelStreamChunk> {\n    this.logger.debug(LOG_MODULE, \"request\", \"Starting streaming chat call\", {\n      model: this.name,\n      messageCount: messages.length,\n      streaming: true,\n      toolCount: options?.tools?.length ?? 0,\n    });\n\n    let stream: AbortableAsyncIterator<ChatResponse>;\n\n    try {\n      stream = await this.client.chat({ ...this.buildRequest(messages, options), stream: true });\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    if (options?.signal) {\n      if (options.signal.aborted) {\n        stream.abort();\n      } else {\n        options.signal.addEventListener(\"abort\", () => stream.abort(), { once: true });\n      }\n    }\n\n    let rawDoneReason: string | undefined;\n    let sawToolCall = false;\n    const usage: Usage = { input: 0, output: 0, total: 0 };\n\n    try {\n      for await (const chunk of stream) {\n        const content = chunk.message?.content;\n\n        if (content) {\n          yield { type: \"delta\", content };\n        }\n\n        for (const call of chunk.message?.tool_calls ?? []) {\n          sawToolCall = true;\n\n          yield {\n            type: \"tool-call\",\n            id: call.function.name,\n            name: call.function.name,\n            input: (call.function.arguments ?? {}) as Record<string, unknown>,\n          };\n        }\n\n        if (chunk.done_reason) {\n          rawDoneReason = chunk.done_reason;\n        }\n\n        if (chunk.done) {\n          usage.input = chunk.prompt_eval_count ?? usage.input;\n          usage.output = chunk.eval_count ?? usage.output;\n          usage.total = usage.input + usage.output;\n        }\n      }\n    } catch (thrown) {\n      throw this.logAndWrap(thrown);\n    }\n\n    const finishReason = sawToolCall ? \"tool_calls\" : mapDoneReason(rawDoneReason);\n\n    this.logger.debug(LOG_MODULE, \"response\", \"streaming chat call succeeded\", {\n      finishReason,\n      usage,\n    });\n\n    yield { type: \"done\", finishReason, usage };\n  }\n\n  /**\n   * Assemble the Ollama chat request shared by `complete()` and\n   * `stream()` (each adds its own `stream` literal so the client's\n   * overload resolves). Maps inference params into Ollama `options`\n   * and conditionally attaches tools + native structured output.\n   */\n  private buildRequest(\n    messages: Message[],\n    options: ModelCallOptions | undefined,\n  ): Omit<ChatRequest, \"stream\"> {\n    const temperature = options?.temperature ?? this.config.temperature;\n    const maxTokens = options?.maxTokens ?? this.config.maxTokens;\n\n    const ollamaOptions: Partial<Options> = {\n      ...(temperature !== undefined ? { temperature } : {}),\n      ...(maxTokens !== undefined ? { num_predict: maxTokens } : {}),\n    };\n\n    return {\n      model: this.name,\n      messages: toOllamaMessages(messages),\n      ...(Object.keys(ollamaOptions).length > 0 ? { options: ollamaOptions } : {}),\n      ...this.buildTools(options?.tools),\n      ...this.buildFormat(options?.responseSchema),\n    };\n  }\n\n  /**\n   * Spread-friendly tools fragment. Empty object when no tools were\n   * supplied so the caller can unconditionally spread it.\n   */\n  private buildTools(tools: ModelCallOptions[\"tools\"]): Pick<ChatRequest, \"tools\"> {\n    const mapped = toOllamaTools(tools);\n\n    return mapped ? { tools: mapped } : {};\n  }\n\n  /**\n   * Translate the neutral `responseSchema` into Ollama's native\n   * structured output (`format` accepts a JSON Schema object).\n   * Emitted only when the model is `structuredOutput`-capable and the\n   * schema is an object root — otherwise the agent's soft prompt hint\n   * + client-side `validate()` carry shape.\n   */\n  private buildFormat(\n    responseSchema: Record<string, unknown> | undefined,\n  ): Pick<ChatRequest, \"format\"> {\n    if (!responseSchema || !this.capabilities.structuredOutput) {\n      return {};\n    }\n\n    if (responseSchema.type !== \"object\" || typeof responseSchema.properties !== \"object\") {\n      return {};\n    }\n\n    return { format: responseSchema };\n  }\n\n  /**\n   * Reshape Ollama's `message.tool_calls` into the neutral\n   * `ModelToolCallRequest[]`. Ollama has no tool-call id, so the\n   * neutral `id` is synthesized from the tool name (see the class\n   * doc). Returns `undefined` when no tools were requested.\n   */\n  private extractToolCalls(\n    message: ChatResponse[\"message\"] | undefined,\n  ): ModelToolCallRequest[] | undefined {\n    const calls = message?.tool_calls;\n\n    if (!calls || calls.length === 0) {\n      return undefined;\n    }\n\n    return calls.map((call) => ({\n      id: call.function.name,\n      name: call.function.name,\n      input: (call.function.arguments ?? {}) as Record<string, unknown>,\n    }));\n  }\n\n  /**\n   * Normalize Ollama's eval counts into the neutral `Usage` shape.\n   * Ollama runs locally with no prompt cache, so there is no\n   * `cachedTokens`; `total` is computed from input + output.\n   */\n  private extractUsage(response: ChatResponse): Usage {\n    const input = response.prompt_eval_count ?? 0;\n    const output = response.eval_count ?? 0;\n\n    return { input, output, total: input + output };\n  }\n\n  /**\n   * Wrap a thrown provider error into the typed `AIError` hierarchy\n   * and emit the standard error log line before it propagates.\n   */\n  private logAndWrap(thrown: unknown) {\n    const wrapped = wrapOllamaError(thrown);\n\n    this.logger.error(LOG_MODULE, \"error\", wrapped.message, {\n      code: wrapped.code,\n      context: wrapped.context,\n    });\n\n    return wrapped;\n  }\n}\n"],"mappings":";;;;;;;;;AAuBA,MAAM,aAAa;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsCnB,IAAa,cAAb,MAAkD;CAUhD,AAAO,YAAY,QAAgB,QAA2B,WAAmB,UAAU;gBAFzD;EAGhC,KAAK,SAAS;EACd,KAAK,SAAS;EACd,KAAK,OAAO,OAAO;EACnB,KAAK,WAAW;EAChB,KAAK,UAAU,OAAO;EACtB,KAAK,eAAe;GAClB,kBAAkB,OAAO,oBAAoB;GAC7C,QAAQ,OAAO,UAAU,sBAAsB,OAAO,IAAI;EAC5D;CACF;;;;;;;CAQA,MAAa,SAAS,UAAqB,SAAoD;EAC7F,KAAK,OAAO,MAAM,YAAY,WAAW,sBAAsB;GAC7D,OAAO,KAAK;GACZ,cAAc,SAAS;GACvB,WAAW;GACX,WAAW,SAAS,OAAO,UAAU;EACvC,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,WAAW,MAAM,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAM,CAAC;EAC9F,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,MAAM,YAAY,KAAK,iBAAiB,SAAS,OAAO;EACxD,MAAM,eAAe,YAAY,eAAe,cAAc,SAAS,WAAW;EAClF,MAAM,QAAQ,KAAK,aAAa,QAAQ;EAExC,KAAK,OAAO,MAAM,YAAY,YAAY,uBAAuB;GAAE;GAAc;EAAM,CAAC;EAExF,OAAO;GACL,SAAS,SAAS,SAAS,WAAW;GACtC;GACA;GACA;EACF;CACF;;;;;;;;CASA,OAAc,OACZ,UACA,SACiC;EACjC,KAAK,OAAO,MAAM,YAAY,WAAW,gCAAgC;GACvE,OAAO,KAAK;GACZ,cAAc,SAAS;GACvB,WAAW;GACX,WAAW,SAAS,OAAO,UAAU;EACvC,CAAC;EAED,IAAI;EAEJ,IAAI;GACF,SAAS,MAAM,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAK,CAAC;EAC3F,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,IAAI,SAAS,QACX,IAAI,QAAQ,OAAO,SACjB,OAAO,MAAM;OAEb,QAAQ,OAAO,iBAAiB,eAAe,OAAO,MAAM,GAAG,EAAE,MAAM,KAAK,CAAC;EAIjF,IAAI;EACJ,IAAI,cAAc;EAClB,MAAM,QAAe;GAAE,OAAO;GAAG,QAAQ;GAAG,OAAO;EAAE;EAErD,IAAI;GACF,WAAW,MAAM,SAAS,QAAQ;IAChC,MAAM,UAAU,MAAM,SAAS;IAE/B,IAAI,SACF,MAAM;KAAE,MAAM;KAAS;IAAQ;IAGjC,KAAK,MAAM,QAAQ,MAAM,SAAS,cAAc,CAAC,GAAG;KAClD,cAAc;KAEd,MAAM;MACJ,MAAM;MACN,IAAI,KAAK,SAAS;MAClB,MAAM,KAAK,SAAS;MACpB,OAAQ,KAAK,SAAS,aAAa,CAAC;KACtC;IACF;IAEA,IAAI,MAAM,aACR,gBAAgB,MAAM;IAGxB,IAAI,MAAM,MAAM;KACd,MAAM,QAAQ,MAAM,qBAAqB,MAAM;KAC/C,MAAM,SAAS,MAAM,cAAc,MAAM;KACzC,MAAM,QAAQ,MAAM,QAAQ,MAAM;IACpC;GACF;EACF,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,MAAM,eAAe,cAAc,eAAe,cAAc,aAAa;EAE7E,KAAK,OAAO,MAAM,YAAY,YAAY,iCAAiC;GACzE;GACA;EACF,CAAC;EAED,MAAM;GAAE,MAAM;GAAQ;GAAc;EAAM;CAC5C;;;;;;;CAQA,AAAQ,aACN,UACA,SAC6B;EAC7B,MAAM,cAAc,SAAS,eAAe,KAAK,OAAO;EACxD,MAAM,YAAY,SAAS,aAAa,KAAK,OAAO;EAEpD,MAAM,gBAAkC;GACtC,GAAI,gBAAgB,SAAY,EAAE,YAAY,IAAI,CAAC;GACnD,GAAI,cAAc,SAAY,EAAE,aAAa,UAAU,IAAI,CAAC;EAC9D;EAEA,OAAO;GACL,OAAO,KAAK;GACZ,UAAU,iBAAiB,QAAQ;GACnC,GAAI,OAAO,KAAK,aAAa,EAAE,SAAS,IAAI,EAAE,SAAS,cAAc,IAAI,CAAC;GAC1E,GAAG,KAAK,WAAW,SAAS,KAAK;GACjC,GAAG,KAAK,YAAY,SAAS,cAAc;EAC7C;CACF;;;;;CAMA,AAAQ,WAAW,OAA8D;EAC/E,MAAM,SAAS,cAAc,KAAK;EAElC,OAAO,SAAS,EAAE,OAAO,OAAO,IAAI,CAAC;CACvC;;;;;;;;CASA,AAAQ,YACN,gBAC6B;EAC7B,IAAI,CAAC,kBAAkB,CAAC,KAAK,aAAa,kBACxC,OAAO,CAAC;EAGV,IAAI,eAAe,SAAS,YAAY,OAAO,eAAe,eAAe,UAC3E,OAAO,CAAC;EAGV,OAAO,EAAE,QAAQ,eAAe;CAClC;;;;;;;CAQA,AAAQ,iBACN,SACoC;EACpC,MAAM,QAAQ,SAAS;EAEvB,IAAI,CAAC,SAAS,MAAM,WAAW,GAC7B;EAGF,OAAO,MAAM,KAAK,UAAU;GAC1B,IAAI,KAAK,SAAS;GAClB,MAAM,KAAK,SAAS;GACpB,OAAQ,KAAK,SAAS,aAAa,CAAC;EACtC,EAAE;CACJ;;;;;;CAOA,AAAQ,aAAa,UAA+B;EAClD,MAAM,QAAQ,SAAS,qBAAqB;EAC5C,MAAM,SAAS,SAAS,cAAc;EAEtC,OAAO;GAAE;GAAO;GAAQ,OAAO,QAAQ;EAAO;CAChD;;;;;CAMA,AAAQ,WAAW,QAAiB;EAClC,MAAM,UAAU,gBAAgB,MAAM;EAEtC,KAAK,OAAO,MAAM,YAAY,SAAS,QAAQ,SAAS;GACtD,MAAM,QAAQ;GACd,SAAS,QAAQ;EACnB,CAAC;EAED,OAAO;CACT;AACF"}