@warlock.js/ai-ollama 5.30.0 → 5.32.0

package/CHANGELOG.md

@@ -4,6 +4,25 @@ All notable changes to `@warlock.js/ai-ollama` are documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). `@warlock.js/*` packages are released in lockstep — every package shares the same version number, so a version below may list only the changes that affected this package.
 
+## 5.32.0 - 2026-10-18
+
+### Changed
+
+- Pin the package development compiler to TypeScript 7.0.2 and use explicit Node types and source roots with supported module resolution.
+
+## 5.31.0 - 2026-10-18
+
+### Fixed
+
+- `complete()` now honours `signal`: the `ollama` client takes no per-request signal, so `OllamaSDK` wraps the client's `fetch` (the caller's own `fetch` option included) to apply each call's signal. Aborting cancels the HTTP request before or after the response starts, on `complete()` and `stream()`, and a call with an already-aborted signal rejects without sending a request. An adapter contract test covers both calls.
+- Provider body options merge plain-object fields one level deep, preserving neutral reasoning, generation and structured-output settings unless the caller overrides their keys.
+
+### Added
+
+- `complete()` and the streaming `done` chunk attach `usage.cost` when the model is given `pricing`.
+- With `streamChunks`, `stream()` yields `reasoning` for each `message.thinking` piece. Provider-defined tools for other providers are left out of the request with a warning.
+- `providerOptions.ollama.body` forwards native per-call request options for completions and streams while protecting the model, conversation and tool definitions.
+
 ## 5.30.0 - 2026-10-16
 
 ### Changed
@@ -81,6 +100,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 ### Changed
 
 - Lockstep release maintenance and dependency refresh.
+
 ## 5.20.0 - 2026-09-24
 
 ### Changed

package/cjs/index.cjs

@@ -2,7 +2,44 @@ Object.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });
 let ollama = require("ollama");
 let _warlock_js_ai = require("@warlock.js/ai");
 let _warlock_js_logger = require("@warlock.js/logger");
+let node_async_hooks = require("node:async_hooks");
 
+//#region ../../packages/ai-ollama/src/utils/call-signal.ts
+/**
+* Per-call cancellation for the `ollama` client.
+*
+* The `ollama` client (0.6) takes no per-request `AbortSignal`: a
+* non-streaming `chat()` cannot be cancelled at all, and a streamed one only
+* through the iterator it returns once the response headers have arrived.
+* To honour `ModelCallOptions.signal` the SDK builds its client with
+* {@link signalAwareFetch}, and the model runs each request inside
+* {@link withCallSignal}; the wrapped `fetch` reads the call's signal from
+* async context and hands it to the real `fetch`, so aborting cancels the
+* HTTP request at any point — before the headers, or mid-body.
+*/
+const callSignal = new node_async_hooks.AsyncLocalStorage();
+/** Run `operation` with `signal` as the cancellation signal for any `fetch` it makes through {@link signalAwareFetch}. */
+function withCallSignal(signal, operation) {
+	return callSignal.run(signal, operation);
+}
+/**
+* Wrap a `fetch` so every request also listens to the signal of the call it
+* belongs to (see {@link withCallSignal}), combined with any signal the
+* client set itself.
+*/
+function signalAwareFetch(baseFetch) {
+	return (input, init) => {
+		const signal = callSignal.getStore();
+		if (!signal) return baseFetch(input, init);
+		const own = init?.signal ?? void 0;
+		return baseFetch(input, {
+			...init,
+			signal: own ? AbortSignal.any([own, signal]) : signal
+		});
+	};
+}
+
+//#endregion
 //#region ../../packages/ai-ollama/src/utils/map-done-reason.ts
 const doneReasonMap = {
 	stop: "stop",
@@ -346,6 +383,38 @@ var OllamaEmbedder = class {
 	}
 };
 
+//#endregion
+//#region ../../packages/ai-ollama/src/provider-options.ts
+const RESERVED_FIELDS = /* @__PURE__ */ new Set([
+	"model",
+	"messages",
+	"tools",
+	"stream"
+]);
+function mergeProviderBody(built, options) {
+	const body = (options?.providerOptions?.ollama)?.body ?? {};
+	const reserved = Object.keys(body).filter((key) => RESERVED_FIELDS.has(key));
+	if (reserved.length > 0) throw new _warlock_js_ai.InvalidRequestError(`providerOptions.ollama.body cannot set ${reserved.join(", ")}: use the call's messages, tools and model options.`, { context: { fields: reserved } });
+	const defaults = built;
+	const overrides = Object.fromEntries(Object.entries(body).map(([key, value]) => {
+		const previous = defaults[key];
+		return [key, isPlainObject(previous) && isPlainObject(value) ? {
+			...previous,
+			...value
+		} : value];
+	}));
+	return {
+		...built,
+		...overrides
+	};
+}
+/** Merge only plain objects; arrays, null and provider scalar values replace the field. */
+function isPlainObject(value) {
+	if (value === null || typeof value !== "object") return false;
+	const prototype = Object.getPrototypeOf(value);
+	return prototype === Object.prototype || prototype === null;
+}
+
 //#endregion
 //#region ../../packages/ai-ollama/src/known-reasoning-models.ts
 /**
@@ -436,6 +505,8 @@ function inferVisionCapability(modelName) {
 //#endregion
 //#region ../../packages/ai-ollama/src/model.ts
 const LOG_MODULE = "ai.ollama";
+/** `provider` value of `ai.providerTool()` entries this adapter sends verbatim. */
+const OLLAMA_WIRE_FORMAT = "ollama";
 /** Ollama supports only low, medium, and high named thinking levels. */
 const EFFORT_TO_THINK_LEVEL = {
 	minimal: "low",
@@ -502,7 +573,8 @@ var OllamaModel = class {
 	* 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.
+	* the instance defaults for this call only. Aborting `options.signal`
+	* cancels the HTTP request through the SDK client's signal-aware `fetch`.
 	*/
 	async complete(messages, options) {
 		this.logger.debug(LOG_MODULE, "request", "Starting chat call", {
@@ -513,16 +585,17 @@ var OllamaModel = class {
 		});
 		let response;
 		try {
-			response = await this.client.chat({
+			options?.signal?.throwIfAborted();
+			response = await withCallSignal(options?.signal, () => 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);
+		const usage = (0, _warlock_js_ai.attachCost)(this.extractUsage(response), this.pricing);
 		this.logger.debug(LOG_MODULE, "response", "chat call succeeded", {
 			finishReason,
 			usage
@@ -539,7 +612,12 @@ var OllamaModel = class {
 	* `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.
+	* Honors `options.signal`: aborting cancels the HTTP request (through
+	* the SDK client's signal-aware `fetch`) and the stream iterator.
+	*
+	* With `options.streamChunks` it also yields `reasoning` for each
+	* `message.thinking` piece. Usage arrives only on the final chunk, so no
+	* `usage` chunk; calls arrive whole, so no `tool-args`.
 	*/
 	async *stream(messages, options) {
 		this.logger.debug(LOG_MODULE, "request", "Starting streaming chat call", {
@@ -550,15 +628,18 @@ var OllamaModel = class {
 		});
 		let stream;
 		try {
-			stream = await this.client.chat({
+			options?.signal?.throwIfAborted();
+			stream = await withCallSignal(options?.signal, () => 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 });
+		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 = {
@@ -566,8 +647,14 @@ var OllamaModel = class {
 			output: 0,
 			total: 0
 		};
+		const want = (0, _warlock_js_ai.resolveStreamChunks)(options);
 		try {
 			for await (const chunk of stream) {
+				const thinking = chunk.message?.thinking;
+				if (want.reasoning && thinking) yield {
+					type: "reasoning",
+					delta: thinking
+				};
 				const content = chunk.message?.content;
 				if (content) yield {
 					type: "delta",
@@ -593,6 +680,7 @@ var OllamaModel = class {
 			throw this.logAndWrap(thrown);
 		}
 		const finishReason = sawToolCall ? "tool_calls" : mapDoneReason(rawDoneReason);
+		(0, _warlock_js_ai.attachCost)(usage, this.pricing);
 		this.logger.debug(LOG_MODULE, "response", "streaming chat call succeeded", {
 			finishReason,
 			usage
@@ -616,14 +704,14 @@ var OllamaModel = class {
 			...temperature !== void 0 ? { temperature } : {},
 			...maxTokens !== void 0 ? { num_predict: maxTokens } : {}
 		};
-		return {
+		return mergeProviderBody({
 			model: this.name,
 			messages: toOllamaMessages(messages),
 			...Object.keys(ollamaOptions).length > 0 ? { options: ollamaOptions } : {},
-			...this.buildTools(options?.tools),
+			...this.buildTools(options),
 			...this.buildFormat(options?.responseSchema),
 			...this.buildThink(options?.reasoning)
-		};
+		}, options);
 	}
 	/**
 	* Translate the neutral `reasoning` hint into Ollama's `think`
@@ -653,9 +741,15 @@ var OllamaModel = class {
 	* 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 } : {};
+	buildTools(options) {
+		const { tools, definitions } = (0, _warlock_js_ai.selectProviderTools)(options, {
+			accepts: [OLLAMA_WIRE_FORMAT, this.provider],
+			module: LOG_MODULE,
+			model: this.name,
+			logger: this.logger
+		});
+		const mapped = [...toOllamaTools(tools) ?? [], ...definitions];
+		return mapped.length > 0 ? { tools: mapped } : {};
 	}
 	/**
 	* Translate the neutral `responseSchema` into Ollama's native
@@ -749,7 +843,10 @@ var OllamaModel = class {
 var OllamaSDK = class {
 	constructor(config = {}) {
 		const { provider, pricing, ...clientConfig } = config;
-		this.client = new ollama.Ollama(clientConfig);
+		this.client = new ollama.Ollama({
+			...clientConfig,
+			fetch: signalAwareFetch(clientConfig.fetch ?? globalThis.fetch)
+		});
 		this.provider = provider ?? "ollama";
 		this.pricing = pricing;
 	}

package/cjs/index.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.cjs","names":["InvalidRequestError","AIError","ProviderTimeoutError","ProviderError","ProviderAuthError","ProviderRateLimitError","ContextLengthExceededError","InvalidRequestError","LOG_MODULE","log","log","Ollama"],"sources":["../../../../../../../packages/ai-ollama/src/utils/map-done-reason.ts","../../../../../../../packages/ai-ollama/src/utils/to-ollama-messages.ts","../../../../../../../packages/ai-ollama/src/utils/to-ollama-tools.ts","../../../../../../../packages/ai-ollama/src/utils/wrap-ollama-error.ts","../../../../../../../packages/ai-ollama/src/embedder.ts","../../../../../../../packages/ai-ollama/src/known-reasoning-models.ts","../../../../../../../packages/ai-ollama/src/known-vision-models.ts","../../../../../../../packages/ai-ollama/src/model.ts","../../../../../../../packages/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 emits a\n * reasoning / \"thinking\" channel before the visible answer.\n *\n * Ollama exposes thinking via the request-side `think` flag and the\n * response-side `message.thinking` string. Only models trained for it\n * honor `think`; sending it to a non-reasoning model is a no-op at best.\n * Tags are family-named with optional size/quant suffixes\n * (`deepseek-r1:7b`, `qwq:32b-preview`), so a substring match tolerates\n * the suffixes. Covers the common reasoning families on the Ollama\n * registry; plain instruct models (`llama3.1`, `mistral`, `phi3`) are\n * excluded. Override per-model via\n * `ollama.model({ name, reasoning: true | false })`.\n */\nconst REASONING_CAPABLE_SUBSTRINGS = [\n  \"deepseek-r1\",\n  \"qwq\",\n  \"qwen3\",\n  \"magistral\",\n  \"phi4-reasoning\",\n  \"phi4-mini-reasoning\",\n  \"cogito\",\n  \"smallthinker\",\n  \"exaone-deep\",\n  \"gpt-oss\",\n];\n\n/**\n * Infer whether an Ollama model tag supports a reasoning / thinking\n * channel based on the known thinking-family substrings. Unknown tags\n * default to `false` so the adapter never sends the `think` flag to a\n * model that cannot honor it (a no-op for plain instruct models).\n *\n * @example\n * inferReasoningCapability(\"deepseek-r1:7b\");  // → true\n * inferReasoningCapability(\"qwq:32b\");          // → true\n * inferReasoningCapability(\"llama3.1\");         // → false\n * inferReasoningCapability(\"nomic-embed-text\"); // → false\n */\nexport function inferReasoningCapability(modelName: string): boolean {\n  const normalized = modelName.toLowerCase();\n\n  return REASONING_CAPABLE_SUBSTRINGS.some((fragment) => normalized.includes(fragment));\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 ReasoningEffort,\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 { inferReasoningCapability } from \"./known-reasoning-models\";\nimport { inferVisionCapability } from \"./known-vision-models\";\nimport { mapDoneReason, toOllamaMessages, toOllamaTools, wrapOllamaError } from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/** Ollama supports only low, medium, and high named thinking levels. */\nconst EFFORT_TO_THINK_LEVEL: Record<\n  Exclude<ReasoningEffort, \"none\">,\n  \"low\" | \"medium\" | \"high\"\n> = {\n  minimal: \"low\",\n  low: \"low\",\n  medium: \"medium\",\n  high: \"high\",\n  xhigh: \"high\",\n  max: \"high\",\n};\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      // Thinking-capable families (deepseek-r1, qwq, qwen3, …) honor the\n      // `think` request flag; plain instruct models do not. Explicit\n      // config wins over the family-substring inference.\n      reasoning: config.reasoning ?? inferReasoningCapability(config.name),\n      // Ollama has no provider-side prompt cache and the chat API takes\n      // no audio / PDF content parts — report these truthfully as false\n      // so the agent rejects unsupported attachments / cache hints\n      // upfront instead of silently dropping them at the wire.\n      promptCaching: false,\n      audio: false,\n      pdf: false,\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      ...this.buildThink(options?.reasoning),\n    };\n  }\n\n  /**\n   * Translate the neutral `reasoning` hint into Ollama's `think`\n   * request flag. Ollama's `think` accepts `boolean | 'low' | 'medium'\n   * | 'high'`, so neutral effort levels are clamped to the nearest named\n   * level; an effort-less `reasoning` (only `maxTokens`, or an empty\n   * object) becomes `think: true` to switch the channel on.\n   * The neutral `effort: \"none\"` (\"run without reasoning\") maps to\n   * `think: false` — Ollama has no `\"none\"` level, and this is its\n   * native reasoning-off switch.\n   *\n   * No-ops unless the model is reasoning-capable, so the `think` flag is\n   * never sent to a plain instruct model that cannot honor it.\n   *\n   * `reasoning.maxTokens` (the thinking-budget hint) has no Ollama\n   * equivalent — the daemon does not accept a thinking-token cap — so it\n   * is honored only as the on/off signal above and otherwise ignored.\n   * `ModelCallOptions.cacheControl` is likewise a no-op: Ollama has no\n   * provider prompt cache, so there is no cache breakpoint to place.\n   */\n  private buildThink(\n    reasoning: ModelCallOptions[\"reasoning\"],\n  ): Pick<ChatRequest, \"think\"> {\n    if (!reasoning || !this.capabilities.reasoning) {\n      return {};\n    }\n\n    // `effort: \"none\"` = explicit reasoning-off → switch the channel off.\n    // Ollama's `think` has no `\"none\"` level, so map it to `false`.\n    if (reasoning.effort === \"none\") {\n      return { think: false };\n    }\n\n    return { think: reasoning.effort ? EFFORT_TO_THINK_LEVEL[reasoning.effort] : true };\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   *\n   * Cost-truth: Ollama reports only `prompt_eval_count` (input) and\n   * `eval_count` (output). It has **no** provider prompt cache, so\n   * `Usage.cachedTokens` / `Usage.cacheWriteTokens` stay undefined\n   * (honest absence, not a false zero). Reasoning models emit their\n   * thinking as the `message.thinking` *string* but the wire format\n   * carries **no separate reasoning-token count** — the thinking tokens\n   * are already folded into `eval_count`. We therefore do not fabricate\n   * a `Usage.reasoningTokens` from the text length; it is left undefined\n   * unless/until the daemon exposes a real count. `total` is input +\n   * 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,CAAC,CAC9E,KAAK,SAAS,KAAK,IAAI,CAAC,CACxB,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,EAAE,CAAC;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;;;;;;;;;;;;;;;;;;AC1GA,MAAM,+BAA+B;CACnC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;;;;;AAcA,SAAgB,yBAAyB,WAA4B;CACnE,MAAM,aAAa,UAAU,YAAY;CAEzC,OAAO,6BAA6B,MAAM,aAAa,WAAW,SAAS,QAAQ,CAAC;AACtF;;;;;;;;;;;;;;;AChCA,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;;;;ACfA,MAAM,aAAa;;AAGnB,MAAM,wBAGF;CACF,SAAS;CACT,KAAK;CACL,QAAQ;CACR,MAAM;CACN,OAAO;CACP,KAAK;AACP;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsCA,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;GAI1D,WAAW,OAAO,aAAa,yBAAyB,OAAO,IAAI;GAKnE,eAAe;GACf,OAAO;GACP,KAAK;EACP;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,CAAC,CAAC,SAAS,IAAI,EAAE,SAAS,cAAc,IAAI,CAAC;GAC1E,GAAG,KAAK,WAAW,SAAS,KAAK;GACjC,GAAG,KAAK,YAAY,SAAS,cAAc;GAC3C,GAAG,KAAK,WAAW,SAAS,SAAS;EACvC;CACF;;;;;;;;;;;;;;;;;;;;CAqBA,AAAQ,WACN,WAC4B;EAC5B,IAAI,CAAC,aAAa,CAAC,KAAK,aAAa,WACnC,OAAO,CAAC;EAKV,IAAI,UAAU,WAAW,QACvB,OAAO,EAAE,OAAO,MAAM;EAGxB,OAAO,EAAE,OAAO,UAAU,SAAS,sBAAsB,UAAU,UAAU,KAAK;CACpF;;;;;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;;;;;;;;;;;;;;;CAgBA,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;;;;;;;;;;;;;;;;;;;;;;;;;;;AClVA,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"}
+{"version":3,"file":"index.cjs","names":["AsyncLocalStorage","InvalidRequestError","extractJsonSchema","AIError","ProviderTimeoutError","ProviderError","ProviderAuthError","ProviderRateLimitError","ContextLengthExceededError","InvalidRequestError","LOG_MODULE","log","InvalidRequestError","log","attachCost","resolveStreamChunks","selectProviderTools","Ollama","approximateTokenCount"],"sources":["../../../../../../../packages/ai-ollama/src/utils/call-signal.ts","../../../../../../../packages/ai-ollama/src/utils/map-done-reason.ts","../../../../../../../packages/ai-ollama/src/utils/to-ollama-messages.ts","../../../../../../../packages/ai-ollama/src/utils/to-ollama-tools.ts","../../../../../../../packages/ai-ollama/src/utils/wrap-ollama-error.ts","../../../../../../../packages/ai-ollama/src/embedder.ts","../../../../../../../packages/ai-ollama/src/provider-options.ts","../../../../../../../packages/ai-ollama/src/known-reasoning-models.ts","../../../../../../../packages/ai-ollama/src/known-vision-models.ts","../../../../../../../packages/ai-ollama/src/model.ts","../../../../../../../packages/ai-ollama/src/sdk.ts"],"sourcesContent":["import { AsyncLocalStorage } from \"node:async_hooks\";\n\n/**\n * Per-call cancellation for the `ollama` client.\n *\n * The `ollama` client (0.6) takes no per-request `AbortSignal`: a\n * non-streaming `chat()` cannot be cancelled at all, and a streamed one only\n * through the iterator it returns once the response headers have arrived.\n * To honour `ModelCallOptions.signal` the SDK builds its client with\n * {@link signalAwareFetch}, and the model runs each request inside\n * {@link withCallSignal}; the wrapped `fetch` reads the call's signal from\n * async context and hands it to the real `fetch`, so aborting cancels the\n * HTTP request at any point — before the headers, or mid-body.\n */\nconst callSignal = new AsyncLocalStorage<AbortSignal | undefined>();\n\n/** Run `operation` with `signal` as the cancellation signal for any `fetch` it makes through {@link signalAwareFetch}. */\nexport function withCallSignal<T>(signal: AbortSignal | undefined, operation: () => T): T {\n  return callSignal.run(signal, operation);\n}\n\n/**\n * Wrap a `fetch` so every request also listens to the signal of the call it\n * belongs to (see {@link withCallSignal}), combined with any signal the\n * client set itself.\n */\nexport function signalAwareFetch(baseFetch: typeof fetch): typeof fetch {\n  return (input, init) => {\n    const signal = callSignal.getStore();\n\n    if (!signal) {\n      return baseFetch(input, init);\n    }\n\n    const own = init?.signal ?? undefined;\n\n    return baseFetch(input, {\n      ...init,\n      signal: own ? AbortSignal.any([own, signal]) : signal,\n    });\n  };\n}\n","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(\n  input: ToolConfig<unknown, unknown>[\"input\"],\n): 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(client: Ollama, config: OllamaEmbedderConfig, provider: string = \"ollama\") {\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","import { InvalidRequestError, type ModelCallOptions } from \"@warlock.js/ai\";\n\n/** Adapter-only fields merged after the neutral request parameters. */\nexport type OllamaCallOptions = { body?: Record<string, unknown> };\n\nconst RESERVED_FIELDS = new Set([\"model\", \"messages\", \"tools\", \"stream\"]);\n\nexport function mergeProviderBody<T extends object>(\n  built: T,\n  options: ModelCallOptions | undefined,\n): T {\n  const provider = options?.providerOptions?.ollama as OllamaCallOptions | undefined;\n  const body = provider?.body ?? {};\n  const reserved = Object.keys(body).filter((key) => RESERVED_FIELDS.has(key));\n  if (reserved.length > 0) {\n    throw new InvalidRequestError(\n      `providerOptions.ollama.body cannot set ${reserved.join(\", \")}: use the call's messages, tools and model options.`,\n      { context: { fields: reserved } },\n    );\n  }\n  const defaults = built as Record<string, unknown>;\n  const overrides = Object.fromEntries(\n    Object.entries(body).map(([key, value]) => {\n      const previous = defaults[key];\n      return [\n        key,\n        isPlainObject(previous) && isPlainObject(value) ? { ...previous, ...value } : value,\n      ];\n    }),\n  );\n  return { ...built, ...overrides };\n}\n\n/** Merge only plain objects; arrays, null and provider scalar values replace the field. */\nfunction isPlainObject(value: unknown): value is Record<string, unknown> {\n  if (value === null || typeof value !== \"object\") return false;\n  const prototype = Object.getPrototypeOf(value);\n  return prototype === Object.prototype || prototype === null;\n}\n","/**\n * Substrings identifying Ollama model tags whose family emits a\n * reasoning / \"thinking\" channel before the visible answer.\n *\n * Ollama exposes thinking via the request-side `think` flag and the\n * response-side `message.thinking` string. Only models trained for it\n * honor `think`; sending it to a non-reasoning model is a no-op at best.\n * Tags are family-named with optional size/quant suffixes\n * (`deepseek-r1:7b`, `qwq:32b-preview`), so a substring match tolerates\n * the suffixes. Covers the common reasoning families on the Ollama\n * registry; plain instruct models (`llama3.1`, `mistral`, `phi3`) are\n * excluded. Override per-model via\n * `ollama.model({ name, reasoning: true | false })`.\n */\nconst REASONING_CAPABLE_SUBSTRINGS = [\n  \"deepseek-r1\",\n  \"qwq\",\n  \"qwen3\",\n  \"magistral\",\n  \"phi4-reasoning\",\n  \"phi4-mini-reasoning\",\n  \"cogito\",\n  \"smallthinker\",\n  \"exaone-deep\",\n  \"gpt-oss\",\n];\n\n/**\n * Infer whether an Ollama model tag supports a reasoning / thinking\n * channel based on the known thinking-family substrings. Unknown tags\n * default to `false` so the adapter never sends the `think` flag to a\n * model that cannot honor it (a no-op for plain instruct models).\n *\n * @example\n * inferReasoningCapability(\"deepseek-r1:7b\");  // → true\n * inferReasoningCapability(\"qwq:32b\");          // → true\n * inferReasoningCapability(\"llama3.1\");         // → false\n * inferReasoningCapability(\"nomic-embed-text\"); // → false\n */\nexport function inferReasoningCapability(modelName: string): boolean {\n  const normalized = modelName.toLowerCase();\n\n  return REASONING_CAPABLE_SUBSTRINGS.some((fragment) => normalized.includes(fragment));\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  attachCost,\n  resolveStreamChunks,\n  selectProviderTools,\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 ReasoningEffort,\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  Tool,\n} from \"ollama\";\nimport { mergeProviderBody } from \"./provider-options\";\nimport type { OllamaModelConfig } from \"./config.type\";\nimport { inferReasoningCapability } from \"./known-reasoning-models\";\nimport { inferVisionCapability } from \"./known-vision-models\";\nimport {\n  mapDoneReason,\n  toOllamaMessages,\n  toOllamaTools,\n  withCallSignal,\n  wrapOllamaError,\n} from \"./utils\";\n\nconst LOG_MODULE = \"ai.ollama\";\n\n/** `provider` value of `ai.providerTool()` entries this adapter sends verbatim. */\nconst OLLAMA_WIRE_FORMAT = \"ollama\";\n\n/** Ollama supports only low, medium, and high named thinking levels. */\nconst EFFORT_TO_THINK_LEVEL: Record<Exclude<ReasoningEffort, \"none\">, \"low\" | \"medium\" | \"high\"> = {\n  minimal: \"low\",\n  low: \"low\",\n  medium: \"medium\",\n  high: \"high\",\n  xhigh: \"high\",\n  max: \"high\",\n};\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      // Thinking-capable families (deepseek-r1, qwq, qwen3, …) honor the\n      // `think` request flag; plain instruct models do not. Explicit\n      // config wins over the family-substring inference.\n      reasoning: config.reasoning ?? inferReasoningCapability(config.name),\n      // Ollama has no provider-side prompt cache and the chat API takes\n      // no audio / PDF content parts — report these truthfully as false\n      // so the agent rejects unsupported attachments / cache hints\n      // upfront instead of silently dropping them at the wire.\n      promptCaching: false,\n      audio: false,\n      pdf: false,\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. Aborting `options.signal`\n   * cancels the HTTP request through the SDK client's signal-aware `fetch`.\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      options?.signal?.throwIfAborted();\n      response = await withCallSignal(options?.signal, () =>\n        this.client.chat({ ...this.buildRequest(messages, options), stream: false }),\n      );\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 = attachCost(this.extractUsage(response), this.pricing);\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`: aborting cancels the HTTP request (through\n   * the SDK client's signal-aware `fetch`) and the stream iterator.\n   *\n   * With `options.streamChunks` it also yields `reasoning` for each\n   * `message.thinking` piece. Usage arrives only on the final chunk, so no\n   * `usage` chunk; calls arrive whole, so no `tool-args`.\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      options?.signal?.throwIfAborted();\n      stream = await withCallSignal(options?.signal, () =>\n        this.client.chat({ ...this.buildRequest(messages, options), stream: true }),\n      );\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    const want = resolveStreamChunks(options);\n\n    try {\n      for await (const chunk of stream) {\n        // `message.thinking` streams when the request sets `think`.\n        const thinking = chunk.message?.thinking;\n\n        if (want.reasoning && thinking) {\n          yield { type: \"reasoning\", delta: thinking };\n        }\n\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    attachCost(usage, this.pricing);\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 mergeProviderBody<Omit<ChatRequest, \"stream\">>(\n      {\n        model: this.name,\n        messages: toOllamaMessages(messages),\n        ...(Object.keys(ollamaOptions).length > 0 ? { options: ollamaOptions } : {}),\n        ...this.buildTools(options),\n        ...this.buildFormat(options?.responseSchema),\n        ...this.buildThink(options?.reasoning),\n      },\n      options,\n    );\n  }\n\n  /**\n   * Translate the neutral `reasoning` hint into Ollama's `think`\n   * request flag. Ollama's `think` accepts `boolean | 'low' | 'medium'\n   * | 'high'`, so neutral effort levels are clamped to the nearest named\n   * level; an effort-less `reasoning` (only `maxTokens`, or an empty\n   * object) becomes `think: true` to switch the channel on.\n   * The neutral `effort: \"none\"` (\"run without reasoning\") maps to\n   * `think: false` — Ollama has no `\"none\"` level, and this is its\n   * native reasoning-off switch.\n   *\n   * No-ops unless the model is reasoning-capable, so the `think` flag is\n   * never sent to a plain instruct model that cannot honor it.\n   *\n   * `reasoning.maxTokens` (the thinking-budget hint) has no Ollama\n   * equivalent — the daemon does not accept a thinking-token cap — so it\n   * is honored only as the on/off signal above and otherwise ignored.\n   * `ModelCallOptions.cacheControl` is likewise a no-op: Ollama has no\n   * provider prompt cache, so there is no cache breakpoint to place.\n   */\n  private buildThink(reasoning: ModelCallOptions[\"reasoning\"]): Pick<ChatRequest, \"think\"> {\n    if (!reasoning || !this.capabilities.reasoning) {\n      return {};\n    }\n\n    // `effort: \"none\"` = explicit reasoning-off → switch the channel off.\n    // Ollama's `think` has no `\"none\"` level, so map it to `false`.\n    if (reasoning.effort === \"none\") {\n      return { think: false };\n    }\n\n    return { think: reasoning.effort ? EFFORT_TO_THINK_LEVEL[reasoning.effort] : true };\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(options: ModelCallOptions | undefined): Pick<ChatRequest, \"tools\"> {\n    // Ollama's chat API has no built-in tools; provider-defined entries\n    // reach it only when they name this adapter, and others are skipped\n    // with a warning.\n    const { tools, definitions } = selectProviderTools(options, {\n      accepts: [OLLAMA_WIRE_FORMAT, this.provider],\n      module: LOG_MODULE,\n      model: this.name,\n      logger: this.logger,\n    });\n    const mapped = [...(toOllamaTools(tools) ?? []), ...(definitions as unknown as Tool[])];\n\n    return mapped.length > 0 ? { 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   *\n   * Cost-truth: Ollama reports only `prompt_eval_count` (input) and\n   * `eval_count` (output). It has **no** provider prompt cache, so\n   * `Usage.cachedTokens` / `Usage.cacheWriteTokens` stay undefined\n   * (honest absence, not a false zero). Reasoning models emit their\n   * thinking as the `message.thinking` *string* but the wire format\n   * carries **no separate reasoning-token count** — the thinking tokens\n   * are already folded into `eval_count`. We therefore do not fabricate\n   * a `Usage.reasoningTokens` from the text length; it is left undefined\n   * unless/until the daemon exposes a real count. `total` is input +\n   * 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 { OllamaEmbedderConfig, OllamaModelConfig, OllamaSDKConfig } from \"./config.type\";\nimport { OllamaEmbedder } from \"./embedder\";\nimport { OllamaModel } from \"./model\";\nimport { signalAwareFetch } from \"./utils\";\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    // Route every request through a fetch that honours the current call's\n    // `ModelCallOptions.signal` — the `ollama` client takes no per-request\n    // signal of its own (see `utils/call-signal.ts`).\n    this.client = new Ollama({\n      ...clientConfig,\n      fetch: signalAwareFetch(clientConfig.fetch ?? globalThis.fetch),\n    });\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":";;;;;;;;;;;;;;;;;;;AAcA,MAAM,aAAa,IAAIA,mCAA2C;;AAGlE,SAAgB,eAAkB,QAAiC,WAAuB;CACxF,OAAO,WAAW,IAAI,QAAQ,SAAS;AACzC;;;;;;AAOA,SAAgB,iBAAiB,WAAuC;CACtE,QAAQ,OAAO,SAAS;EACtB,MAAM,SAAS,WAAW,SAAS;EAEnC,IAAI,CAAC,QACH,OAAO,UAAU,OAAO,IAAI;EAG9B,MAAM,MAAM,MAAM,UAAU;EAE5B,OAAO,UAAU,OAAO;GACtB,GAAG;GACH,QAAQ,MAAM,YAAY,IAAI,CAAC,KAAK,MAAM,CAAC,IAAI;EACjD,CAAC;CACH;AACF;;;;ACvCA,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,IAAIC,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,CAAC,CAC9E,KAAK,SAAS,KAAK,IAAI,CAAC,CACxB,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,aACP,OACgC;CAChC,MAAM,aAASC,kCAAkB,KAAK;CAEtC,IAAI,UAAU,OAAO,SAAS,UAC5B,OAAO;CAGT,OAAO,EAAE,MAAM,SAAS;AAC1B;;;;;;;;;;;;;;;;;;;;;;;;;ACFA,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,YAAY,QAAgB,QAA8B,WAAmB,UAAU;gBAF5DC;EAGhC,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,EAAE,CAAC;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;;;;AC/GA,MAAM,kCAAkB,IAAI,IAAI;CAAC;CAAS;CAAY;CAAS;AAAQ,CAAC;AAExE,SAAgB,kBACd,OACA,SACG;CAEH,MAAM,QADW,SAAS,iBAAiB,OACtB,EAAE,QAAQ,CAAC;CAChC,MAAM,WAAW,OAAO,KAAK,IAAI,CAAC,CAAC,QAAQ,QAAQ,gBAAgB,IAAI,GAAG,CAAC;CAC3E,IAAI,SAAS,SAAS,GACpB,MAAM,IAAIE,mCACR,0CAA0C,SAAS,KAAK,IAAI,EAAE,sDAC9D,EAAE,SAAS,EAAE,QAAQ,SAAS,EAAE,CAClC;CAEF,MAAM,WAAW;CACjB,MAAM,YAAY,OAAO,YACvB,OAAO,QAAQ,IAAI,CAAC,CAAC,KAAK,CAAC,KAAK,WAAW;EACzC,MAAM,WAAW,SAAS;EAC1B,OAAO,CACL,KACA,cAAc,QAAQ,KAAK,cAAc,KAAK,IAAI;GAAE,GAAG;GAAU,GAAG;EAAM,IAAI,KAChF;CACF,CAAC,CACH;CACA,OAAO;EAAE,GAAG;EAAO,GAAG;CAAU;AAClC;;AAGA,SAAS,cAAc,OAAkD;CACvE,IAAI,UAAU,QAAQ,OAAO,UAAU,UAAU,OAAO;CACxD,MAAM,YAAY,OAAO,eAAe,KAAK;CAC7C,OAAO,cAAc,OAAO,aAAa,cAAc;AACzD;;;;;;;;;;;;;;;;;;ACxBA,MAAM,+BAA+B;CACnC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;;;;;AAcA,SAAgB,yBAAyB,WAA4B;CACnE,MAAM,aAAa,UAAU,YAAY;CAEzC,OAAO,6BAA6B,MAAM,aAAa,WAAW,SAAS,QAAQ,CAAC;AACtF;;;;;;;;;;;;;;;AChCA,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;;;;ACJA,MAAM,aAAa;;AAGnB,MAAM,qBAAqB;;AAG3B,MAAM,wBAA6F;CACjG,SAAS;CACT,KAAK;CACL,QAAQ;CACR,MAAM;CACN,OAAO;CACP,KAAK;AACP;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsCA,IAAa,cAAb,MAAkD;CAUhD,AAAO,YAAY,QAAgB,QAA2B,WAAmB,UAAU;gBAFzDC;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;GAI1D,WAAW,OAAO,aAAa,yBAAyB,OAAO,IAAI;GAKnE,eAAe;GACf,OAAO;GACP,KAAK;EACP;CACF;;;;;;;;CASA,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,SAAS,QAAQ,eAAe;GAChC,WAAW,MAAM,eAAe,SAAS,cACvC,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAM,CAAC,CAC7E;EACF,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,YAAQC,2BAAW,KAAK,aAAa,QAAQ,GAAG,KAAK,OAAO;EAElE,KAAK,OAAO,MAAM,YAAY,YAAY,uBAAuB;GAAE;GAAc;EAAM,CAAC;EAExF,OAAO;GACL,SAAS,SAAS,SAAS,WAAW;GACtC;GACA;GACA;EACF;CACF;;;;;;;;;;;;;CAcA,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,QAAQ,eAAe;GAChC,SAAS,MAAM,eAAe,SAAS,cACrC,KAAK,OAAO,KAAK;IAAE,GAAG,KAAK,aAAa,UAAU,OAAO;IAAG,QAAQ;GAAK,CAAC,CAC5E;EACF,SAAS,QAAQ;GACf,MAAM,KAAK,WAAW,MAAM;EAC9B;EAEA,IAAI,SAAS,QAAQ;GACnB,IAAI,QAAQ,OAAO,SACjB,OAAO,MAAM;QAEb,QAAQ,OAAO,iBAAiB,eAAe,OAAO,MAAM,GAAG,EAAE,MAAM,KAAK,CAAC;EAEjF;EAEA,IAAI;EACJ,IAAI,cAAc;EAClB,MAAM,QAAe;GAAE,OAAO;GAAG,QAAQ;GAAG,OAAO;EAAE;EACrD,MAAM,WAAOC,oCAAoB,OAAO;EAExC,IAAI;GACF,WAAW,MAAM,SAAS,QAAQ;IAEhC,MAAM,WAAW,MAAM,SAAS;IAEhC,IAAI,KAAK,aAAa,UACpB,MAAM;KAAE,MAAM;KAAa,OAAO;IAAS;IAG7C,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,+BAAW,OAAO,KAAK,OAAO;EAE9B,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,kBACL;GACE,OAAO,KAAK;GACZ,UAAU,iBAAiB,QAAQ;GACnC,GAAI,OAAO,KAAK,aAAa,CAAC,CAAC,SAAS,IAAI,EAAE,SAAS,cAAc,IAAI,CAAC;GAC1E,GAAG,KAAK,WAAW,OAAO;GAC1B,GAAG,KAAK,YAAY,SAAS,cAAc;GAC3C,GAAG,KAAK,WAAW,SAAS,SAAS;EACvC,GACA,OACF;CACF;;;;;;;;;;;;;;;;;;;;CAqBA,AAAQ,WAAW,WAAsE;EACvF,IAAI,CAAC,aAAa,CAAC,KAAK,aAAa,WACnC,OAAO,CAAC;EAKV,IAAI,UAAU,WAAW,QACvB,OAAO,EAAE,OAAO,MAAM;EAGxB,OAAO,EAAE,OAAO,UAAU,SAAS,sBAAsB,UAAU,UAAU,KAAK;CACpF;;;;;CAMA,AAAQ,WAAW,SAAmE;EAIpF,MAAM,EAAE,OAAO,oBAAgBC,oCAAoB,SAAS;GAC1D,SAAS,CAAC,oBAAoB,KAAK,QAAQ;GAC3C,QAAQ;GACR,OAAO,KAAK;GACZ,QAAQ,KAAK;EACf,CAAC;EACD,MAAM,SAAS,CAAC,GAAI,cAAc,KAAK,KAAK,CAAC,GAAI,GAAI,WAAiC;EAEtF,OAAO,OAAO,SAAS,IAAI,EAAE,OAAO,OAAO,IAAI,CAAC;CAClD;;;;;;;;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;;;;;;;;;;;;;;;CAgBA,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;;;;;;;;;;;;;;;;;;;;;;;;;;;AChYA,IAAa,YAAb,MAAqD;CAKnD,AAAO,YAAY,SAA0B,CAAC,GAAG;EAC/C,MAAM,EAAE,UAAU,SAAS,GAAG,iBAAiB;EAK/C,KAAK,SAAS,IAAIC,cAAO;GACvB,GAAG;GACH,OAAO,iBAAiB,aAAa,SAAS,WAAW,KAAK;EAChE,CAAC;EACD,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,WAAOC,sCAAsB,IAAI;CACnC;;;;;;;;CASA,AAAO,SAAS,QAAgD;EAC9D,OAAO,IAAI,eAAe,KAAK,QAAQ,QAAQ,KAAK,QAAQ;CAC9D;AACF"}

package/esm/config.type.d.mts

@@ -1,6 +1,5 @@
 import { Config } from "ollama";
 import { EmbedderConfig, ModelConfig, ModelPricing } from "@warlock.js/ai";
-
 //#region ../../packages/ai-ollama/src/config.type.d.ts
 /**
  * Configuration for the Ollama SDK adapter.
@@ -30,7 +29,7 @@ import { EmbedderConfig, ModelConfig, ModelPricing } from "@warlock.js/ai";
  *   headers: { Authorization: `Bearer ${process.env.OLLAMA_TOKEN}` },
  * });
  */
-type OllamaSDKConfig = Partial<Config> & {
+export type OllamaSDKConfig = Partial<Config> & {
   provider?: string;
   /**
    * Per-model USD pricing registry, keyed by model name. Surfaced onto
@@ -48,7 +47,7 @@ type OllamaSDKConfig = Partial<Config> & {
  * ollama.model({ name: "llama3.1" });
  * ollama.model({ name: "llama3.2-vision", vision: true });
  */
-type OllamaModelConfig = ModelConfig & {
+export type OllamaModelConfig = ModelConfig & {
   /**
    * Override the auto-inferred vision capability. When omitted, the
    * adapter checks the model tag against the known multimodal Ollama
@@ -84,7 +83,6 @@ type OllamaModelConfig = ModelConfig & {
  * @example
  * ollama.embedder({ name: "nomic-embed-text" });
  */
-type OllamaEmbedderConfig = EmbedderConfig;
+export type OllamaEmbedderConfig = EmbedderConfig;
 //#endregion
-export { OllamaEmbedderConfig, OllamaModelConfig, OllamaSDKConfig };
 //# sourceMappingURL=config.type.d.mts.map

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

@@ -1 +1 @@
-{"version":3,"file":"config.type.d.mts","names":[],"sources":["../../../../../../../packages/ai-ollama/src/config.type.ts"],"mappings":";;;;;;AA+BA;;;;;;;;;;;;;;;;AAOuC;AAYvC;;;;;;;;;KAnBY,eAAA,GAAkB,OAAA,CAAQ,MAAA;EACpC,QAAA;EAuD8B;;;AAAiB;;EAjD/C,OAAA,GAAU,MAAA,SAAe,YAAA;AAAA;;;;;;;;;;KAYf,iBAAA,GAAoB,WAAW;;;;;;;EAOzC,MAAA;;;;;;;;EAQA,gBAAA;;;;;;;;;;EAUA,SAAA;AAAA;;;;;;;;;;KAYU,oBAAA,GAAuB,cAAc"}
+{"version":3,"file":"config.type.d.mts","names":[],"sources":["../../../../../../../packages/ai-ollama/src/config.type.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;YA+BY,kBAAkB,QAAQ;EACpC;;;;;;EAMA,UAAU,eAAe;;;;;;;;;;;YAYf,oBAAoB;;;;;;;EAO9B;;;;;;;;EAQA;;;;;;;;;;EAUA;;;;;;;;;;;YAYU,uBAAuB"}

package/esm/embedder.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"embedder.mjs","names":[],"sources":["../../../../../../../packages/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,EAAE,CAAC;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"}
+{"version":3,"file":"embedder.mjs","names":[],"sources":["../../../../../../../packages/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(client: Ollama, config: OllamaEmbedderConfig, provider: string = \"ollama\") {\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,YAAY,QAAgB,QAA8B,WAAmB,UAAU;gBAF5D;EAGhC,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,EAAE,CAAC;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

@@ -1,3 +1,4 @@
 import { OllamaEmbedderConfig, OllamaModelConfig, OllamaSDKConfig } from "./config.type.mjs";
 import { OllamaSDK } from "./sdk.mjs";
-export { type OllamaEmbedderConfig, type OllamaModelConfig, OllamaSDK, type OllamaSDKConfig };
+import { OllamaCallOptions } from "./provider-options.mjs";
+export { type OllamaCallOptions, type OllamaEmbedderConfig, type OllamaModelConfig, OllamaSDK, type OllamaSDKConfig };