@ziro-agent/ollama 0.2.0

package/LICENSE

@@ -0,0 +1,190 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for describing the origin of the Work and
+      reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright 2026 ZiroAgent Contributors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,62 @@
+# @ziro-agent/ollama
+
+Ollama provider for ZiroAgent SDK — run open-weight LLMs locally with
+the same interface as `@ziro-agent/openai` / `@ziro-agent/anthropic`.
+
+> **Sovereign pillar primitive.** Part of the v0.1.9 *trust recovery*
+> milestone (RFC 0004). Lets the SDK keep its OSS-first promise: you
+> can build an agent end-to-end without ever sending bytes to a
+> third-party API.
+
+## Install
+
+```bash
+npm install @ziro-agent/ollama @ziro-agent/agent
+```
+
+## Usage
+
+```ts
+import { createAgent, generateText } from '@ziro-agent/agent';
+import { ollama } from '@ziro-agent/ollama';
+
+const text = await generateText({
+  model: ollama('llama3.1'),
+  prompt: 'Explain RAG to a junior engineer.',
+});
+
+const agent = createAgent({ model: ollama('qwen2.5'), tools: { /* ... */ } });
+```
+
+For a remote / containerised daemon:
+
+```ts
+import { createOllama } from '@ziro-agent/ollama';
+
+const ollama = createOllama({
+  baseURL: process.env.OLLAMA_BASE_URL ?? 'http://ollama:11434',
+  defaultOptions: { num_ctx: 8192, repeat_penalty: 1.1 },
+});
+```
+
+## Budget Guard
+
+Local models are free at runtime, so `Budget Guard` reports `$0` and
+`pricingAvailable: true`. `maxUsd` constraints simply never trip
+(instead of being silently disabled). The meaningful local-runtime
+limits keep working exactly as documented:
+
+- `maxTokens`
+- `maxLlmCalls`
+- `maxDurationMs`
+- `maxSteps`
+
+## Tools
+
+Uses Ollama's native function-calling protocol (not the OpenAI compat
+shim — it loses tool-call fidelity on several models). Recommended
+tool-capable models: `llama3.1`, `qwen2.5`, `mistral-nemo`, `gemma3`.
+
+## License
+
+Apache-2.0 © ZiroAgent

package/dist/index.cjs

@@ -0,0 +1,280 @@
+'use strict';
+
+var core = require('@ziro-agent/core');
+
+// src/ollama-chat-model.ts
+
+// src/util/ndjson.ts
+async function* parseNDJSON(body) {
+  const reader = body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+  try {
+    while (true) {
+      const { value, done } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      let nl = buffer.indexOf("\n");
+      while (nl !== -1) {
+        const line = buffer.slice(0, nl).replace(/\r$/, "").trim();
+        buffer = buffer.slice(nl + 1);
+        if (line.length > 0) {
+          yield JSON.parse(line);
+        }
+        nl = buffer.indexOf("\n");
+      }
+    }
+    const tail = buffer.trim();
+    if (tail.length > 0) yield JSON.parse(tail);
+  } finally {
+    reader.releaseLock();
+  }
+}
+
+// src/ollama-chat-model.ts
+var OllamaChatModel = class {
+  provider = "ollama";
+  modelId;
+  config;
+  constructor(config) {
+    this.modelId = config.modelId;
+    this.config = config;
+  }
+  async generate(options) {
+    const body = this.buildBody(options, false);
+    const res = await this.fetch("/api/chat", body, options);
+    const json = await res.json();
+    const text = json.message?.content ?? "";
+    const toolCalls = json.message?.tool_calls?.map((tc, i) => ({
+      type: "tool-call",
+      toolCallId: synthesizeToolCallId(tc.function?.name ?? "unknown", i),
+      toolName: tc.function?.name ?? "",
+      args: tc.function?.arguments ?? {}
+    })) ?? [];
+    return {
+      text,
+      content: [...text.length > 0 ? [{ type: "text", text }] : [], ...toolCalls],
+      toolCalls,
+      finishReason: mapFinishReason(json),
+      usage: mapUsage(json),
+      rawResponse: json
+    };
+  }
+  async stream(options) {
+    const body = this.buildBody(options, true);
+    const res = await this.fetch("/api/chat", body, options);
+    if (!res.body) {
+      throw new core.APICallError({
+        message: "Ollama streaming response has no body.",
+        statusCode: res.status
+      });
+    }
+    const ndjson = parseNDJSON(res.body);
+    return new ReadableStream({
+      async start(controller) {
+        let usage;
+        let finish = "unknown";
+        let toolIdx = 0;
+        try {
+          for await (const chunk of ndjson) {
+            if (chunk.message?.content) {
+              controller.enqueue({ type: "text-delta", textDelta: chunk.message.content });
+            }
+            if (chunk.message?.tool_calls) {
+              for (const tc of chunk.message.tool_calls) {
+                const id = synthesizeToolCallId(tc.function?.name ?? "unknown", toolIdx++);
+                controller.enqueue({
+                  type: "tool-call",
+                  toolCallId: id,
+                  toolName: tc.function?.name ?? "",
+                  args: tc.function?.arguments ?? {}
+                });
+              }
+            }
+            if (chunk.done) {
+              finish = mapFinishReason(chunk);
+              usage = mapUsage(chunk);
+            }
+          }
+          controller.enqueue({ type: "finish", finishReason: finish, usage: usage ?? {} });
+          controller.close();
+        } catch (err) {
+          controller.enqueue({ type: "error", error: err });
+          controller.close();
+        }
+      }
+    });
+  }
+  /**
+   * Local models are free at runtime — their cost is electricity / GPU
+   * time, not per-token billing. We report `pricingAvailable: true`
+   * with `$0` so `Budget Guard.maxUsd` constraints simply never trip
+   * for Ollama runs (instead of being silently disabled, which would
+   * surprise callers migrating from OpenAI).
+   *
+   * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets
+   * keep working as documented — those are the meaningful limits when
+   * GPU time is the bottleneck.
+   */
+  estimateCost(options) {
+    const inputTokens = core.estimateTokensFromMessages(asChatMessages(options.messages));
+    const maxOut = options.maxTokens ?? 8192;
+    const minOut = Math.min(16, maxOut);
+    return {
+      minTokens: inputTokens + minOut,
+      maxTokens: inputTokens + maxOut,
+      minUsd: 0,
+      maxUsd: 0,
+      pricingAvailable: true
+    };
+  }
+  buildBody(options, stream) {
+    const ollamaOptions = {
+      ...this.config.defaultOptions ?? {}
+    };
+    if (options.temperature !== void 0) ollamaOptions.temperature = options.temperature;
+    if (options.topP !== void 0) ollamaOptions.top_p = options.topP;
+    if (options.maxTokens !== void 0) ollamaOptions.num_predict = options.maxTokens;
+    if (options.stopSequences !== void 0) ollamaOptions.stop = options.stopSequences;
+    if (options.seed !== void 0) ollamaOptions.seed = options.seed;
+    const body = {
+      model: this.modelId,
+      messages: options.messages.map(toOllamaMessage),
+      stream
+    };
+    if (Object.keys(ollamaOptions).length > 0) body.options = ollamaOptions;
+    if (options.tools?.length) {
+      body.tools = options.tools.map((t) => ({
+        type: "function",
+        function: {
+          name: t.name,
+          ...t.description !== void 0 ? { description: t.description } : {},
+          parameters: t.parameters
+        }
+      }));
+    }
+    if (options.providerOptions) Object.assign(body, options.providerOptions);
+    return body;
+  }
+  async fetch(path, body, options) {
+    const url = `${this.config.baseURL}${path}`;
+    const headers = { ...this.config.headers, ...options.headers };
+    const init = {
+      method: "POST",
+      headers,
+      body: JSON.stringify(body)
+    };
+    if (options.abortSignal) init.signal = options.abortSignal;
+    const res = await this.config.fetcher(url, init);
+    if (!res.ok) {
+      const text = await res.text().catch(() => "");
+      throw new core.APICallError({
+        message: `Ollama API error: ${res.status} ${res.statusText}`,
+        url,
+        statusCode: res.status,
+        responseBody: text
+      });
+    }
+    return res;
+  }
+};
+function toOllamaMessage(m) {
+  switch (m.role) {
+    case "system":
+    case "user": {
+      const text = m.content.filter((p) => p.type === "text").map((p) => p.text).join("");
+      const images = m.content.filter((p) => p.type === "image").map((p) => {
+        const img = p.image;
+        if (typeof img === "string") return img.replace(/^data:[^;]+;base64,/, "");
+        if (img instanceof URL) return img.toString();
+        return uint8ToBase64(img);
+      });
+      const out = { role: m.role, content: text };
+      if (images.length > 0) out.images = images;
+      return out;
+    }
+    case "assistant": {
+      const text = m.content.filter((p) => p.type === "text").map((p) => p.text).join("");
+      const toolCalls = m.content.filter((p) => p.type === "tool-call");
+      const out = { role: "assistant", content: text };
+      if (toolCalls.length > 0) {
+        out.tool_calls = toolCalls.map((tc) => ({
+          function: {
+            name: tc.toolName,
+            arguments: tc.args ?? {}
+          }
+        }));
+      }
+      return out;
+    }
+    case "tool": {
+      const first = m.content[0];
+      if (!first || first.type !== "tool-result") {
+        return { role: "tool", content: "" };
+      }
+      return {
+        role: "tool",
+        content: typeof first.result === "string" ? first.result : JSON.stringify(first.result)
+      };
+    }
+  }
+}
+function synthesizeToolCallId(name, index) {
+  return `ollama_${name}_${index}_${Math.random().toString(36).slice(2, 8)}`;
+}
+function asChatMessages(messages) {
+  return messages;
+}
+function uint8ToBase64(arr) {
+  let s = "";
+  for (let i = 0; i < arr.byteLength; i++) s += String.fromCharCode(arr[i]);
+  return typeof btoa !== "undefined" ? btoa(s) : Buffer.from(s, "binary").toString("base64");
+}
+function mapFinishReason(json) {
+  if (json.message?.tool_calls?.length) return "tool-calls";
+  if (json.done_reason === "stop" || json.done) return "stop";
+  if (json.done_reason === "length") return "length";
+  return "unknown";
+}
+function mapUsage(json) {
+  const out = {};
+  if (json.prompt_eval_count !== void 0) out.promptTokens = json.prompt_eval_count;
+  if (json.eval_count !== void 0) out.completionTokens = json.eval_count;
+  if (json.prompt_eval_count !== void 0 && json.eval_count !== void 0) {
+    out.totalTokens = json.prompt_eval_count + json.eval_count;
+  }
+  return out;
+}
+
+// src/ollama-provider.ts
+function createOllama(options = {}) {
+  const baseURL = (options.baseURL ?? loadEnv("OLLAMA_BASE_URL") ?? "http://localhost:11434").replace(/\/+$/, "");
+  const fetcher = options.fetch ?? globalThis.fetch;
+  const headers = {
+    "Content-Type": "application/json",
+    ...options.headers
+  };
+  const make = (modelId) => new OllamaChatModel({
+    modelId,
+    baseURL,
+    headers,
+    fetcher,
+    ...options.defaultOptions ? { defaultOptions: options.defaultOptions } : {}
+  });
+  const provider = ((modelId) => make(modelId));
+  provider.chat = make;
+  return provider;
+}
+var ollama = createOllama();
+function loadEnv(name) {
+  if (typeof process !== "undefined" && process.env) {
+    return process.env[name];
+  }
+  return void 0;
+}
+
+exports.OllamaChatModel = OllamaChatModel;
+exports.createOllama = createOllama;
+exports.ollama = ollama;
+//# sourceMappingURL=index.cjs.map
+//# sourceMappingURL=index.cjs.map

package/dist/index.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/util/ndjson.ts","../src/ollama-chat-model.ts","../src/ollama-provider.ts"],"names":["APICallError","estimateTokensFromMessages"],"mappings":";;;;;;;AAQA,gBAAuB,YACrB,IAAA,EACkB;AAClB,EAAA,MAAM,MAAA,GAAS,KAAK,SAAA,EAAU;AAC9B,EAAA,MAAM,OAAA,GAAU,IAAI,WAAA,EAAY;AAChC,EAAA,IAAI,MAAA,GAAS,EAAA;AACb,EAAA,IAAI;AACF,IAAA,OAAO,IAAA,EAAM;AACX,MAAA,MAAM,EAAE,KAAA,EAAO,IAAA,EAAK,GAAI,MAAM,OAAO,IAAA,EAAK;AAC1C,MAAA,IAAI,IAAA,EAAM;AACV,MAAA,MAAA,IAAU,QAAQ,MAAA,CAAO,KAAA,EAAO,EAAE,MAAA,EAAQ,MAAM,CAAA;AAChD,MAAA,IAAI,EAAA,GAAK,MAAA,CAAO,OAAA,CAAQ,IAAI,CAAA;AAC5B,MAAA,OAAO,OAAO,CAAA,CAAA,EAAI;AAChB,QAAA,MAAM,IAAA,GAAO,MAAA,CAAO,KAAA,CAAM,CAAA,EAAG,EAAE,EAAE,OAAA,CAAQ,KAAA,EAAO,EAAE,CAAA,CAAE,IAAA,EAAK;AACzD,QAAA,MAAA,GAAS,MAAA,CAAO,KAAA,CAAM,EAAA,GAAK,CAAC,CAAA;AAC5B,QAAA,IAAI,IAAA,CAAK,SAAS,CAAA,EAAG;AACnB,UAAA,MAAM,IAAA,CAAK,MAAM,IAAI,CAAA;AAAA,QACvB;AACA,QAAA,EAAA,GAAK,MAAA,CAAO,QAAQ,IAAI,CAAA;AAAA,MAC1B;AAAA,IACF;AACA,IAAA,MAAM,IAAA,GAAO,OAAO,IAAA,EAAK;AACzB,IAAA,IAAI,KAAK,MAAA,GAAS,CAAA,EAAG,MAAM,IAAA,CAAK,MAAM,IAAI,CAAA;AAAA,EAC5C,CAAA,SAAE;AACA,IAAA,MAAA,CAAO,WAAA,EAAY;AAAA,EACrB;AACF;;;AC2BO,IAAM,kBAAN,MAA+C;AAAA,EAC3C,QAAA,GAAW,QAAA;AAAA,EACX,OAAA;AAAA,EACQ,MAAA;AAAA,EAEjB,YAAY,MAAA,EAA+B;AACzC,IAAA,IAAA,CAAK,UAAU,MAAA,CAAO,OAAA;AACtB,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AAAA,EAChB;AAAA,EAEA,MAAM,SAAS,OAAA,EAAyD;AACtE,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,SAAA,CAAU,OAAA,EAAS,KAAK,CAAA;AAC1C,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,KAAA,CAAM,WAAA,EAAa,MAAM,OAAO,CAAA;AACvD,IAAA,MAAM,IAAA,GAAQ,MAAM,GAAA,CAAI,IAAA,EAAK;AAE7B,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,OAAA,EAAS,OAAA,IAAW,EAAA;AACtC,IAAA,MAAM,YACJ,IAAA,CAAK,OAAA,EAAS,YAAY,GAAA,CAAI,CAAC,IAAI,CAAA,MAAO;AAAA,MACxC,IAAA,EAAM,WAAA;AAAA,MACN,YAAY,oBAAA,CAAqB,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,WAAW,CAAC,CAAA;AAAA,MAClE,QAAA,EAAU,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,EAAA;AAAA,MAC/B,IAAA,EAAM,EAAA,CAAG,QAAA,EAAU,SAAA,IAAa;AAAC,KACnC,CAAE,KAAK,EAAC;AAEV,IAAA,OAAO;AAAA,MACL,IAAA;AAAA,MACA,SAAS,CAAC,GAAI,IAAA,CAAK,MAAA,GAAS,IAAI,CAAC,EAAE,IAAA,EAAM,MAAA,EAAiB,MAAM,CAAA,GAAI,EAAC,EAAI,GAAG,SAAS,CAAA;AAAA,MACrF,SAAA;AAAA,MACA,YAAA,EAAc,gBAAgB,IAAI,CAAA;AAAA,MAClC,KAAA,EAAO,SAAS,IAAI,CAAA;AAAA,MACpB,WAAA,EAAa;AAAA,KACf;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,OAAA,EAAqE;AAChF,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,SAAA,CAAU,OAAA,EAAS,IAAI,CAAA;AACzC,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,KAAA,CAAM,WAAA,EAAa,MAAM,OAAO,CAAA;AACvD,IAAA,IAAI,CAAC,IAAI,IAAA,EAAM;AACb,MAAA,MAAM,IAAIA,iBAAA,CAAa;AAAA,QACrB,OAAA,EAAS,wCAAA;AAAA,QACT,YAAY,GAAA,CAAI;AAAA,OACjB,CAAA;AAAA,IACH;AAEA,IAAA,MAAM,MAAA,GAAS,WAAA,CAAgC,GAAA,CAAI,IAAI,CAAA;AAEvD,IAAA,OAAO,IAAI,cAAA,CAAgC;AAAA,MACzC,MAAM,MAAM,UAAA,EAAY;AACtB,QAAA,IAAI,KAAA;AACJ,QAAA,IAAI,MAAA,GAAuB,SAAA;AAC3B,QAAA,IAAI,OAAA,GAAU,CAAA;AAEd,QAAA,IAAI;AACF,UAAA,WAAA,MAAiB,SAAS,MAAA,EAAQ;AAChC,YAAA,IAAI,KAAA,CAAM,SAAS,OAAA,EAAS;AAC1B,cAAA,UAAA,CAAW,OAAA,CAAQ,EAAE,IAAA,EAAM,YAAA,EAAc,WAAW,KAAA,CAAM,OAAA,CAAQ,SAAS,CAAA;AAAA,YAC7E;AACA,YAAA,IAAI,KAAA,CAAM,SAAS,UAAA,EAAY;AAC7B,cAAA,KAAA,MAAW,EAAA,IAAM,KAAA,CAAM,OAAA,CAAQ,UAAA,EAAY;AACzC,gBAAA,MAAM,KAAK,oBAAA,CAAqB,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,WAAW,OAAA,EAAS,CAAA;AAKzE,gBAAA,UAAA,CAAW,OAAA,CAAQ;AAAA,kBACjB,IAAA,EAAM,WAAA;AAAA,kBACN,UAAA,EAAY,EAAA;AAAA,kBACZ,QAAA,EAAU,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,EAAA;AAAA,kBAC/B,IAAA,EAAM,EAAA,CAAG,QAAA,EAAU,SAAA,IAAa;AAAC,iBAClC,CAAA;AAAA,cACH;AAAA,YACF;AACA,YAAA,IAAI,MAAM,IAAA,EAAM;AACd,cAAA,MAAA,GAAS,gBAAgB,KAAK,CAAA;AAC9B,cAAA,KAAA,GAAQ,SAAS,KAAK,CAAA;AAAA,YACxB;AAAA,UACF;AACA,UAAA,UAAA,CAAW,OAAA,CAAQ,EAAE,IAAA,EAAM,QAAA,EAAU,YAAA,EAAc,QAAQ,KAAA,EAAO,KAAA,IAAS,EAAC,EAAG,CAAA;AAC/E,UAAA,UAAA,CAAW,KAAA,EAAM;AAAA,QACnB,SAAS,GAAA,EAAK;AACZ,UAAA,UAAA,CAAW,QAAQ,EAAE,IAAA,EAAM,OAAA,EAAS,KAAA,EAAO,KAAK,CAAA;AAChD,UAAA,UAAA,CAAW,KAAA,EAAM;AAAA,QACnB;AAAA,MACF;AAAA,KACD,CAAA;AAAA,EACH;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAaA,aAAa,OAAA,EAAyC;AACpD,IAAA,MAAM,WAAA,GAAcC,+BAAA,CAA2B,cAAA,CAAe,OAAA,CAAQ,QAAQ,CAAC,CAAA;AAC/E,IAAA,MAAM,MAAA,GAAS,QAAQ,SAAA,IAAa,IAAA;AACpC,IAAA,MAAM,MAAA,GAAS,IAAA,CAAK,GAAA,CAAI,EAAA,EAAI,MAAM,CAAA;AAClC,IAAA,OAAO;AAAA,MACL,WAAW,WAAA,GAAc,MAAA;AAAA,MACzB,WAAW,WAAA,GAAc,MAAA;AAAA,MACzB,MAAA,EAAQ,CAAA;AAAA,MACR,MAAA,EAAQ,CAAA;AAAA,MACR,gBAAA,EAAkB;AAAA,KACpB;AAAA,EACF;AAAA,EAEQ,SAAA,CAAU,SAA2B,MAAA,EAA0C;AACrF,IAAA,MAAM,aAAA,GAAyC;AAAA,MAC7C,GAAI,IAAA,CAAK,MAAA,CAAO,cAAA,IAAkB;AAAC,KACrC;AACA,IAAA,IAAI,OAAA,CAAQ,WAAA,KAAgB,MAAA,EAAW,aAAA,CAAc,cAAc,OAAA,CAAQ,WAAA;AAC3E,IAAA,IAAI,OAAA,CAAQ,IAAA,KAAS,MAAA,EAAW,aAAA,CAAc,QAAQ,OAAA,CAAQ,IAAA;AAC9D,IAAA,IAAI,OAAA,CAAQ,SAAA,KAAc,MAAA,EAAW,aAAA,CAAc,cAAc,OAAA,CAAQ,SAAA;AACzE,IAAA,IAAI,OAAA,CAAQ,aAAA,KAAkB,MAAA,EAAW,aAAA,CAAc,OAAO,OAAA,CAAQ,aAAA;AACtE,IAAA,IAAI,OAAA,CAAQ,IAAA,KAAS,MAAA,EAAW,aAAA,CAAc,OAAO,OAAA,CAAQ,IAAA;AAE7D,IAAA,MAAM,IAAA,GAAgC;AAAA,MACpC,OAAO,IAAA,CAAK,OAAA;AAAA,MACZ,QAAA,EAAU,OAAA,CAAQ,QAAA,CAAS,GAAA,CAAI,eAAe,CAAA;AAAA,MAC9C;AAAA,KACF;AACA,IAAA,IAAI,OAAO,IAAA,CAAK,aAAa,EAAE,MAAA,GAAS,CAAA,OAAQ,OAAA,GAAU,aAAA;AAC1D,IAAA,IAAI,OAAA,CAAQ,OAAO,MAAA,EAAQ;AACzB,MAAA,IAAA,CAAK,KAAA,GAAQ,OAAA,CAAQ,KAAA,CAAM,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,QACrC,IAAA,EAAM,UAAA;AAAA,QACN,QAAA,EAAU;AAAA,UACR,MAAM,CAAA,CAAE,IAAA;AAAA,UACR,GAAI,EAAE,WAAA,KAAgB,MAAA,GAAY,EAAE,WAAA,EAAa,CAAA,CAAE,WAAA,EAAY,GAAI,EAAC;AAAA,UACpE,YAAY,CAAA,CAAE;AAAA;AAChB,OACF,CAAE,CAAA;AAAA,IACJ;AACA,IAAA,IAAI,QAAQ,eAAA,EAAiB,MAAA,CAAO,MAAA,CAAO,IAAA,EAAM,QAAQ,eAAe,CAAA;AACxE,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,MAAc,KAAA,CAAM,IAAA,EAAc,IAAA,EAAe,OAAA,EAA8C;AAC7F,IAAA,MAAM,MAAM,CAAA,EAAG,IAAA,CAAK,MAAA,CAAO,OAAO,GAAG,IAAI,CAAA,CAAA;AACzC,IAAA,MAAM,OAAA,GAAU,EAAE,GAAG,IAAA,CAAK,OAAO,OAAA,EAAS,GAAG,QAAQ,OAAA,EAAQ;AAC7D,IAAA,MAAM,IAAA,GAAoB;AAAA,MACxB,MAAA,EAAQ,MAAA;AAAA,MACR,OAAA;AAAA,MACA,IAAA,EAAM,IAAA,CAAK,SAAA,CAAU,IAAI;AAAA,KAC3B;AACA,IAAA,IAAI,OAAA,CAAQ,WAAA,EAAa,IAAA,CAAK,MAAA,GAAS,OAAA,CAAQ,WAAA;AAE/C,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,MAAA,CAAO,OAAA,CAAQ,KAAK,IAAI,CAAA;AAC/C,IAAA,IAAI,CAAC,IAAI,EAAA,EAAI;AACX,MAAA,MAAM,OAAO,MAAM,GAAA,CAAI,MAAK,CAAE,KAAA,CAAM,MAAM,EAAE,CAAA;AAC5C,MAAA,MAAM,IAAID,iBAAA,CAAa;AAAA,QACrB,SAAS,CAAA,kBAAA,EAAqB,GAAA,CAAI,MAAM,CAAA,CAAA,EAAI,IAAI,UAAU,CAAA,CAAA;AAAA,QAC1D,GAAA;AAAA,QACA,YAAY,GAAA,CAAI,MAAA;AAAA,QAChB,YAAA,EAAc;AAAA,OACf,CAAA;AAAA,IACH;AACA,IAAA,OAAO,GAAA;AAAA,EACT;AACF;AAEA,SAAS,gBAAgB,CAAA,EAA+B;AACtD,EAAA,QAAQ,EAAE,IAAA;AAAM,IACd,KAAK,QAAA;AAAA,IACL,KAAK,MAAA,EAAQ;AACX,MAAA,MAAM,OAAO,CAAA,CAAE,OAAA,CACZ,MAAA,CAAO,CAAC,MAAM,CAAA,CAAE,IAAA,KAAS,MAAM,CAAA,CAC/B,IAAI,CAAC,CAAA,KAAO,EAAuB,IAAI,CAAA,CACvC,KAAK,EAAE,CAAA;AACV,MAAA,MAAM,MAAA,GAAS,CAAA,CAAE,OAAA,CACd,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,IAAA,KAAS,OAAO,CAAA,CAChC,GAAA,CAAI,CAAC,CAAA,KAAM;AACV,QAAA,MAAM,MAAO,CAAA,CAA2C,KAAA;AACxD,QAAA,IAAI,OAAO,GAAA,KAAQ,QAAA,SAAiB,GAAA,CAAI,OAAA,CAAQ,uBAAuB,EAAE,CAAA;AACzE,QAAA,IAAI,GAAA,YAAe,GAAA,EAAK,OAAO,GAAA,CAAI,QAAA,EAAS;AAC5C,QAAA,OAAO,cAAc,GAAG,CAAA;AAAA,MAC1B,CAAC,CAAA;AACH,MAAA,MAAM,MAA+B,EAAE,IAAA,EAAM,CAAA,CAAE,IAAA,EAAM,SAAS,IAAA,EAAK;AACnE,MAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,EAAG,GAAA,CAAI,MAAA,GAAS,MAAA;AACpC,MAAA,OAAO,GAAA;AAAA,IACT;AAAA,IACA,KAAK,WAAA,EAAa;AAChB,MAAA,MAAM,OAAO,CAAA,CAAE,OAAA,CACZ,MAAA,CAAO,CAAC,MAAM,CAAA,CAAE,IAAA,KAAS,MAAM,CAAA,CAC/B,IAAI,CAAC,CAAA,KAAO,EAAuB,IAAI,CAAA,CACvC,KAAK,EAAE,CAAA;AACV,MAAA,MAAM,SAAA,GAAY,EAAE,OAAA,CAAQ,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,SAAS,WAAW,CAAA;AAChE,MAAA,MAAM,GAAA,GAA+B,EAAE,IAAA,EAAM,WAAA,EAAa,SAAS,IAAA,EAAK;AACxE,MAAA,IAAI,SAAA,CAAU,SAAS,CAAA,EAAG;AACxB,QAAA,GAAA,CAAI,UAAA,GAAa,SAAA,CAAU,GAAA,CAAI,CAAC,EAAA,MAAQ;AAAA,UACtC,QAAA,EAAU;AAAA,YACR,MAAO,EAAA,CAAoB,QAAA;AAAA,YAC3B,SAAA,EAAY,EAAA,CAAoB,IAAA,IAAQ;AAAC;AAC3C,SACF,CAAE,CAAA;AAAA,MACJ;AACA,MAAA,OAAO,GAAA;AAAA,IACT;AAAA,IACA,KAAK,MAAA,EAAQ;AACX,MAAA,MAAM,KAAA,GAAQ,CAAA,CAAE,OAAA,CAAQ,CAAC,CAAA;AACzB,MAAA,IAAI,CAAC,KAAA,IAAS,KAAA,CAAM,IAAA,KAAS,aAAA,EAAe;AAC1C,QAAA,OAAO,EAAE,IAAA,EAAM,MAAA,EAAQ,OAAA,EAAS,EAAA,EAAG;AAAA,MACrC;AACA,MAAA,OAAO;AAAA,QACL,IAAA,EAAM,MAAA;AAAA,QACN,OAAA,EAAS,OAAO,KAAA,CAAM,MAAA,KAAW,QAAA,GAAW,MAAM,MAAA,GAAS,IAAA,CAAK,SAAA,CAAU,KAAA,CAAM,MAAM;AAAA,OACxF;AAAA,IACF;AAAA;AAEJ;AAQA,SAAS,oBAAA,CAAqB,MAAc,KAAA,EAAuB;AACjE,EAAA,OAAO,CAAA,OAAA,EAAU,IAAI,CAAA,CAAA,EAAI,KAAK,IAAI,IAAA,CAAK,MAAA,EAAO,CAAE,QAAA,CAAS,EAAE,CAAA,CAAE,KAAA,CAAM,CAAA,EAAG,CAAC,CAAC,CAAA,CAAA;AAC1E;AAEA,SAAS,eACP,QAAA,EACkD;AAClD,EAAA,OAAO,QAAA;AACT;AAEA,SAAS,cAAc,GAAA,EAAyB;AAC9C,EAAA,IAAI,CAAA,GAAI,EAAA;AACR,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,GAAA,CAAI,UAAA,EAAY,CAAA,EAAA,EAAK,CAAA,IAAK,MAAA,CAAO,YAAA,CAAa,GAAA,CAAI,CAAC,CAAW,CAAA;AAClF,EAAA,OAAO,OAAO,IAAA,KAAS,WAAA,GAAc,IAAA,CAAK,CAAC,CAAA,GAAI,MAAA,CAAO,IAAA,CAAK,CAAA,EAAG,QAAQ,CAAA,CAAE,QAAA,CAAS,QAAQ,CAAA;AAC3F;AAEA,SAAS,gBAAgB,IAAA,EAAwC;AAC/D,EAAA,IAAI,IAAA,CAAK,OAAA,EAAS,UAAA,EAAY,MAAA,EAAQ,OAAO,YAAA;AAC7C,EAAA,IAAI,IAAA,CAAK,WAAA,KAAgB,MAAA,IAAU,IAAA,CAAK,MAAM,OAAO,MAAA;AACrD,EAAA,IAAI,IAAA,CAAK,WAAA,KAAgB,QAAA,EAAU,OAAO,QAAA;AAC1C,EAAA,OAAO,SAAA;AACT;AAEA,SAAS,SAAS,IAAA,EAAsC;AACtD,EAAA,MAAM,MAAkB,EAAC;AACzB,EAAA,IAAI,IAAA,CAAK,iBAAA,KAAsB,MAAA,EAAW,GAAA,CAAI,eAAe,IAAA,CAAK,iBAAA;AAClE,EAAA,IAAI,IAAA,CAAK,UAAA,KAAe,MAAA,EAAW,GAAA,CAAI,mBAAmB,IAAA,CAAK,UAAA;AAC/D,EAAA,IAAI,IAAA,CAAK,iBAAA,KAAsB,MAAA,IAAa,IAAA,CAAK,eAAe,MAAA,EAAW;AACzE,IAAA,GAAA,CAAI,WAAA,GAAc,IAAA,CAAK,iBAAA,GAAoB,IAAA,CAAK,UAAA;AAAA,EAClD;AACA,EAAA,OAAO,GAAA;AACT;;;ACtRO,SAAS,YAAA,CAAa,OAAA,GAAiC,EAAC,EAAmB;AAChF,EAAA,MAAM,OAAA,GAAA,CACJ,QAAQ,OAAA,IACR,OAAA,CAAQ,iBAAiB,CAAA,IACzB,wBAAA,EACA,OAAA,CAAQ,MAAA,EAAQ,EAAE,CAAA;AACpB,EAAA,MAAM,OAAA,GAAU,OAAA,CAAQ,KAAA,IAAS,UAAA,CAAW,KAAA;AAC5C,EAAA,MAAM,OAAA,GAAkC;AAAA,IACtC,cAAA,EAAgB,kBAAA;AAAA,IAChB,GAAG,OAAA,CAAQ;AAAA,GACb;AAEA,EAAA,MAAM,IAAA,GAAO,CAAC,OAAA,KACZ,IAAI,eAAA,CAAgB;AAAA,IAClB,OAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAA;AAAA,IACA,GAAI,QAAQ,cAAA,GAAiB,EAAE,gBAAgB,OAAA,CAAQ,cAAA,KAAmB;AAAC,GAC5E,CAAA;AAEH,EAAA,MAAM,QAAA,IAAY,CAAC,OAAA,KAA+B,IAAA,CAAK,OAAO,CAAA,CAAA;AAC9D,EAAA,QAAA,CAAS,IAAA,GAAO,IAAA;AAChB,EAAA,OAAO,QAAA;AACT;AAQO,IAAM,SAAyB,YAAA;AAEtC,SAAS,QAAQ,IAAA,EAAkC;AACjD,EAAA,IAAI,OAAO,OAAA,KAAY,WAAA,IAAe,OAAA,CAAQ,GAAA,EAAK;AACjD,IAAA,OAAO,OAAA,CAAQ,IAAI,IAAI,CAAA;AAAA,EACzB;AACA,EAAA,OAAO,MAAA;AACT","file":"index.cjs","sourcesContent":["/**\n * Parse a `ReadableStream<Uint8Array>` of newline-delimited JSON into an\n * async iterator of parsed records. Tolerates `\\r\\n`, blank lines, and\n * partial chunks.\n *\n * Ollama's `/api/chat?stream=true` emits one JSON object per line —\n * unlike OpenAI / Anthropic which use SSE (`data: {...}\\n\\n`).\n */\nexport async function* parseNDJSON<T = unknown>(\n  body: ReadableStream<Uint8Array>,\n): AsyncIterable<T> {\n  const reader = body.getReader();\n  const decoder = new TextDecoder();\n  let buffer = '';\n  try {\n    while (true) {\n      const { value, done } = await reader.read();\n      if (done) break;\n      buffer += decoder.decode(value, { stream: true });\n      let nl = buffer.indexOf('\\n');\n      while (nl !== -1) {\n        const line = buffer.slice(0, nl).replace(/\\r$/, '').trim();\n        buffer = buffer.slice(nl + 1);\n        if (line.length > 0) {\n          yield JSON.parse(line) as T;\n        }\n        nl = buffer.indexOf('\\n');\n      }\n    }\n    const tail = buffer.trim();\n    if (tail.length > 0) yield JSON.parse(tail) as T;\n  } finally {\n    reader.releaseLock();\n  }\n}\n","import {\n  APICallError,\n  type CostEstimate,\n  estimateTokensFromMessages,\n  type FinishReason,\n  type LanguageModel,\n  type ModelCallOptions,\n  type ModelGenerateResult,\n  type ModelStreamPart,\n  type NormalizedMessage,\n  type TokenUsage,\n  type ToolCallPart,\n} from '@ziro-agent/core';\nimport { parseNDJSON } from './util/ndjson.js';\n\n/**\n * Common Ollama model ids. Ollama is open-weight and the catalogue\n * grows weekly, so we leave the type open via `(string & {})` — pulling\n * a model is `ollama pull <name>` and any tag works.\n *\n * The enumerated entries are the most-pulled tool-capable models on\n * ollama.com as of v0.1.9. They are deliberately conservative — listing\n * every quant would explode the union without changing runtime behaviour.\n */\nexport type OllamaChatModelId =\n  | 'llama3.1'\n  | 'llama3.1:8b'\n  | 'llama3.1:70b'\n  | 'llama3.2'\n  | 'llama3.2:3b'\n  | 'llama3.3'\n  | 'qwen2.5'\n  | 'qwen2.5:7b'\n  | 'qwen2.5:14b'\n  | 'qwen2.5:32b'\n  | 'qwen2.5-coder'\n  | 'mistral'\n  | 'mistral-nemo'\n  | 'mixtral'\n  | 'gemma3'\n  | 'phi4'\n  | (string & {});\n\ninterface OllamaChatModelConfig {\n  modelId: OllamaChatModelId;\n  baseURL: string;\n  headers: Record<string, string>;\n  fetcher: typeof fetch;\n  /** Forwarded into Ollama's per-request `options` block. */\n  defaultOptions?: Record<string, unknown>;\n}\n\n/**\n * `LanguageModel` adapter for the local Ollama HTTP API\n * (`http://localhost:11434/api/chat`). Talks the native Ollama protocol\n * (NDJSON streaming, function-calling shape) — not the OpenAI-compat\n * shim, which loses tool-call fidelity on several models.\n *\n * Sovereign-pillar primitive (RFC 0004 §v0.1.9): an open-weight default\n * provider so the SDK is not assumed-cloud-only.\n */\nexport class OllamaChatModel implements LanguageModel {\n  readonly provider = 'ollama';\n  readonly modelId: string;\n  private readonly config: OllamaChatModelConfig;\n\n  constructor(config: OllamaChatModelConfig) {\n    this.modelId = config.modelId;\n    this.config = config;\n  }\n\n  async generate(options: ModelCallOptions): Promise<ModelGenerateResult> {\n    const body = this.buildBody(options, false);\n    const res = await this.fetch('/api/chat', body, options);\n    const json = (await res.json()) as OllamaChatResponse;\n\n    const text = json.message?.content ?? '';\n    const toolCalls: ToolCallPart[] =\n      json.message?.tool_calls?.map((tc, i) => ({\n        type: 'tool-call' as const,\n        toolCallId: synthesizeToolCallId(tc.function?.name ?? 'unknown', i),\n        toolName: tc.function?.name ?? '',\n        args: tc.function?.arguments ?? {},\n      })) ?? [];\n\n    return {\n      text,\n      content: [...(text.length > 0 ? [{ type: 'text' as const, text }] : []), ...toolCalls],\n      toolCalls,\n      finishReason: mapFinishReason(json),\n      usage: mapUsage(json),\n      rawResponse: json,\n    };\n  }\n\n  async stream(options: ModelCallOptions): Promise<ReadableStream<ModelStreamPart>> {\n    const body = this.buildBody(options, true);\n    const res = await this.fetch('/api/chat', body, options);\n    if (!res.body) {\n      throw new APICallError({\n        message: 'Ollama streaming response has no body.',\n        statusCode: res.status,\n      });\n    }\n\n    const ndjson = parseNDJSON<OllamaChatResponse>(res.body);\n\n    return new ReadableStream<ModelStreamPart>({\n      async start(controller) {\n        let usage: TokenUsage | undefined;\n        let finish: FinishReason = 'unknown';\n        let toolIdx = 0;\n\n        try {\n          for await (const chunk of ndjson) {\n            if (chunk.message?.content) {\n              controller.enqueue({ type: 'text-delta', textDelta: chunk.message.content });\n            }\n            if (chunk.message?.tool_calls) {\n              for (const tc of chunk.message.tool_calls) {\n                const id = synthesizeToolCallId(tc.function?.name ?? 'unknown', toolIdx++);\n                // Ollama emits each tool call complete (no incremental\n                // arg streaming), so we forward the full call directly\n                // — matching the OpenAI provider's terminal `tool-call`\n                // event shape.\n                controller.enqueue({\n                  type: 'tool-call',\n                  toolCallId: id,\n                  toolName: tc.function?.name ?? '',\n                  args: tc.function?.arguments ?? {},\n                });\n              }\n            }\n            if (chunk.done) {\n              finish = mapFinishReason(chunk);\n              usage = mapUsage(chunk);\n            }\n          }\n          controller.enqueue({ type: 'finish', finishReason: finish, usage: usage ?? {} });\n          controller.close();\n        } catch (err) {\n          controller.enqueue({ type: 'error', error: err });\n          controller.close();\n        }\n      },\n    });\n  }\n\n  /**\n   * Local models are free at runtime — their cost is electricity / GPU\n   * time, not per-token billing. We report `pricingAvailable: true`\n   * with `$0` so `Budget Guard.maxUsd` constraints simply never trip\n   * for Ollama runs (instead of being silently disabled, which would\n   * surprise callers migrating from OpenAI).\n   *\n   * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets\n   * keep working as documented — those are the meaningful limits when\n   * GPU time is the bottleneck.\n   */\n  estimateCost(options: ModelCallOptions): CostEstimate {\n    const inputTokens = estimateTokensFromMessages(asChatMessages(options.messages));\n    const maxOut = options.maxTokens ?? 8_192;\n    const minOut = Math.min(16, maxOut);\n    return {\n      minTokens: inputTokens + minOut,\n      maxTokens: inputTokens + maxOut,\n      minUsd: 0,\n      maxUsd: 0,\n      pricingAvailable: true,\n    };\n  }\n\n  private buildBody(options: ModelCallOptions, stream: boolean): Record<string, unknown> {\n    const ollamaOptions: Record<string, unknown> = {\n      ...(this.config.defaultOptions ?? {}),\n    };\n    if (options.temperature !== undefined) ollamaOptions.temperature = options.temperature;\n    if (options.topP !== undefined) ollamaOptions.top_p = options.topP;\n    if (options.maxTokens !== undefined) ollamaOptions.num_predict = options.maxTokens;\n    if (options.stopSequences !== undefined) ollamaOptions.stop = options.stopSequences;\n    if (options.seed !== undefined) ollamaOptions.seed = options.seed;\n\n    const body: Record<string, unknown> = {\n      model: this.modelId,\n      messages: options.messages.map(toOllamaMessage),\n      stream,\n    };\n    if (Object.keys(ollamaOptions).length > 0) body.options = ollamaOptions;\n    if (options.tools?.length) {\n      body.tools = options.tools.map((t) => ({\n        type: 'function',\n        function: {\n          name: t.name,\n          ...(t.description !== undefined ? { description: t.description } : {}),\n          parameters: t.parameters,\n        },\n      }));\n    }\n    if (options.providerOptions) Object.assign(body, options.providerOptions);\n    return body;\n  }\n\n  private async fetch(path: string, body: unknown, options: ModelCallOptions): Promise<Response> {\n    const url = `${this.config.baseURL}${path}`;\n    const headers = { ...this.config.headers, ...options.headers };\n    const init: RequestInit = {\n      method: 'POST',\n      headers,\n      body: JSON.stringify(body),\n    };\n    if (options.abortSignal) init.signal = options.abortSignal;\n\n    const res = await this.config.fetcher(url, init);\n    if (!res.ok) {\n      const text = await res.text().catch(() => '');\n      throw new APICallError({\n        message: `Ollama API error: ${res.status} ${res.statusText}`,\n        url,\n        statusCode: res.status,\n        responseBody: text,\n      });\n    }\n    return res;\n  }\n}\n\nfunction toOllamaMessage(m: NormalizedMessage): unknown {\n  switch (m.role) {\n    case 'system':\n    case 'user': {\n      const text = m.content\n        .filter((p) => p.type === 'text')\n        .map((p) => (p as { text: string }).text)\n        .join('');\n      const images = m.content\n        .filter((p) => p.type === 'image')\n        .map((p) => {\n          const img = (p as { image: string | URL | Uint8Array }).image;\n          if (typeof img === 'string') return img.replace(/^data:[^;]+;base64,/, '');\n          if (img instanceof URL) return img.toString();\n          return uint8ToBase64(img);\n        });\n      const out: Record<string, unknown> = { role: m.role, content: text };\n      if (images.length > 0) out.images = images;\n      return out;\n    }\n    case 'assistant': {\n      const text = m.content\n        .filter((p) => p.type === 'text')\n        .map((p) => (p as { text: string }).text)\n        .join('');\n      const toolCalls = m.content.filter((p) => p.type === 'tool-call');\n      const out: Record<string, unknown> = { role: 'assistant', content: text };\n      if (toolCalls.length > 0) {\n        out.tool_calls = toolCalls.map((tc) => ({\n          function: {\n            name: (tc as ToolCallPart).toolName,\n            arguments: (tc as ToolCallPart).args ?? {},\n          },\n        }));\n      }\n      return out;\n    }\n    case 'tool': {\n      const first = m.content[0];\n      if (!first || first.type !== 'tool-result') {\n        return { role: 'tool', content: '' };\n      }\n      return {\n        role: 'tool',\n        content: typeof first.result === 'string' ? first.result : JSON.stringify(first.result),\n      };\n    }\n  }\n}\n\n/**\n * Ollama's `tool_calls[]` entries don't carry a stable id — they're\n * just `{ function: { name, arguments } }`. We synthesise an id so the\n * downstream agent loop can match tool results back to calls without\n * collisions across a batch.\n */\nfunction synthesizeToolCallId(name: string, index: number): string {\n  return `ollama_${name}_${index}_${Math.random().toString(36).slice(2, 8)}`;\n}\n\nfunction asChatMessages(\n  messages: NormalizedMessage[],\n): Parameters<typeof estimateTokensFromMessages>[0] {\n  return messages as unknown as Parameters<typeof estimateTokensFromMessages>[0];\n}\n\nfunction uint8ToBase64(arr: Uint8Array): string {\n  let s = '';\n  for (let i = 0; i < arr.byteLength; i++) s += String.fromCharCode(arr[i] as number);\n  return typeof btoa !== 'undefined' ? btoa(s) : Buffer.from(s, 'binary').toString('base64');\n}\n\nfunction mapFinishReason(json: OllamaChatResponse): FinishReason {\n  if (json.message?.tool_calls?.length) return 'tool-calls';\n  if (json.done_reason === 'stop' || json.done) return 'stop';\n  if (json.done_reason === 'length') return 'length';\n  return 'unknown';\n}\n\nfunction mapUsage(json: OllamaChatResponse): TokenUsage {\n  const out: TokenUsage = {};\n  if (json.prompt_eval_count !== undefined) out.promptTokens = json.prompt_eval_count;\n  if (json.eval_count !== undefined) out.completionTokens = json.eval_count;\n  if (json.prompt_eval_count !== undefined && json.eval_count !== undefined) {\n    out.totalTokens = json.prompt_eval_count + json.eval_count;\n  }\n  return out;\n}\n\ninterface OllamaChatResponse {\n  model?: string;\n  created_at?: string;\n  message?: {\n    role?: string;\n    content?: string;\n    tool_calls?: Array<{\n      function?: {\n        name?: string;\n        arguments?: Record<string, unknown>;\n      };\n    }>;\n  };\n  done?: boolean;\n  done_reason?: 'stop' | 'length' | 'load' | string;\n  total_duration?: number;\n  load_duration?: number;\n  prompt_eval_count?: number;\n  prompt_eval_duration?: number;\n  eval_count?: number;\n  eval_duration?: number;\n}\n","import type { LanguageModel } from '@ziro-agent/core';\nimport { OllamaChatModel, type OllamaChatModelId } from './ollama-chat-model.js';\n\nexport interface OllamaProviderOptions {\n  /**\n   * Base URL of the Ollama daemon. Defaults to\n   * `process.env.OLLAMA_BASE_URL` then `http://localhost:11434`.\n   *\n   * For remote / containerised Ollama (e.g. compose stack, Coolify),\n   * set this to `http://ollama:11434` or whatever your network exposes.\n   */\n  baseURL?: string;\n  /** Extra default headers attached to every request. */\n  headers?: Record<string, string>;\n  /** Custom `fetch`. Defaults to `globalThis.fetch`. */\n  fetch?: typeof fetch;\n  /**\n   * Default Ollama-native sampling options (`num_ctx`, `mirostat`,\n   * `repeat_penalty`, …) merged into every request's `options` block.\n   * Per-call overrides via `ModelCallOptions.providerOptions.options`\n   * still win.\n   */\n  defaultOptions?: Record<string, unknown>;\n}\n\nexport interface OllamaProvider {\n  (modelId: OllamaChatModelId): LanguageModel;\n  chat(modelId: OllamaChatModelId): LanguageModel;\n}\n\n/**\n * Build an {@link OllamaProvider} bound to a specific Ollama daemon.\n * Mirrors `createOpenAI` / `createAnthropic` so swapping providers in\n * `createAgent({ model })` is a one-line change.\n */\nexport function createOllama(options: OllamaProviderOptions = {}): OllamaProvider {\n  const baseURL = (\n    options.baseURL ??\n    loadEnv('OLLAMA_BASE_URL') ??\n    'http://localhost:11434'\n  ).replace(/\\/+$/, '');\n  const fetcher = options.fetch ?? globalThis.fetch;\n  const headers: Record<string, string> = {\n    'Content-Type': 'application/json',\n    ...options.headers,\n  };\n\n  const make = (modelId: OllamaChatModelId): LanguageModel =>\n    new OllamaChatModel({\n      modelId,\n      baseURL,\n      headers,\n      fetcher,\n      ...(options.defaultOptions ? { defaultOptions: options.defaultOptions } : {}),\n    });\n\n  const provider = ((modelId: OllamaChatModelId) => make(modelId)) as OllamaProvider;\n  provider.chat = make;\n  return provider;\n}\n\n/**\n * Default singleton — connects to `localhost:11434` (or\n * `OLLAMA_BASE_URL` if set). Suitable for the typical\n * `ollama serve` setup; build a custom provider with `createOllama()`\n * for remote / authenticated daemons.\n */\nexport const ollama: OllamaProvider = createOllama();\n\nfunction loadEnv(name: string): string | undefined {\n  if (typeof process !== 'undefined' && process.env) {\n    return process.env[name];\n  }\n  return undefined;\n}\n"]}

package/dist/index.d.cts

@@ -0,0 +1,92 @@
+import { LanguageModel, ModelCallOptions, ModelGenerateResult, ModelStreamPart, CostEstimate } from '@ziro-agent/core';
+
+/**
+ * Common Ollama model ids. Ollama is open-weight and the catalogue
+ * grows weekly, so we leave the type open via `(string & {})` — pulling
+ * a model is `ollama pull <name>` and any tag works.
+ *
+ * The enumerated entries are the most-pulled tool-capable models on
+ * ollama.com as of v0.1.9. They are deliberately conservative — listing
+ * every quant would explode the union without changing runtime behaviour.
+ */
+type OllamaChatModelId = 'llama3.1' | 'llama3.1:8b' | 'llama3.1:70b' | 'llama3.2' | 'llama3.2:3b' | 'llama3.3' | 'qwen2.5' | 'qwen2.5:7b' | 'qwen2.5:14b' | 'qwen2.5:32b' | 'qwen2.5-coder' | 'mistral' | 'mistral-nemo' | 'mixtral' | 'gemma3' | 'phi4' | (string & {});
+interface OllamaChatModelConfig {
+    modelId: OllamaChatModelId;
+    baseURL: string;
+    headers: Record<string, string>;
+    fetcher: typeof fetch;
+    /** Forwarded into Ollama's per-request `options` block. */
+    defaultOptions?: Record<string, unknown>;
+}
+/**
+ * `LanguageModel` adapter for the local Ollama HTTP API
+ * (`http://localhost:11434/api/chat`). Talks the native Ollama protocol
+ * (NDJSON streaming, function-calling shape) — not the OpenAI-compat
+ * shim, which loses tool-call fidelity on several models.
+ *
+ * Sovereign-pillar primitive (RFC 0004 §v0.1.9): an open-weight default
+ * provider so the SDK is not assumed-cloud-only.
+ */
+declare class OllamaChatModel implements LanguageModel {
+    readonly provider = "ollama";
+    readonly modelId: string;
+    private readonly config;
+    constructor(config: OllamaChatModelConfig);
+    generate(options: ModelCallOptions): Promise<ModelGenerateResult>;
+    stream(options: ModelCallOptions): Promise<ReadableStream<ModelStreamPart>>;
+    /**
+     * Local models are free at runtime — their cost is electricity / GPU
+     * time, not per-token billing. We report `pricingAvailable: true`
+     * with `$0` so `Budget Guard.maxUsd` constraints simply never trip
+     * for Ollama runs (instead of being silently disabled, which would
+     * surprise callers migrating from OpenAI).
+     *
+     * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets
+     * keep working as documented — those are the meaningful limits when
+     * GPU time is the bottleneck.
+     */
+    estimateCost(options: ModelCallOptions): CostEstimate;
+    private buildBody;
+    private fetch;
+}
+
+interface OllamaProviderOptions {
+    /**
+     * Base URL of the Ollama daemon. Defaults to
+     * `process.env.OLLAMA_BASE_URL` then `http://localhost:11434`.
+     *
+     * For remote / containerised Ollama (e.g. compose stack, Coolify),
+     * set this to `http://ollama:11434` or whatever your network exposes.
+     */
+    baseURL?: string;
+    /** Extra default headers attached to every request. */
+    headers?: Record<string, string>;
+    /** Custom `fetch`. Defaults to `globalThis.fetch`. */
+    fetch?: typeof fetch;
+    /**
+     * Default Ollama-native sampling options (`num_ctx`, `mirostat`,
+     * `repeat_penalty`, …) merged into every request's `options` block.
+     * Per-call overrides via `ModelCallOptions.providerOptions.options`
+     * still win.
+     */
+    defaultOptions?: Record<string, unknown>;
+}
+interface OllamaProvider {
+    (modelId: OllamaChatModelId): LanguageModel;
+    chat(modelId: OllamaChatModelId): LanguageModel;
+}
+/**
+ * Build an {@link OllamaProvider} bound to a specific Ollama daemon.
+ * Mirrors `createOpenAI` / `createAnthropic` so swapping providers in
+ * `createAgent({ model })` is a one-line change.
+ */
+declare function createOllama(options?: OllamaProviderOptions): OllamaProvider;
+/**
+ * Default singleton — connects to `localhost:11434` (or
+ * `OLLAMA_BASE_URL` if set). Suitable for the typical
+ * `ollama serve` setup; build a custom provider with `createOllama()`
+ * for remote / authenticated daemons.
+ */
+declare const ollama: OllamaProvider;
+
+export { OllamaChatModel, type OllamaChatModelId, type OllamaProviderOptions, createOllama, ollama };

package/dist/index.d.ts

@@ -0,0 +1,92 @@
+import { LanguageModel, ModelCallOptions, ModelGenerateResult, ModelStreamPart, CostEstimate } from '@ziro-agent/core';
+
+/**
+ * Common Ollama model ids. Ollama is open-weight and the catalogue
+ * grows weekly, so we leave the type open via `(string & {})` — pulling
+ * a model is `ollama pull <name>` and any tag works.
+ *
+ * The enumerated entries are the most-pulled tool-capable models on
+ * ollama.com as of v0.1.9. They are deliberately conservative — listing
+ * every quant would explode the union without changing runtime behaviour.
+ */
+type OllamaChatModelId = 'llama3.1' | 'llama3.1:8b' | 'llama3.1:70b' | 'llama3.2' | 'llama3.2:3b' | 'llama3.3' | 'qwen2.5' | 'qwen2.5:7b' | 'qwen2.5:14b' | 'qwen2.5:32b' | 'qwen2.5-coder' | 'mistral' | 'mistral-nemo' | 'mixtral' | 'gemma3' | 'phi4' | (string & {});
+interface OllamaChatModelConfig {
+    modelId: OllamaChatModelId;
+    baseURL: string;
+    headers: Record<string, string>;
+    fetcher: typeof fetch;
+    /** Forwarded into Ollama's per-request `options` block. */
+    defaultOptions?: Record<string, unknown>;
+}
+/**
+ * `LanguageModel` adapter for the local Ollama HTTP API
+ * (`http://localhost:11434/api/chat`). Talks the native Ollama protocol
+ * (NDJSON streaming, function-calling shape) — not the OpenAI-compat
+ * shim, which loses tool-call fidelity on several models.
+ *
+ * Sovereign-pillar primitive (RFC 0004 §v0.1.9): an open-weight default
+ * provider so the SDK is not assumed-cloud-only.
+ */
+declare class OllamaChatModel implements LanguageModel {
+    readonly provider = "ollama";
+    readonly modelId: string;
+    private readonly config;
+    constructor(config: OllamaChatModelConfig);
+    generate(options: ModelCallOptions): Promise<ModelGenerateResult>;
+    stream(options: ModelCallOptions): Promise<ReadableStream<ModelStreamPart>>;
+    /**
+     * Local models are free at runtime — their cost is electricity / GPU
+     * time, not per-token billing. We report `pricingAvailable: true`
+     * with `$0` so `Budget Guard.maxUsd` constraints simply never trip
+     * for Ollama runs (instead of being silently disabled, which would
+     * surprise callers migrating from OpenAI).
+     *
+     * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets
+     * keep working as documented — those are the meaningful limits when
+     * GPU time is the bottleneck.
+     */
+    estimateCost(options: ModelCallOptions): CostEstimate;
+    private buildBody;
+    private fetch;
+}
+
+interface OllamaProviderOptions {
+    /**
+     * Base URL of the Ollama daemon. Defaults to
+     * `process.env.OLLAMA_BASE_URL` then `http://localhost:11434`.
+     *
+     * For remote / containerised Ollama (e.g. compose stack, Coolify),
+     * set this to `http://ollama:11434` or whatever your network exposes.
+     */
+    baseURL?: string;
+    /** Extra default headers attached to every request. */
+    headers?: Record<string, string>;
+    /** Custom `fetch`. Defaults to `globalThis.fetch`. */
+    fetch?: typeof fetch;
+    /**
+     * Default Ollama-native sampling options (`num_ctx`, `mirostat`,
+     * `repeat_penalty`, …) merged into every request's `options` block.
+     * Per-call overrides via `ModelCallOptions.providerOptions.options`
+     * still win.
+     */
+    defaultOptions?: Record<string, unknown>;
+}
+interface OllamaProvider {
+    (modelId: OllamaChatModelId): LanguageModel;
+    chat(modelId: OllamaChatModelId): LanguageModel;
+}
+/**
+ * Build an {@link OllamaProvider} bound to a specific Ollama daemon.
+ * Mirrors `createOpenAI` / `createAnthropic` so swapping providers in
+ * `createAgent({ model })` is a one-line change.
+ */
+declare function createOllama(options?: OllamaProviderOptions): OllamaProvider;
+/**
+ * Default singleton — connects to `localhost:11434` (or
+ * `OLLAMA_BASE_URL` if set). Suitable for the typical
+ * `ollama serve` setup; build a custom provider with `createOllama()`
+ * for remote / authenticated daemons.
+ */
+declare const ollama: OllamaProvider;
+
+export { OllamaChatModel, type OllamaChatModelId, type OllamaProviderOptions, createOllama, ollama };

package/dist/index.js

@@ -0,0 +1,276 @@
+import { APICallError, estimateTokensFromMessages } from '@ziro-agent/core';
+
+// src/ollama-chat-model.ts
+
+// src/util/ndjson.ts
+async function* parseNDJSON(body) {
+  const reader = body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+  try {
+    while (true) {
+      const { value, done } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      let nl = buffer.indexOf("\n");
+      while (nl !== -1) {
+        const line = buffer.slice(0, nl).replace(/\r$/, "").trim();
+        buffer = buffer.slice(nl + 1);
+        if (line.length > 0) {
+          yield JSON.parse(line);
+        }
+        nl = buffer.indexOf("\n");
+      }
+    }
+    const tail = buffer.trim();
+    if (tail.length > 0) yield JSON.parse(tail);
+  } finally {
+    reader.releaseLock();
+  }
+}
+
+// src/ollama-chat-model.ts
+var OllamaChatModel = class {
+  provider = "ollama";
+  modelId;
+  config;
+  constructor(config) {
+    this.modelId = config.modelId;
+    this.config = config;
+  }
+  async generate(options) {
+    const body = this.buildBody(options, false);
+    const res = await this.fetch("/api/chat", body, options);
+    const json = await res.json();
+    const text = json.message?.content ?? "";
+    const toolCalls = json.message?.tool_calls?.map((tc, i) => ({
+      type: "tool-call",
+      toolCallId: synthesizeToolCallId(tc.function?.name ?? "unknown", i),
+      toolName: tc.function?.name ?? "",
+      args: tc.function?.arguments ?? {}
+    })) ?? [];
+    return {
+      text,
+      content: [...text.length > 0 ? [{ type: "text", text }] : [], ...toolCalls],
+      toolCalls,
+      finishReason: mapFinishReason(json),
+      usage: mapUsage(json),
+      rawResponse: json
+    };
+  }
+  async stream(options) {
+    const body = this.buildBody(options, true);
+    const res = await this.fetch("/api/chat", body, options);
+    if (!res.body) {
+      throw new APICallError({
+        message: "Ollama streaming response has no body.",
+        statusCode: res.status
+      });
+    }
+    const ndjson = parseNDJSON(res.body);
+    return new ReadableStream({
+      async start(controller) {
+        let usage;
+        let finish = "unknown";
+        let toolIdx = 0;
+        try {
+          for await (const chunk of ndjson) {
+            if (chunk.message?.content) {
+              controller.enqueue({ type: "text-delta", textDelta: chunk.message.content });
+            }
+            if (chunk.message?.tool_calls) {
+              for (const tc of chunk.message.tool_calls) {
+                const id = synthesizeToolCallId(tc.function?.name ?? "unknown", toolIdx++);
+                controller.enqueue({
+                  type: "tool-call",
+                  toolCallId: id,
+                  toolName: tc.function?.name ?? "",
+                  args: tc.function?.arguments ?? {}
+                });
+              }
+            }
+            if (chunk.done) {
+              finish = mapFinishReason(chunk);
+              usage = mapUsage(chunk);
+            }
+          }
+          controller.enqueue({ type: "finish", finishReason: finish, usage: usage ?? {} });
+          controller.close();
+        } catch (err) {
+          controller.enqueue({ type: "error", error: err });
+          controller.close();
+        }
+      }
+    });
+  }
+  /**
+   * Local models are free at runtime — their cost is electricity / GPU
+   * time, not per-token billing. We report `pricingAvailable: true`
+   * with `$0` so `Budget Guard.maxUsd` constraints simply never trip
+   * for Ollama runs (instead of being silently disabled, which would
+   * surprise callers migrating from OpenAI).
+   *
+   * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets
+   * keep working as documented — those are the meaningful limits when
+   * GPU time is the bottleneck.
+   */
+  estimateCost(options) {
+    const inputTokens = estimateTokensFromMessages(asChatMessages(options.messages));
+    const maxOut = options.maxTokens ?? 8192;
+    const minOut = Math.min(16, maxOut);
+    return {
+      minTokens: inputTokens + minOut,
+      maxTokens: inputTokens + maxOut,
+      minUsd: 0,
+      maxUsd: 0,
+      pricingAvailable: true
+    };
+  }
+  buildBody(options, stream) {
+    const ollamaOptions = {
+      ...this.config.defaultOptions ?? {}
+    };
+    if (options.temperature !== void 0) ollamaOptions.temperature = options.temperature;
+    if (options.topP !== void 0) ollamaOptions.top_p = options.topP;
+    if (options.maxTokens !== void 0) ollamaOptions.num_predict = options.maxTokens;
+    if (options.stopSequences !== void 0) ollamaOptions.stop = options.stopSequences;
+    if (options.seed !== void 0) ollamaOptions.seed = options.seed;
+    const body = {
+      model: this.modelId,
+      messages: options.messages.map(toOllamaMessage),
+      stream
+    };
+    if (Object.keys(ollamaOptions).length > 0) body.options = ollamaOptions;
+    if (options.tools?.length) {
+      body.tools = options.tools.map((t) => ({
+        type: "function",
+        function: {
+          name: t.name,
+          ...t.description !== void 0 ? { description: t.description } : {},
+          parameters: t.parameters
+        }
+      }));
+    }
+    if (options.providerOptions) Object.assign(body, options.providerOptions);
+    return body;
+  }
+  async fetch(path, body, options) {
+    const url = `${this.config.baseURL}${path}`;
+    const headers = { ...this.config.headers, ...options.headers };
+    const init = {
+      method: "POST",
+      headers,
+      body: JSON.stringify(body)
+    };
+    if (options.abortSignal) init.signal = options.abortSignal;
+    const res = await this.config.fetcher(url, init);
+    if (!res.ok) {
+      const text = await res.text().catch(() => "");
+      throw new APICallError({
+        message: `Ollama API error: ${res.status} ${res.statusText}`,
+        url,
+        statusCode: res.status,
+        responseBody: text
+      });
+    }
+    return res;
+  }
+};
+function toOllamaMessage(m) {
+  switch (m.role) {
+    case "system":
+    case "user": {
+      const text = m.content.filter((p) => p.type === "text").map((p) => p.text).join("");
+      const images = m.content.filter((p) => p.type === "image").map((p) => {
+        const img = p.image;
+        if (typeof img === "string") return img.replace(/^data:[^;]+;base64,/, "");
+        if (img instanceof URL) return img.toString();
+        return uint8ToBase64(img);
+      });
+      const out = { role: m.role, content: text };
+      if (images.length > 0) out.images = images;
+      return out;
+    }
+    case "assistant": {
+      const text = m.content.filter((p) => p.type === "text").map((p) => p.text).join("");
+      const toolCalls = m.content.filter((p) => p.type === "tool-call");
+      const out = { role: "assistant", content: text };
+      if (toolCalls.length > 0) {
+        out.tool_calls = toolCalls.map((tc) => ({
+          function: {
+            name: tc.toolName,
+            arguments: tc.args ?? {}
+          }
+        }));
+      }
+      return out;
+    }
+    case "tool": {
+      const first = m.content[0];
+      if (!first || first.type !== "tool-result") {
+        return { role: "tool", content: "" };
+      }
+      return {
+        role: "tool",
+        content: typeof first.result === "string" ? first.result : JSON.stringify(first.result)
+      };
+    }
+  }
+}
+function synthesizeToolCallId(name, index) {
+  return `ollama_${name}_${index}_${Math.random().toString(36).slice(2, 8)}`;
+}
+function asChatMessages(messages) {
+  return messages;
+}
+function uint8ToBase64(arr) {
+  let s = "";
+  for (let i = 0; i < arr.byteLength; i++) s += String.fromCharCode(arr[i]);
+  return typeof btoa !== "undefined" ? btoa(s) : Buffer.from(s, "binary").toString("base64");
+}
+function mapFinishReason(json) {
+  if (json.message?.tool_calls?.length) return "tool-calls";
+  if (json.done_reason === "stop" || json.done) return "stop";
+  if (json.done_reason === "length") return "length";
+  return "unknown";
+}
+function mapUsage(json) {
+  const out = {};
+  if (json.prompt_eval_count !== void 0) out.promptTokens = json.prompt_eval_count;
+  if (json.eval_count !== void 0) out.completionTokens = json.eval_count;
+  if (json.prompt_eval_count !== void 0 && json.eval_count !== void 0) {
+    out.totalTokens = json.prompt_eval_count + json.eval_count;
+  }
+  return out;
+}
+
+// src/ollama-provider.ts
+function createOllama(options = {}) {
+  const baseURL = (options.baseURL ?? loadEnv("OLLAMA_BASE_URL") ?? "http://localhost:11434").replace(/\/+$/, "");
+  const fetcher = options.fetch ?? globalThis.fetch;
+  const headers = {
+    "Content-Type": "application/json",
+    ...options.headers
+  };
+  const make = (modelId) => new OllamaChatModel({
+    modelId,
+    baseURL,
+    headers,
+    fetcher,
+    ...options.defaultOptions ? { defaultOptions: options.defaultOptions } : {}
+  });
+  const provider = ((modelId) => make(modelId));
+  provider.chat = make;
+  return provider;
+}
+var ollama = createOllama();
+function loadEnv(name) {
+  if (typeof process !== "undefined" && process.env) {
+    return process.env[name];
+  }
+  return void 0;
+}
+
+export { OllamaChatModel, createOllama, ollama };
+//# sourceMappingURL=index.js.map
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/util/ndjson.ts","../src/ollama-chat-model.ts","../src/ollama-provider.ts"],"names":[],"mappings":";;;;;AAQA,gBAAuB,YACrB,IAAA,EACkB;AAClB,EAAA,MAAM,MAAA,GAAS,KAAK,SAAA,EAAU;AAC9B,EAAA,MAAM,OAAA,GAAU,IAAI,WAAA,EAAY;AAChC,EAAA,IAAI,MAAA,GAAS,EAAA;AACb,EAAA,IAAI;AACF,IAAA,OAAO,IAAA,EAAM;AACX,MAAA,MAAM,EAAE,KAAA,EAAO,IAAA,EAAK,GAAI,MAAM,OAAO,IAAA,EAAK;AAC1C,MAAA,IAAI,IAAA,EAAM;AACV,MAAA,MAAA,IAAU,QAAQ,MAAA,CAAO,KAAA,EAAO,EAAE,MAAA,EAAQ,MAAM,CAAA;AAChD,MAAA,IAAI,EAAA,GAAK,MAAA,CAAO,OAAA,CAAQ,IAAI,CAAA;AAC5B,MAAA,OAAO,OAAO,CAAA,CAAA,EAAI;AAChB,QAAA,MAAM,IAAA,GAAO,MAAA,CAAO,KAAA,CAAM,CAAA,EAAG,EAAE,EAAE,OAAA,CAAQ,KAAA,EAAO,EAAE,CAAA,CAAE,IAAA,EAAK;AACzD,QAAA,MAAA,GAAS,MAAA,CAAO,KAAA,CAAM,EAAA,GAAK,CAAC,CAAA;AAC5B,QAAA,IAAI,IAAA,CAAK,SAAS,CAAA,EAAG;AACnB,UAAA,MAAM,IAAA,CAAK,MAAM,IAAI,CAAA;AAAA,QACvB;AACA,QAAA,EAAA,GAAK,MAAA,CAAO,QAAQ,IAAI,CAAA;AAAA,MAC1B;AAAA,IACF;AACA,IAAA,MAAM,IAAA,GAAO,OAAO,IAAA,EAAK;AACzB,IAAA,IAAI,KAAK,MAAA,GAAS,CAAA,EAAG,MAAM,IAAA,CAAK,MAAM,IAAI,CAAA;AAAA,EAC5C,CAAA,SAAE;AACA,IAAA,MAAA,CAAO,WAAA,EAAY;AAAA,EACrB;AACF;;;AC2BO,IAAM,kBAAN,MAA+C;AAAA,EAC3C,QAAA,GAAW,QAAA;AAAA,EACX,OAAA;AAAA,EACQ,MAAA;AAAA,EAEjB,YAAY,MAAA,EAA+B;AACzC,IAAA,IAAA,CAAK,UAAU,MAAA,CAAO,OAAA;AACtB,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AAAA,EAChB;AAAA,EAEA,MAAM,SAAS,OAAA,EAAyD;AACtE,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,SAAA,CAAU,OAAA,EAAS,KAAK,CAAA;AAC1C,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,KAAA,CAAM,WAAA,EAAa,MAAM,OAAO,CAAA;AACvD,IAAA,MAAM,IAAA,GAAQ,MAAM,GAAA,CAAI,IAAA,EAAK;AAE7B,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,OAAA,EAAS,OAAA,IAAW,EAAA;AACtC,IAAA,MAAM,YACJ,IAAA,CAAK,OAAA,EAAS,YAAY,GAAA,CAAI,CAAC,IAAI,CAAA,MAAO;AAAA,MACxC,IAAA,EAAM,WAAA;AAAA,MACN,YAAY,oBAAA,CAAqB,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,WAAW,CAAC,CAAA;AAAA,MAClE,QAAA,EAAU,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,EAAA;AAAA,MAC/B,IAAA,EAAM,EAAA,CAAG,QAAA,EAAU,SAAA,IAAa;AAAC,KACnC,CAAE,KAAK,EAAC;AAEV,IAAA,OAAO;AAAA,MACL,IAAA;AAAA,MACA,SAAS,CAAC,GAAI,IAAA,CAAK,MAAA,GAAS,IAAI,CAAC,EAAE,IAAA,EAAM,MAAA,EAAiB,MAAM,CAAA,GAAI,EAAC,EAAI,GAAG,SAAS,CAAA;AAAA,MACrF,SAAA;AAAA,MACA,YAAA,EAAc,gBAAgB,IAAI,CAAA;AAAA,MAClC,KAAA,EAAO,SAAS,IAAI,CAAA;AAAA,MACpB,WAAA,EAAa;AAAA,KACf;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,OAAA,EAAqE;AAChF,IAAA,MAAM,IAAA,GAAO,IAAA,CAAK,SAAA,CAAU,OAAA,EAAS,IAAI,CAAA;AACzC,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,KAAA,CAAM,WAAA,EAAa,MAAM,OAAO,CAAA;AACvD,IAAA,IAAI,CAAC,IAAI,IAAA,EAAM;AACb,MAAA,MAAM,IAAI,YAAA,CAAa;AAAA,QACrB,OAAA,EAAS,wCAAA;AAAA,QACT,YAAY,GAAA,CAAI;AAAA,OACjB,CAAA;AAAA,IACH;AAEA,IAAA,MAAM,MAAA,GAAS,WAAA,CAAgC,GAAA,CAAI,IAAI,CAAA;AAEvD,IAAA,OAAO,IAAI,cAAA,CAAgC;AAAA,MACzC,MAAM,MAAM,UAAA,EAAY;AACtB,QAAA,IAAI,KAAA;AACJ,QAAA,IAAI,MAAA,GAAuB,SAAA;AAC3B,QAAA,IAAI,OAAA,GAAU,CAAA;AAEd,QAAA,IAAI;AACF,UAAA,WAAA,MAAiB,SAAS,MAAA,EAAQ;AAChC,YAAA,IAAI,KAAA,CAAM,SAAS,OAAA,EAAS;AAC1B,cAAA,UAAA,CAAW,OAAA,CAAQ,EAAE,IAAA,EAAM,YAAA,EAAc,WAAW,KAAA,CAAM,OAAA,CAAQ,SAAS,CAAA;AAAA,YAC7E;AACA,YAAA,IAAI,KAAA,CAAM,SAAS,UAAA,EAAY;AAC7B,cAAA,KAAA,MAAW,EAAA,IAAM,KAAA,CAAM,OAAA,CAAQ,UAAA,EAAY;AACzC,gBAAA,MAAM,KAAK,oBAAA,CAAqB,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,WAAW,OAAA,EAAS,CAAA;AAKzE,gBAAA,UAAA,CAAW,OAAA,CAAQ;AAAA,kBACjB,IAAA,EAAM,WAAA;AAAA,kBACN,UAAA,EAAY,EAAA;AAAA,kBACZ,QAAA,EAAU,EAAA,CAAG,QAAA,EAAU,IAAA,IAAQ,EAAA;AAAA,kBAC/B,IAAA,EAAM,EAAA,CAAG,QAAA,EAAU,SAAA,IAAa;AAAC,iBAClC,CAAA;AAAA,cACH;AAAA,YACF;AACA,YAAA,IAAI,MAAM,IAAA,EAAM;AACd,cAAA,MAAA,GAAS,gBAAgB,KAAK,CAAA;AAC9B,cAAA,KAAA,GAAQ,SAAS,KAAK,CAAA;AAAA,YACxB;AAAA,UACF;AACA,UAAA,UAAA,CAAW,OAAA,CAAQ,EAAE,IAAA,EAAM,QAAA,EAAU,YAAA,EAAc,QAAQ,KAAA,EAAO,KAAA,IAAS,EAAC,EAAG,CAAA;AAC/E,UAAA,UAAA,CAAW,KAAA,EAAM;AAAA,QACnB,SAAS,GAAA,EAAK;AACZ,UAAA,UAAA,CAAW,QAAQ,EAAE,IAAA,EAAM,OAAA,EAAS,KAAA,EAAO,KAAK,CAAA;AAChD,UAAA,UAAA,CAAW,KAAA,EAAM;AAAA,QACnB;AAAA,MACF;AAAA,KACD,CAAA;AAAA,EACH;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAaA,aAAa,OAAA,EAAyC;AACpD,IAAA,MAAM,WAAA,GAAc,0BAAA,CAA2B,cAAA,CAAe,OAAA,CAAQ,QAAQ,CAAC,CAAA;AAC/E,IAAA,MAAM,MAAA,GAAS,QAAQ,SAAA,IAAa,IAAA;AACpC,IAAA,MAAM,MAAA,GAAS,IAAA,CAAK,GAAA,CAAI,EAAA,EAAI,MAAM,CAAA;AAClC,IAAA,OAAO;AAAA,MACL,WAAW,WAAA,GAAc,MAAA;AAAA,MACzB,WAAW,WAAA,GAAc,MAAA;AAAA,MACzB,MAAA,EAAQ,CAAA;AAAA,MACR,MAAA,EAAQ,CAAA;AAAA,MACR,gBAAA,EAAkB;AAAA,KACpB;AAAA,EACF;AAAA,EAEQ,SAAA,CAAU,SAA2B,MAAA,EAA0C;AACrF,IAAA,MAAM,aAAA,GAAyC;AAAA,MAC7C,GAAI,IAAA,CAAK,MAAA,CAAO,cAAA,IAAkB;AAAC,KACrC;AACA,IAAA,IAAI,OAAA,CAAQ,WAAA,KAAgB,MAAA,EAAW,aAAA,CAAc,cAAc,OAAA,CAAQ,WAAA;AAC3E,IAAA,IAAI,OAAA,CAAQ,IAAA,KAAS,MAAA,EAAW,aAAA,CAAc,QAAQ,OAAA,CAAQ,IAAA;AAC9D,IAAA,IAAI,OAAA,CAAQ,SAAA,KAAc,MAAA,EAAW,aAAA,CAAc,cAAc,OAAA,CAAQ,SAAA;AACzE,IAAA,IAAI,OAAA,CAAQ,aAAA,KAAkB,MAAA,EAAW,aAAA,CAAc,OAAO,OAAA,CAAQ,aAAA;AACtE,IAAA,IAAI,OAAA,CAAQ,IAAA,KAAS,MAAA,EAAW,aAAA,CAAc,OAAO,OAAA,CAAQ,IAAA;AAE7D,IAAA,MAAM,IAAA,GAAgC;AAAA,MACpC,OAAO,IAAA,CAAK,OAAA;AAAA,MACZ,QAAA,EAAU,OAAA,CAAQ,QAAA,CAAS,GAAA,CAAI,eAAe,CAAA;AAAA,MAC9C;AAAA,KACF;AACA,IAAA,IAAI,OAAO,IAAA,CAAK,aAAa,EAAE,MAAA,GAAS,CAAA,OAAQ,OAAA,GAAU,aAAA;AAC1D,IAAA,IAAI,OAAA,CAAQ,OAAO,MAAA,EAAQ;AACzB,MAAA,IAAA,CAAK,KAAA,GAAQ,OAAA,CAAQ,KAAA,CAAM,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,QACrC,IAAA,EAAM,UAAA;AAAA,QACN,QAAA,EAAU;AAAA,UACR,MAAM,CAAA,CAAE,IAAA;AAAA,UACR,GAAI,EAAE,WAAA,KAAgB,MAAA,GAAY,EAAE,WAAA,EAAa,CAAA,CAAE,WAAA,EAAY,GAAI,EAAC;AAAA,UACpE,YAAY,CAAA,CAAE;AAAA;AAChB,OACF,CAAE,CAAA;AAAA,IACJ;AACA,IAAA,IAAI,QAAQ,eAAA,EAAiB,MAAA,CAAO,MAAA,CAAO,IAAA,EAAM,QAAQ,eAAe,CAAA;AACxE,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,MAAc,KAAA,CAAM,IAAA,EAAc,IAAA,EAAe,OAAA,EAA8C;AAC7F,IAAA,MAAM,MAAM,CAAA,EAAG,IAAA,CAAK,MAAA,CAAO,OAAO,GAAG,IAAI,CAAA,CAAA;AACzC,IAAA,MAAM,OAAA,GAAU,EAAE,GAAG,IAAA,CAAK,OAAO,OAAA,EAAS,GAAG,QAAQ,OAAA,EAAQ;AAC7D,IAAA,MAAM,IAAA,GAAoB;AAAA,MACxB,MAAA,EAAQ,MAAA;AAAA,MACR,OAAA;AAAA,MACA,IAAA,EAAM,IAAA,CAAK,SAAA,CAAU,IAAI;AAAA,KAC3B;AACA,IAAA,IAAI,OAAA,CAAQ,WAAA,EAAa,IAAA,CAAK,MAAA,GAAS,OAAA,CAAQ,WAAA;AAE/C,IAAA,MAAM,MAAM,MAAM,IAAA,CAAK,MAAA,CAAO,OAAA,CAAQ,KAAK,IAAI,CAAA;AAC/C,IAAA,IAAI,CAAC,IAAI,EAAA,EAAI;AACX,MAAA,MAAM,OAAO,MAAM,GAAA,CAAI,MAAK,CAAE,KAAA,CAAM,MAAM,EAAE,CAAA;AAC5C,MAAA,MAAM,IAAI,YAAA,CAAa;AAAA,QACrB,SAAS,CAAA,kBAAA,EAAqB,GAAA,CAAI,MAAM,CAAA,CAAA,EAAI,IAAI,UAAU,CAAA,CAAA;AAAA,QAC1D,GAAA;AAAA,QACA,YAAY,GAAA,CAAI,MAAA;AAAA,QAChB,YAAA,EAAc;AAAA,OACf,CAAA;AAAA,IACH;AACA,IAAA,OAAO,GAAA;AAAA,EACT;AACF;AAEA,SAAS,gBAAgB,CAAA,EAA+B;AACtD,EAAA,QAAQ,EAAE,IAAA;AAAM,IACd,KAAK,QAAA;AAAA,IACL,KAAK,MAAA,EAAQ;AACX,MAAA,MAAM,OAAO,CAAA,CAAE,OAAA,CACZ,MAAA,CAAO,CAAC,MAAM,CAAA,CAAE,IAAA,KAAS,MAAM,CAAA,CAC/B,IAAI,CAAC,CAAA,KAAO,EAAuB,IAAI,CAAA,CACvC,KAAK,EAAE,CAAA;AACV,MAAA,MAAM,MAAA,GAAS,CAAA,CAAE,OAAA,CACd,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,IAAA,KAAS,OAAO,CAAA,CAChC,GAAA,CAAI,CAAC,CAAA,KAAM;AACV,QAAA,MAAM,MAAO,CAAA,CAA2C,KAAA;AACxD,QAAA,IAAI,OAAO,GAAA,KAAQ,QAAA,SAAiB,GAAA,CAAI,OAAA,CAAQ,uBAAuB,EAAE,CAAA;AACzE,QAAA,IAAI,GAAA,YAAe,GAAA,EAAK,OAAO,GAAA,CAAI,QAAA,EAAS;AAC5C,QAAA,OAAO,cAAc,GAAG,CAAA;AAAA,MAC1B,CAAC,CAAA;AACH,MAAA,MAAM,MAA+B,EAAE,IAAA,EAAM,CAAA,CAAE,IAAA,EAAM,SAAS,IAAA,EAAK;AACnE,MAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,EAAG,GAAA,CAAI,MAAA,GAAS,MAAA;AACpC,MAAA,OAAO,GAAA;AAAA,IACT;AAAA,IACA,KAAK,WAAA,EAAa;AAChB,MAAA,MAAM,OAAO,CAAA,CAAE,OAAA,CACZ,MAAA,CAAO,CAAC,MAAM,CAAA,CAAE,IAAA,KAAS,MAAM,CAAA,CAC/B,IAAI,CAAC,CAAA,KAAO,EAAuB,IAAI,CAAA,CACvC,KAAK,EAAE,CAAA;AACV,MAAA,MAAM,SAAA,GAAY,EAAE,OAAA,CAAQ,MAAA,CAAO,CAAC,CAAA,KAAM,CAAA,CAAE,SAAS,WAAW,CAAA;AAChE,MAAA,MAAM,GAAA,GAA+B,EAAE,IAAA,EAAM,WAAA,EAAa,SAAS,IAAA,EAAK;AACxE,MAAA,IAAI,SAAA,CAAU,SAAS,CAAA,EAAG;AACxB,QAAA,GAAA,CAAI,UAAA,GAAa,SAAA,CAAU,GAAA,CAAI,CAAC,EAAA,MAAQ;AAAA,UACtC,QAAA,EAAU;AAAA,YACR,MAAO,EAAA,CAAoB,QAAA;AAAA,YAC3B,SAAA,EAAY,EAAA,CAAoB,IAAA,IAAQ;AAAC;AAC3C,SACF,CAAE,CAAA;AAAA,MACJ;AACA,MAAA,OAAO,GAAA;AAAA,IACT;AAAA,IACA,KAAK,MAAA,EAAQ;AACX,MAAA,MAAM,KAAA,GAAQ,CAAA,CAAE,OAAA,CAAQ,CAAC,CAAA;AACzB,MAAA,IAAI,CAAC,KAAA,IAAS,KAAA,CAAM,IAAA,KAAS,aAAA,EAAe;AAC1C,QAAA,OAAO,EAAE,IAAA,EAAM,MAAA,EAAQ,OAAA,EAAS,EAAA,EAAG;AAAA,MACrC;AACA,MAAA,OAAO;AAAA,QACL,IAAA,EAAM,MAAA;AAAA,QACN,OAAA,EAAS,OAAO,KAAA,CAAM,MAAA,KAAW,QAAA,GAAW,MAAM,MAAA,GAAS,IAAA,CAAK,SAAA,CAAU,KAAA,CAAM,MAAM;AAAA,OACxF;AAAA,IACF;AAAA;AAEJ;AAQA,SAAS,oBAAA,CAAqB,MAAc,KAAA,EAAuB;AACjE,EAAA,OAAO,CAAA,OAAA,EAAU,IAAI,CAAA,CAAA,EAAI,KAAK,IAAI,IAAA,CAAK,MAAA,EAAO,CAAE,QAAA,CAAS,EAAE,CAAA,CAAE,KAAA,CAAM,CAAA,EAAG,CAAC,CAAC,CAAA,CAAA;AAC1E;AAEA,SAAS,eACP,QAAA,EACkD;AAClD,EAAA,OAAO,QAAA;AACT;AAEA,SAAS,cAAc,GAAA,EAAyB;AAC9C,EAAA,IAAI,CAAA,GAAI,EAAA;AACR,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,GAAA,CAAI,UAAA,EAAY,CAAA,EAAA,EAAK,CAAA,IAAK,MAAA,CAAO,YAAA,CAAa,GAAA,CAAI,CAAC,CAAW,CAAA;AAClF,EAAA,OAAO,OAAO,IAAA,KAAS,WAAA,GAAc,IAAA,CAAK,CAAC,CAAA,GAAI,MAAA,CAAO,IAAA,CAAK,CAAA,EAAG,QAAQ,CAAA,CAAE,QAAA,CAAS,QAAQ,CAAA;AAC3F;AAEA,SAAS,gBAAgB,IAAA,EAAwC;AAC/D,EAAA,IAAI,IAAA,CAAK,OAAA,EAAS,UAAA,EAAY,MAAA,EAAQ,OAAO,YAAA;AAC7C,EAAA,IAAI,IAAA,CAAK,WAAA,KAAgB,MAAA,IAAU,IAAA,CAAK,MAAM,OAAO,MAAA;AACrD,EAAA,IAAI,IAAA,CAAK,WAAA,KAAgB,QAAA,EAAU,OAAO,QAAA;AAC1C,EAAA,OAAO,SAAA;AACT;AAEA,SAAS,SAAS,IAAA,EAAsC;AACtD,EAAA,MAAM,MAAkB,EAAC;AACzB,EAAA,IAAI,IAAA,CAAK,iBAAA,KAAsB,MAAA,EAAW,GAAA,CAAI,eAAe,IAAA,CAAK,iBAAA;AAClE,EAAA,IAAI,IAAA,CAAK,UAAA,KAAe,MAAA,EAAW,GAAA,CAAI,mBAAmB,IAAA,CAAK,UAAA;AAC/D,EAAA,IAAI,IAAA,CAAK,iBAAA,KAAsB,MAAA,IAAa,IAAA,CAAK,eAAe,MAAA,EAAW;AACzE,IAAA,GAAA,CAAI,WAAA,GAAc,IAAA,CAAK,iBAAA,GAAoB,IAAA,CAAK,UAAA;AAAA,EAClD;AACA,EAAA,OAAO,GAAA;AACT;;;ACtRO,SAAS,YAAA,CAAa,OAAA,GAAiC,EAAC,EAAmB;AAChF,EAAA,MAAM,OAAA,GAAA,CACJ,QAAQ,OAAA,IACR,OAAA,CAAQ,iBAAiB,CAAA,IACzB,wBAAA,EACA,OAAA,CAAQ,MAAA,EAAQ,EAAE,CAAA;AACpB,EAAA,MAAM,OAAA,GAAU,OAAA,CAAQ,KAAA,IAAS,UAAA,CAAW,KAAA;AAC5C,EAAA,MAAM,OAAA,GAAkC;AAAA,IACtC,cAAA,EAAgB,kBAAA;AAAA,IAChB,GAAG,OAAA,CAAQ;AAAA,GACb;AAEA,EAAA,MAAM,IAAA,GAAO,CAAC,OAAA,KACZ,IAAI,eAAA,CAAgB;AAAA,IAClB,OAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAA;AAAA,IACA,GAAI,QAAQ,cAAA,GAAiB,EAAE,gBAAgB,OAAA,CAAQ,cAAA,KAAmB;AAAC,GAC5E,CAAA;AAEH,EAAA,MAAM,QAAA,IAAY,CAAC,OAAA,KAA+B,IAAA,CAAK,OAAO,CAAA,CAAA;AAC9D,EAAA,QAAA,CAAS,IAAA,GAAO,IAAA;AAChB,EAAA,OAAO,QAAA;AACT;AAQO,IAAM,SAAyB,YAAA;AAEtC,SAAS,QAAQ,IAAA,EAAkC;AACjD,EAAA,IAAI,OAAO,OAAA,KAAY,WAAA,IAAe,OAAA,CAAQ,GAAA,EAAK;AACjD,IAAA,OAAO,OAAA,CAAQ,IAAI,IAAI,CAAA;AAAA,EACzB;AACA,EAAA,OAAO,MAAA;AACT","file":"index.js","sourcesContent":["/**\n * Parse a `ReadableStream<Uint8Array>` of newline-delimited JSON into an\n * async iterator of parsed records. Tolerates `\\r\\n`, blank lines, and\n * partial chunks.\n *\n * Ollama's `/api/chat?stream=true` emits one JSON object per line —\n * unlike OpenAI / Anthropic which use SSE (`data: {...}\\n\\n`).\n */\nexport async function* parseNDJSON<T = unknown>(\n  body: ReadableStream<Uint8Array>,\n): AsyncIterable<T> {\n  const reader = body.getReader();\n  const decoder = new TextDecoder();\n  let buffer = '';\n  try {\n    while (true) {\n      const { value, done } = await reader.read();\n      if (done) break;\n      buffer += decoder.decode(value, { stream: true });\n      let nl = buffer.indexOf('\\n');\n      while (nl !== -1) {\n        const line = buffer.slice(0, nl).replace(/\\r$/, '').trim();\n        buffer = buffer.slice(nl + 1);\n        if (line.length > 0) {\n          yield JSON.parse(line) as T;\n        }\n        nl = buffer.indexOf('\\n');\n      }\n    }\n    const tail = buffer.trim();\n    if (tail.length > 0) yield JSON.parse(tail) as T;\n  } finally {\n    reader.releaseLock();\n  }\n}\n","import {\n  APICallError,\n  type CostEstimate,\n  estimateTokensFromMessages,\n  type FinishReason,\n  type LanguageModel,\n  type ModelCallOptions,\n  type ModelGenerateResult,\n  type ModelStreamPart,\n  type NormalizedMessage,\n  type TokenUsage,\n  type ToolCallPart,\n} from '@ziro-agent/core';\nimport { parseNDJSON } from './util/ndjson.js';\n\n/**\n * Common Ollama model ids. Ollama is open-weight and the catalogue\n * grows weekly, so we leave the type open via `(string & {})` — pulling\n * a model is `ollama pull <name>` and any tag works.\n *\n * The enumerated entries are the most-pulled tool-capable models on\n * ollama.com as of v0.1.9. They are deliberately conservative — listing\n * every quant would explode the union without changing runtime behaviour.\n */\nexport type OllamaChatModelId =\n  | 'llama3.1'\n  | 'llama3.1:8b'\n  | 'llama3.1:70b'\n  | 'llama3.2'\n  | 'llama3.2:3b'\n  | 'llama3.3'\n  | 'qwen2.5'\n  | 'qwen2.5:7b'\n  | 'qwen2.5:14b'\n  | 'qwen2.5:32b'\n  | 'qwen2.5-coder'\n  | 'mistral'\n  | 'mistral-nemo'\n  | 'mixtral'\n  | 'gemma3'\n  | 'phi4'\n  | (string & {});\n\ninterface OllamaChatModelConfig {\n  modelId: OllamaChatModelId;\n  baseURL: string;\n  headers: Record<string, string>;\n  fetcher: typeof fetch;\n  /** Forwarded into Ollama's per-request `options` block. */\n  defaultOptions?: Record<string, unknown>;\n}\n\n/**\n * `LanguageModel` adapter for the local Ollama HTTP API\n * (`http://localhost:11434/api/chat`). Talks the native Ollama protocol\n * (NDJSON streaming, function-calling shape) — not the OpenAI-compat\n * shim, which loses tool-call fidelity on several models.\n *\n * Sovereign-pillar primitive (RFC 0004 §v0.1.9): an open-weight default\n * provider so the SDK is not assumed-cloud-only.\n */\nexport class OllamaChatModel implements LanguageModel {\n  readonly provider = 'ollama';\n  readonly modelId: string;\n  private readonly config: OllamaChatModelConfig;\n\n  constructor(config: OllamaChatModelConfig) {\n    this.modelId = config.modelId;\n    this.config = config;\n  }\n\n  async generate(options: ModelCallOptions): Promise<ModelGenerateResult> {\n    const body = this.buildBody(options, false);\n    const res = await this.fetch('/api/chat', body, options);\n    const json = (await res.json()) as OllamaChatResponse;\n\n    const text = json.message?.content ?? '';\n    const toolCalls: ToolCallPart[] =\n      json.message?.tool_calls?.map((tc, i) => ({\n        type: 'tool-call' as const,\n        toolCallId: synthesizeToolCallId(tc.function?.name ?? 'unknown', i),\n        toolName: tc.function?.name ?? '',\n        args: tc.function?.arguments ?? {},\n      })) ?? [];\n\n    return {\n      text,\n      content: [...(text.length > 0 ? [{ type: 'text' as const, text }] : []), ...toolCalls],\n      toolCalls,\n      finishReason: mapFinishReason(json),\n      usage: mapUsage(json),\n      rawResponse: json,\n    };\n  }\n\n  async stream(options: ModelCallOptions): Promise<ReadableStream<ModelStreamPart>> {\n    const body = this.buildBody(options, true);\n    const res = await this.fetch('/api/chat', body, options);\n    if (!res.body) {\n      throw new APICallError({\n        message: 'Ollama streaming response has no body.',\n        statusCode: res.status,\n      });\n    }\n\n    const ndjson = parseNDJSON<OllamaChatResponse>(res.body);\n\n    return new ReadableStream<ModelStreamPart>({\n      async start(controller) {\n        let usage: TokenUsage | undefined;\n        let finish: FinishReason = 'unknown';\n        let toolIdx = 0;\n\n        try {\n          for await (const chunk of ndjson) {\n            if (chunk.message?.content) {\n              controller.enqueue({ type: 'text-delta', textDelta: chunk.message.content });\n            }\n            if (chunk.message?.tool_calls) {\n              for (const tc of chunk.message.tool_calls) {\n                const id = synthesizeToolCallId(tc.function?.name ?? 'unknown', toolIdx++);\n                // Ollama emits each tool call complete (no incremental\n                // arg streaming), so we forward the full call directly\n                // — matching the OpenAI provider's terminal `tool-call`\n                // event shape.\n                controller.enqueue({\n                  type: 'tool-call',\n                  toolCallId: id,\n                  toolName: tc.function?.name ?? '',\n                  args: tc.function?.arguments ?? {},\n                });\n              }\n            }\n            if (chunk.done) {\n              finish = mapFinishReason(chunk);\n              usage = mapUsage(chunk);\n            }\n          }\n          controller.enqueue({ type: 'finish', finishReason: finish, usage: usage ?? {} });\n          controller.close();\n        } catch (err) {\n          controller.enqueue({ type: 'error', error: err });\n          controller.close();\n        }\n      },\n    });\n  }\n\n  /**\n   * Local models are free at runtime — their cost is electricity / GPU\n   * time, not per-token billing. We report `pricingAvailable: true`\n   * with `$0` so `Budget Guard.maxUsd` constraints simply never trip\n   * for Ollama runs (instead of being silently disabled, which would\n   * surprise callers migrating from OpenAI).\n   *\n   * `maxTokens`, `maxLlmCalls`, `maxDurationMs`, and `maxSteps` budgets\n   * keep working as documented — those are the meaningful limits when\n   * GPU time is the bottleneck.\n   */\n  estimateCost(options: ModelCallOptions): CostEstimate {\n    const inputTokens = estimateTokensFromMessages(asChatMessages(options.messages));\n    const maxOut = options.maxTokens ?? 8_192;\n    const minOut = Math.min(16, maxOut);\n    return {\n      minTokens: inputTokens + minOut,\n      maxTokens: inputTokens + maxOut,\n      minUsd: 0,\n      maxUsd: 0,\n      pricingAvailable: true,\n    };\n  }\n\n  private buildBody(options: ModelCallOptions, stream: boolean): Record<string, unknown> {\n    const ollamaOptions: Record<string, unknown> = {\n      ...(this.config.defaultOptions ?? {}),\n    };\n    if (options.temperature !== undefined) ollamaOptions.temperature = options.temperature;\n    if (options.topP !== undefined) ollamaOptions.top_p = options.topP;\n    if (options.maxTokens !== undefined) ollamaOptions.num_predict = options.maxTokens;\n    if (options.stopSequences !== undefined) ollamaOptions.stop = options.stopSequences;\n    if (options.seed !== undefined) ollamaOptions.seed = options.seed;\n\n    const body: Record<string, unknown> = {\n      model: this.modelId,\n      messages: options.messages.map(toOllamaMessage),\n      stream,\n    };\n    if (Object.keys(ollamaOptions).length > 0) body.options = ollamaOptions;\n    if (options.tools?.length) {\n      body.tools = options.tools.map((t) => ({\n        type: 'function',\n        function: {\n          name: t.name,\n          ...(t.description !== undefined ? { description: t.description } : {}),\n          parameters: t.parameters,\n        },\n      }));\n    }\n    if (options.providerOptions) Object.assign(body, options.providerOptions);\n    return body;\n  }\n\n  private async fetch(path: string, body: unknown, options: ModelCallOptions): Promise<Response> {\n    const url = `${this.config.baseURL}${path}`;\n    const headers = { ...this.config.headers, ...options.headers };\n    const init: RequestInit = {\n      method: 'POST',\n      headers,\n      body: JSON.stringify(body),\n    };\n    if (options.abortSignal) init.signal = options.abortSignal;\n\n    const res = await this.config.fetcher(url, init);\n    if (!res.ok) {\n      const text = await res.text().catch(() => '');\n      throw new APICallError({\n        message: `Ollama API error: ${res.status} ${res.statusText}`,\n        url,\n        statusCode: res.status,\n        responseBody: text,\n      });\n    }\n    return res;\n  }\n}\n\nfunction toOllamaMessage(m: NormalizedMessage): unknown {\n  switch (m.role) {\n    case 'system':\n    case 'user': {\n      const text = m.content\n        .filter((p) => p.type === 'text')\n        .map((p) => (p as { text: string }).text)\n        .join('');\n      const images = m.content\n        .filter((p) => p.type === 'image')\n        .map((p) => {\n          const img = (p as { image: string | URL | Uint8Array }).image;\n          if (typeof img === 'string') return img.replace(/^data:[^;]+;base64,/, '');\n          if (img instanceof URL) return img.toString();\n          return uint8ToBase64(img);\n        });\n      const out: Record<string, unknown> = { role: m.role, content: text };\n      if (images.length > 0) out.images = images;\n      return out;\n    }\n    case 'assistant': {\n      const text = m.content\n        .filter((p) => p.type === 'text')\n        .map((p) => (p as { text: string }).text)\n        .join('');\n      const toolCalls = m.content.filter((p) => p.type === 'tool-call');\n      const out: Record<string, unknown> = { role: 'assistant', content: text };\n      if (toolCalls.length > 0) {\n        out.tool_calls = toolCalls.map((tc) => ({\n          function: {\n            name: (tc as ToolCallPart).toolName,\n            arguments: (tc as ToolCallPart).args ?? {},\n          },\n        }));\n      }\n      return out;\n    }\n    case 'tool': {\n      const first = m.content[0];\n      if (!first || first.type !== 'tool-result') {\n        return { role: 'tool', content: '' };\n      }\n      return {\n        role: 'tool',\n        content: typeof first.result === 'string' ? first.result : JSON.stringify(first.result),\n      };\n    }\n  }\n}\n\n/**\n * Ollama's `tool_calls[]` entries don't carry a stable id — they're\n * just `{ function: { name, arguments } }`. We synthesise an id so the\n * downstream agent loop can match tool results back to calls without\n * collisions across a batch.\n */\nfunction synthesizeToolCallId(name: string, index: number): string {\n  return `ollama_${name}_${index}_${Math.random().toString(36).slice(2, 8)}`;\n}\n\nfunction asChatMessages(\n  messages: NormalizedMessage[],\n): Parameters<typeof estimateTokensFromMessages>[0] {\n  return messages as unknown as Parameters<typeof estimateTokensFromMessages>[0];\n}\n\nfunction uint8ToBase64(arr: Uint8Array): string {\n  let s = '';\n  for (let i = 0; i < arr.byteLength; i++) s += String.fromCharCode(arr[i] as number);\n  return typeof btoa !== 'undefined' ? btoa(s) : Buffer.from(s, 'binary').toString('base64');\n}\n\nfunction mapFinishReason(json: OllamaChatResponse): FinishReason {\n  if (json.message?.tool_calls?.length) return 'tool-calls';\n  if (json.done_reason === 'stop' || json.done) return 'stop';\n  if (json.done_reason === 'length') return 'length';\n  return 'unknown';\n}\n\nfunction mapUsage(json: OllamaChatResponse): TokenUsage {\n  const out: TokenUsage = {};\n  if (json.prompt_eval_count !== undefined) out.promptTokens = json.prompt_eval_count;\n  if (json.eval_count !== undefined) out.completionTokens = json.eval_count;\n  if (json.prompt_eval_count !== undefined && json.eval_count !== undefined) {\n    out.totalTokens = json.prompt_eval_count + json.eval_count;\n  }\n  return out;\n}\n\ninterface OllamaChatResponse {\n  model?: string;\n  created_at?: string;\n  message?: {\n    role?: string;\n    content?: string;\n    tool_calls?: Array<{\n      function?: {\n        name?: string;\n        arguments?: Record<string, unknown>;\n      };\n    }>;\n  };\n  done?: boolean;\n  done_reason?: 'stop' | 'length' | 'load' | string;\n  total_duration?: number;\n  load_duration?: number;\n  prompt_eval_count?: number;\n  prompt_eval_duration?: number;\n  eval_count?: number;\n  eval_duration?: number;\n}\n","import type { LanguageModel } from '@ziro-agent/core';\nimport { OllamaChatModel, type OllamaChatModelId } from './ollama-chat-model.js';\n\nexport interface OllamaProviderOptions {\n  /**\n   * Base URL of the Ollama daemon. Defaults to\n   * `process.env.OLLAMA_BASE_URL` then `http://localhost:11434`.\n   *\n   * For remote / containerised Ollama (e.g. compose stack, Coolify),\n   * set this to `http://ollama:11434` or whatever your network exposes.\n   */\n  baseURL?: string;\n  /** Extra default headers attached to every request. */\n  headers?: Record<string, string>;\n  /** Custom `fetch`. Defaults to `globalThis.fetch`. */\n  fetch?: typeof fetch;\n  /**\n   * Default Ollama-native sampling options (`num_ctx`, `mirostat`,\n   * `repeat_penalty`, …) merged into every request's `options` block.\n   * Per-call overrides via `ModelCallOptions.providerOptions.options`\n   * still win.\n   */\n  defaultOptions?: Record<string, unknown>;\n}\n\nexport interface OllamaProvider {\n  (modelId: OllamaChatModelId): LanguageModel;\n  chat(modelId: OllamaChatModelId): LanguageModel;\n}\n\n/**\n * Build an {@link OllamaProvider} bound to a specific Ollama daemon.\n * Mirrors `createOpenAI` / `createAnthropic` so swapping providers in\n * `createAgent({ model })` is a one-line change.\n */\nexport function createOllama(options: OllamaProviderOptions = {}): OllamaProvider {\n  const baseURL = (\n    options.baseURL ??\n    loadEnv('OLLAMA_BASE_URL') ??\n    'http://localhost:11434'\n  ).replace(/\\/+$/, '');\n  const fetcher = options.fetch ?? globalThis.fetch;\n  const headers: Record<string, string> = {\n    'Content-Type': 'application/json',\n    ...options.headers,\n  };\n\n  const make = (modelId: OllamaChatModelId): LanguageModel =>\n    new OllamaChatModel({\n      modelId,\n      baseURL,\n      headers,\n      fetcher,\n      ...(options.defaultOptions ? { defaultOptions: options.defaultOptions } : {}),\n    });\n\n  const provider = ((modelId: OllamaChatModelId) => make(modelId)) as OllamaProvider;\n  provider.chat = make;\n  return provider;\n}\n\n/**\n * Default singleton — connects to `localhost:11434` (or\n * `OLLAMA_BASE_URL` if set). Suitable for the typical\n * `ollama serve` setup; build a custom provider with `createOllama()`\n * for remote / authenticated daemons.\n */\nexport const ollama: OllamaProvider = createOllama();\n\nfunction loadEnv(name: string): string | undefined {\n  if (typeof process !== 'undefined' && process.env) {\n    return process.env[name];\n  }\n  return undefined;\n}\n"]}

package/package.json

@@ -0,0 +1,75 @@
+{
+  "name": "@ziro-agent/ollama",
+  "version": "0.2.0",
+  "description": "Ollama provider for ZiroAgent SDK — run open-weight LLMs locally with the same interface as OpenAI / Anthropic. Sovereign pillar primitive.",
+  "license": "Apache-2.0",
+  "homepage": "https://ziroagent.com",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/ziroagent/sdk-typescript.git",
+    "directory": "packages/providers-ollama"
+  },
+  "bugs": {
+    "url": "https://github.com/ziroagent/sdk-typescript/issues"
+  },
+  "keywords": [
+    "ai",
+    "llm",
+    "ollama",
+    "local",
+    "open-weight",
+    "sovereign",
+    "ziroagent"
+  ],
+  "type": "module",
+  "sideEffects": false,
+  "main": "./dist/index.cjs",
+  "module": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": {
+        "types": "./dist/index.d.ts",
+        "default": "./dist/index.js"
+      },
+      "require": {
+        "types": "./dist/index.d.cts",
+        "default": "./dist/index.cjs"
+      }
+    },
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "dependencies": {
+    "@ziro-agent/core": "0.4.0"
+  },
+  "devDependencies": {
+    "@arethetypeswrong/cli": "^0.18.2",
+    "msw": "^2.11.7",
+    "publint": "^0.3.18",
+    "tsup": "^8.5.1",
+    "typescript": "^5.9.3",
+    "vitest": "^4.1.4"
+  },
+  "engines": {
+    "node": ">=20.10.0"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "author": "ZiroAgent <hello@ziroagent.com> (https://ziroagent.com)",
+  "scripts": {
+    "build": "tsup",
+    "dev": "tsup --watch",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "typecheck": "tsc --noEmit",
+    "clean": "rm -rf dist .turbo *.tsbuildinfo",
+    "publint": "publint",
+    "attw": "attw --pack . --profile=node16"
+  }
+}