@nemesis-oss/ollama-sdk 1.12.1 → 1.14.0

package/dist/index.cjs

@@ -2817,12 +2817,29 @@ var ModelsClient = class {
    * await client.models.createModelFromGguf('my-model', '/path/to/model.gguf', {
    *   template: '{{ .Prompt }}',
    * });
+   *
+   * // Speculative decoding: primary + draft shards + tuning:
+   * await client.models.createModelFromGguf('my-model', '/path/to/model.gguf', {
+   *   draftGguf: '/path/to/draft.gguf',
+   *   draftQuantize: 'q8_0',
+   *   parameters: { draft_num_predict: 8 },
+   * });
    * ```
    *
+   * Speculative-decoding models assemble the same way: pass the draft
+   * network's GGUF(s) as `draftGguf` (uploaded into `draft_files`), optionally
+   * set `draft_quantize` to compress the draft weights independently of the
+   * primary model, and set `parameters.draft_num_predict` to tune how many
+   * tokens the draft network predicts per step (Ollama defaults separate
+   * draft models to 4; `0` disables drafting; models with embedded MTP
+   * tensors require setting it explicitly).
+   *
    * @param model Name for the new model (e.g. `'my-model:latest'`).
    * @param gguf Path(s) to GGUF file(s) on the local filesystem.
    * @param opts Everything else `/api/create` accepts (`template`, `system`,
-   *   `parameters`, `license`, `quantize`, …) except `files` and `from`.
+   *   `parameters`, `license`, `quantize`, `draftGguf`, …) except `files`,
+   *   `draft_files` (use `draftGguf` for path-based drafts; an explicit digest
+   *   map is still accepted and merged), `from`, and `stream`.
    */
   async createModelFromGguf(model, gguf, opts) {
     const paths = typeof gguf === "string" ? [gguf] : gguf;
@@ -2836,14 +2853,29 @@ var ModelsClient = class {
       const uploaded = await this.createBlobFromFile(path, { signal: opts?.signal });
       files[uploaded.fileName] = uploaded.digest;
     }
+    const draftOpt = opts?.draftGguf;
+    const draftPaths = typeof draftOpt === "string" ? [draftOpt] : draftOpt ?? [];
     const {
       model: _model,
       files: _files,
       from: _from,
       stream: _stream,
+      draftGguf: _draftGguf,
+      draft_files: _draftFiles,
       ...rest
     } = opts ?? {};
-    return this.create({ ...rest, model, files, stream: false });
+    const draftFiles = { ..._draftFiles ?? {} };
+    for (const path of draftPaths) {
+      const uploaded = await this.createBlobFromFile(path, { signal: opts?.signal });
+      draftFiles[uploaded.fileName] = uploaded.digest;
+    }
+    return this.create({
+      ...rest,
+      model,
+      files,
+      ...Object.keys(draftFiles).length > 0 ? { draft_files: draftFiles } : {},
+      stream: false
+    });
   }
 };
 
@@ -2851,7 +2883,11 @@ var ModelsClient = class {
 var OpenAIResponsesStreamError = class extends OllamaClientError {
   responsePayload;
   constructor(message, responsePayload) {
-    super(message, { code: "openai_responses_stream_error", retryable: false, response: { body: responsePayload } });
+    super(message, {
+      code: "openai_responses_stream_error",
+      retryable: false,
+      response: { body: responsePayload }
+    });
     this.responsePayload = responsePayload;
   }
 };
@@ -2993,7 +3029,9 @@ var OpenAIChatCompletionStream = class {
     } finally {
       if (!completed) {
         this.source.abort?.();
-        this.rejectFinal(new OllamaAbortError("OpenAI Chat Completion stream ended before completion"));
+        this.rejectFinal(
+          new OllamaAbortError("OpenAI Chat Completion stream ended before completion")
+        );
       }
     }
   }
@@ -3122,7 +3160,14 @@ var OpenAIResponsesStream = class {
         existing.kind = kind;
         return existing;
       }
-      const state = { itemId, kind, content: /* @__PURE__ */ new Map(), summary: /* @__PURE__ */ new Map(), arguments: "", reasoningText: "" };
+      const state = {
+        itemId,
+        kind,
+        content: /* @__PURE__ */ new Map(),
+        summary: /* @__PURE__ */ new Map(),
+        arguments: "",
+        reasoningText: ""
+      };
       outputs.set(outputIndex, state);
       return state;
     };
@@ -3153,9 +3198,33 @@ var OpenAIResponsesStream = class {
     const reconstructedOutput = () => [...outputs.entries()].sort(([a], [b]) => a - b).map(([outputIndex, state]) => {
       const canonical = canonicalOutputItems.get(outputIndex);
       if (canonical) return canonical;
-      if (state.kind === "message") return { type: "message", ...state.itemId ? { id: state.itemId } : {}, ...state.status ? { status: state.status } : {}, role: state.role ?? "assistant", ...state.phase ? { phase: state.phase } : {}, content: [...state.content.entries()].sort(([a], [b]) => a - b).map(([, p]) => p) };
-      if (state.kind === "function_call") return { type: "function_call", ...state.itemId ? { id: state.itemId } : {}, ...state.callId ? { call_id: state.callId } : {}, ...state.status ? { status: state.status } : {}, name: state.name ?? "", arguments: state.arguments };
-      return { type: "reasoning", ...state.itemId ? { id: state.itemId } : {}, ...state.status ? { status: state.status } : {}, ...state.summary.size ? { summary: [...state.summary.entries()].sort(([a], [b]) => a - b).map(([, text]) => ({ type: "summary_text", text })) } : {}, ...state.reasoningText ? { text: state.reasoningText } : {} };
+      if (state.kind === "message")
+        return {
+          type: "message",
+          ...state.itemId ? { id: state.itemId } : {},
+          ...state.status ? { status: state.status } : {},
+          role: state.role ?? "assistant",
+          ...state.phase ? { phase: state.phase } : {},
+          content: [...state.content.entries()].sort(([a], [b]) => a - b).map(([, p]) => p)
+        };
+      if (state.kind === "function_call")
+        return {
+          type: "function_call",
+          ...state.itemId ? { id: state.itemId } : {},
+          ...state.callId ? { call_id: state.callId } : {},
+          ...state.status ? { status: state.status } : {},
+          name: state.name ?? "",
+          arguments: state.arguments
+        };
+      return {
+        type: "reasoning",
+        ...state.itemId ? { id: state.itemId } : {},
+        ...state.status ? { status: state.status } : {},
+        ...state.summary.size ? {
+          summary: [...state.summary.entries()].sort(([a], [b]) => a - b).map(([, text]) => ({ type: "summary_text", text }))
+        } : {},
+        ...state.reasoningText ? { text: state.reasoningText } : {}
+      };
     });
     try {
       for await (const event of this.source) {
@@ -3199,7 +3268,12 @@ var OpenAIResponsesStream = class {
           case "response.content_part.done": {
             const e = payload;
             if (e.part.type === "output_text" && "text" in e.part && typeof e.part.text === "string" || e.part.type === "refusal" && "refusal" in e.part && typeof e.part.refusal === "string") {
-              applyContentPart(e.output_index, e.item_id, e.content_index, e.part);
+              applyContentPart(
+                e.output_index,
+                e.item_id,
+                e.content_index,
+                e.part
+              );
             }
             break;
           }
@@ -3207,7 +3281,10 @@ var OpenAIResponsesStream = class {
             const e = payload;
             const s = ensureState(e.output_index, e.item_id, "message");
             const p = s.content.get(e.content_index);
-            s.content.set(e.content_index, { type: "output_text", text: (p?.type === "output_text" ? p.text : "") + e.delta });
+            s.content.set(e.content_index, {
+              type: "output_text",
+              text: (p?.type === "output_text" ? p.text : "") + e.delta
+            });
             break;
           }
           case "response.output_text.done": {
@@ -3220,7 +3297,10 @@ var OpenAIResponsesStream = class {
             const e = payload;
             const s = ensureState(e.output_index, e.item_id, "message");
             const p = s.content.get(e.content_index);
-            s.content.set(e.content_index, { type: "refusal", refusal: (p?.type === "refusal" ? p.refusal : "") + e.delta });
+            s.content.set(e.content_index, {
+              type: "refusal",
+              refusal: (p?.type === "refusal" ? p.refusal : "") + e.delta
+            });
             break;
           }
           case "response.refusal.done": {
@@ -3282,20 +3362,31 @@ var OpenAIResponsesStream = class {
         const response = terminalFailure ?? terminalIncomplete;
         const status = terminalFailure ? "failed" : "incomplete";
         const detail = response.error?.message ?? response.incomplete_details?.reason ?? "unknown error";
-        throw new OpenAIResponsesStreamError("OpenAI Responses stream " + status + ": " + detail, response);
+        throw new OpenAIResponsesStreamError(
+          "OpenAI Responses stream " + status + ": " + detail,
+          response
+        );
       }
       if (!finalResponse) {
-        if (!responseMeta) throw new Error("OpenAI Responses stream ended without a terminal response payload");
+        if (!responseMeta)
+          throw new Error("OpenAI Responses stream ended without a terminal response payload");
         if (responseMeta.status === "failed" || responseMeta.status === "incomplete") {
           const detail = responseMeta.error?.message ?? responseMeta.incomplete_details?.reason ?? "unknown error";
-          throw new OpenAIResponsesStreamError("OpenAI Responses stream " + responseMeta.status + ": " + detail, responseMeta);
+          throw new OpenAIResponsesStreamError(
+            "OpenAI Responses stream " + responseMeta.status + ": " + detail,
+            responseMeta
+          );
         }
         finalResponse = { ...responseMeta, output: reconstructedOutput(), status: "completed" };
       } else if (finalResponse.status === "failed" || finalResponse.status === "incomplete") {
         const detail = finalResponse.error?.message ?? finalResponse.incomplete_details?.reason ?? "unknown error";
-        throw new OpenAIResponsesStreamError("OpenAI Responses stream " + finalResponse.status + ": " + detail, finalResponse);
+        throw new OpenAIResponsesStreamError(
+          "OpenAI Responses stream " + finalResponse.status + ": " + detail,
+          finalResponse
+        );
       }
-      if (finalResponse.output.length === 0 && outputs.size > 0) finalResponse = { ...finalResponse, output: reconstructedOutput() };
+      if (finalResponse.output.length === 0 && outputs.size > 0)
+        finalResponse = { ...finalResponse, output: reconstructedOutput() };
       completed = true;
       this.removeAbortListener?.();
       this.removeAbortListener = void 0;
@@ -3320,14 +3411,11 @@ var OpenAICompatClient = class {
   }
   request(operation, model, signal, holdUntil) {
     if (this.runner) {
-      return this.runner(
-        (http, runnerSignal) => operation(http, runnerSignal),
-        {
-          ...model !== void 0 ? { model } : {},
-          ...signal !== void 0 ? { signal } : {},
-          ...holdUntil !== void 0 ? { holdUntil } : {}
-        }
-      );
+      return this.runner((http, runnerSignal) => operation(http, runnerSignal), {
+        ...model !== void 0 ? { model } : {},
+        ...signal !== void 0 ? { signal } : {},
+        ...holdUntil !== void 0 ? { holdUntil } : {}
+      });
     }
     return operation(this.http, signal);
   }
@@ -3956,7 +4044,7 @@ var SystemOneContentSchema = zod.z.union([
 var SystemOneChoiceQuestionSchema = zod.z.object({
   type: zod.z.enum(["choice"]),
   instructions: SystemOneContentSchema,
-  criteria: zod.z.object({}).catchall(zod.z.union([zod.z.string(), zod.z.null()])).refine((v) => Object.keys(v).length >= 2, { message: "must have at least 2 properties" }).refine((v) => Object.keys(v).length <= 26, { message: "must have at most 26 properties" }).refine((v) => Object.keys(v).every((k) => new RegExp("\\S").test(k)), {
+  criteria: zod.z.object({}).catchall(zod.z.union([zod.z.string(), zod.z.null()])).refine((v) => Object.keys(v).length >= 2, { message: "must have at least 2 properties" }).refine((v) => Object.keys(v).length <= 255, { message: "must have at most 255 properties" }).refine((v) => Object.keys(v).every((k) => new RegExp("\\S").test(k)), {
     message: 'property names must match pattern "\\S"'
   })
 });
@@ -4095,6 +4183,7 @@ var requestSchemas = {
 var responseSchemas = {
   balance: BalanceResponseSchema,
   chat: ChatResponseSchema,
+  copy: StatusResponseSchema,
   create: StatusResponseSchema,
   embed: EmbedResponseSchema,
   generate: GenerateResponseSchema,
@@ -5001,6 +5090,7 @@ var copyOp = {
   method: "POST",
   path: "/api/copy",
   request: { $ref: "#/schemas/CopyRequest" },
+  response: { $ref: "#/schemas/StatusResponse" },
   environment: { local: true, cloud: false },
   transport: { mode: "json", streaming: false },
   capabilities: {},
@@ -6887,6 +6977,93 @@ var ToolRegistry = class {
     };
   }
 };
+var WEB_SEARCH_TOOL_NAME = "web_search";
+var WEB_FETCH_TOOL_NAME = "web_fetch";
+var DEFAULT_MAX_OUTPUT_CHARS = 2e3 * 4;
+var DEFAULT_MAX_LINKS = 25;
+var TRUNCATION_MARKER = "\u2026[truncated]";
+var webSearchSchema = zod.z.object({
+  query: zod.z.string().min(1).describe("The search query string"),
+  max_results: zod.z.number().int().min(1).max(10).optional().describe("Maximum results to return (default 5, max 10)")
+});
+var webFetchSchema = zod.z.object({
+  url: zod.z.string().min(1).describe("The URL of the web page to fetch")
+});
+function positiveInt2(value, fallback, name) {
+  if (value === void 0) return fallback;
+  if (!Number.isInteger(value) || value < 1) {
+    throw new RangeError(`${name} must be a positive integer, received ${String(value)}`);
+  }
+  return value;
+}
+function truncate(text, maxChars) {
+  if (text.length <= maxChars) return text;
+  if (maxChars <= TRUNCATION_MARKER.length) return text.slice(0, maxChars);
+  return text.slice(0, maxChars - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
+}
+function formatWebSearchResult(res, maxChars) {
+  const results = res.results.map(({ title, url, content }) => ({ title, url, content }));
+  const full = JSON.stringify({ results });
+  if (full.length <= maxChars || results.length === 0) return truncate(full, maxChars);
+  const overhead = JSON.stringify({
+    results: results.map((r) => ({ ...r, content: "" }))
+  }).length;
+  const perResult = Math.max(0, Math.floor((maxChars - overhead) / results.length));
+  const trimmed = JSON.stringify({
+    results: results.map((r) => ({ ...r, content: truncate(r.content, perResult) }))
+  });
+  return trimmed.length <= maxChars ? trimmed : truncate(trimmed, maxChars);
+}
+function formatWebFetchResult(res, maxChars, maxLinks = DEFAULT_MAX_LINKS) {
+  const page = {
+    title: res.title ?? "",
+    content: res.content,
+    links: (res.links ?? []).slice(0, maxLinks)
+  };
+  const full = JSON.stringify(page);
+  if (full.length <= maxChars) return full;
+  const overhead = JSON.stringify({ ...page, content: "" }).length;
+  const trimmed = JSON.stringify({
+    ...page,
+    content: truncate(page.content, Math.max(0, maxChars - overhead))
+  });
+  return trimmed.length <= maxChars ? trimmed : truncate(trimmed, maxChars);
+}
+function createWebSearchTool(client, options = {}) {
+  const maxChars = positiveInt2(options.maxOutputChars, DEFAULT_MAX_OUTPUT_CHARS, "maxOutputChars");
+  const defaultMaxResults = options.defaultMaxResults === void 0 ? void 0 : Math.min(positiveInt2(options.defaultMaxResults, 5, "defaultMaxResults"), 10);
+  return defineTool({
+    name: WEB_SEARCH_TOOL_NAME,
+    description: "Search the web for current information. Returns a list of results, each with a title, url, and relevant content snippet.",
+    schema: webSearchSchema,
+    timeoutMs: options.timeoutMs,
+    execute: async ({ query, max_results }, { signal }) => {
+      const res = await client.webSearch({
+        query,
+        ...(max_results ?? defaultMaxResults) !== void 0 ? { max_results: max_results ?? defaultMaxResults } : {},
+        ...signal !== void 0 ? { signal } : {}
+      });
+      return formatWebSearchResult(res, maxChars);
+    }
+  });
+}
+function createWebFetchTool(client, options = {}) {
+  const maxChars = positiveInt2(options.maxOutputChars, DEFAULT_MAX_OUTPUT_CHARS, "maxOutputChars");
+  const maxLinks = positiveInt2(options.maxLinks, DEFAULT_MAX_LINKS, "maxLinks");
+  return defineTool({
+    name: WEB_FETCH_TOOL_NAME,
+    description: "Fetch a single web page by URL. Returns the page title, its main content, and links found on the page.",
+    schema: webFetchSchema,
+    timeoutMs: options.timeoutMs,
+    execute: async ({ url }, { signal }) => {
+      const res = await client.webFetch({ url, ...signal !== void 0 ? { signal } : {} });
+      return formatWebFetchResult(res, maxChars, maxLinks);
+    }
+  });
+}
+function createWebTools(client, options = {}) {
+  return [createWebSearchTool(client, options), createWebFetchTool(client, options)];
+}
 
 // src/mcp/json-schema.ts
 function label(path) {
@@ -7823,6 +8000,8 @@ exports.RETRYABLE_STATUS_CODES = RETRYABLE_STATUS_CODES;
 exports.ResponsesModule = ResponsesModule;
 exports.ToolCallAccumulator = ToolCallAccumulator;
 exports.ToolRegistry = ToolRegistry;
+exports.WEB_FETCH_TOOL_NAME = WEB_FETCH_TOOL_NAME;
+exports.WEB_SEARCH_TOOL_NAME = WEB_SEARCH_TOOL_NAME;
 exports.addSpanEvent = addSpanEvent;
 exports.applySkill = applySkill;
 exports.batchEmbed = batchEmbed;
@@ -7839,6 +8018,9 @@ exports.createDecision = createDecision;
 exports.createModelAffinityScheduler = createModelAffinityScheduler;
 exports.createOllamaCloudFreeTierQuota = createOllamaCloudFreeTierQuota;
 exports.createTimeoutSignal = createTimeoutSignal;
+exports.createWebFetchTool = createWebFetchTool;
+exports.createWebSearchTool = createWebSearchTool;
+exports.createWebTools = createWebTools;
 exports.defineTool = defineTool;
 exports.detectModelCapabilities = detectModelCapabilities;
 exports.embedBatchOverflowMessage = embedBatchOverflowMessage;
@@ -7855,6 +8037,8 @@ exports.fetchWithBackoff = fetchWithBackoff;
 exports.findOversizedEmbedInputs = findOversizedEmbedInputs;
 exports.findRunningModelContextLength = findRunningModelContextLength;
 exports.formatTelemetry = formatTelemetry;
+exports.formatWebFetchResult = formatWebFetchResult;
+exports.formatWebSearchResult = formatWebSearchResult;
 exports.imageStringNeedsResolution = imageStringNeedsResolution;
 exports.inferRuntimeMode = inferRuntimeMode;
 exports.isKeepAliveSugar = isKeepAliveSugar;