adding ollama adapter

README.md

@@ -187,6 +187,33 @@ const team = orchestrator.createTeam('mixed-team', {
 const result = await orchestrator.runTeam(team, 'Build a CLI tool that converts JSON to CSV.')
 ```
 
+### Local Ollama Support
+
+```typescript
+const orchestrator = new OpenMultiAgent({
+  defaultProvider: 'ollama',
+  defaultModel: 'llama2',
+})
+
+const localAgent: AgentConfig = {
+  name: 'assistant',
+  model: 'llama2',
+  provider: 'ollama',
+  systemPrompt: 'You are a local assistant running on Ollama.',
+  tools: ['bash', 'file_read', 'file_write'],
+}
+
+const team = orchestrator.createTeam('local-team', {
+  name: 'local-team',
+  agents: [localAgent],
+  sharedMemory: true,
+})
+
+const result = await orchestrator.runTeam(team, 'Create a small script that lists files in the current directory.')
+```
+
+Set `OLLAMA_API_KEY` when your local Ollama instance requires authentication. The adapter defaults to `http://localhost:11434`.
+
 ### Streaming Output
 
 ```typescript

examples/01-single-agent.ts

@@ -8,18 +8,22 @@
  *   npx tsx examples/01-single-agent.ts
  *
  * Prerequisites:
- *   ANTHROPIC_API_KEY env var must be set.
+ *   Ollama server running at http://localhost:11434
+ *   OLLAMA_API_KEY env var may be set if your server requires auth.
  */
 
 import { OpenMultiAgent, Agent, ToolRegistry, ToolExecutor, registerBuiltInTools } from '../src/index.js'
 import type { OrchestratorEvent } from '../src/types.js'
 
+const OLLAMA_MODEL = process.env.OLLAMA_MODEL ?? 'llama2'
+
 // ---------------------------------------------------------------------------
 // Part 1: Single agent via OpenMultiAgent (simplest path)
 // ---------------------------------------------------------------------------
 
 const orchestrator = new OpenMultiAgent({
-  defaultModel: 'claude-sonnet-4-6',
+  defaultProvider: 'ollama',
+  defaultModel: OLLAMA_MODEL,
   onProgress: (event: OrchestratorEvent) => {
     if (event.type === 'agent_start') {
       console.log(`[start]    agent=${event.agent}`)
@@ -34,10 +38,11 @@ console.log('Part 1: runAgent() — single one-shot task\n')
 const result = await orchestrator.runAgent(
   {
     name: 'coder',
-    model: 'claude-sonnet-4-6',
+    provider: 'ollama',
-    systemPrompt: `You are a focused TypeScript developer.
+    model: OLLAMA_MODEL,
-When asked to implement something, write clean, minimal code with no extra commentary.
+    systemPrompt: `You are a digital marketing agency.
-Use the bash tool to run commands and the file tools to read/write files.`,
+When asked to getting clients for our cybersecurity solution company, find some startup with funding, webscarp thier niche and collect there contact info.
+Use that info and send a cold eamil`,
     tools: ['bash', 'file_read', 'file_write'],
     maxTurns: 8,
   },
@@ -80,7 +85,8 @@ const executor = new ToolExecutor(registry)
 const streamingAgent = new Agent(
   {
     name: 'explainer',
-    model: 'claude-sonnet-4-6',
+    provider: 'ollama',
+    model: OLLAMA_MODEL,
     systemPrompt: 'You are a concise technical writer. Keep explanations brief.',
     maxTurns: 3,
   },
@@ -111,7 +117,8 @@ console.log('\nPart 3: Agent.prompt() — multi-turn conversation\n')
 const conversationAgent = new Agent(
   {
     name: 'tutor',
-    model: 'claude-sonnet-4-6',
+    provider: 'ollama',
+    model: OLLAMA_MODEL,
     systemPrompt: 'You are a TypeScript tutor. Give short, direct answers.',
     maxTurns: 2,
   },

package-lock.json

@@ -1,11 +1,11 @@
 {
-  "name": "maestro-agents",
+  "name": "open-multi-agent",
   "version": "0.1.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
-      "name": "maestro-agents",
+      "name": "open-multi-agent",
       "version": "0.1.0",
       "license": "MIT",
       "dependencies": {

src/index.ts

@@ -104,6 +104,7 @@ export {
 // ---------------------------------------------------------------------------
 
 export { createAdapter } from './llm/adapter.js'
+export { OllamaAdapter } from './llm/ollama.js'
 export type { SupportedProvider } from './llm/adapter.js'
 
 // ---------------------------------------------------------------------------

src/llm/adapter.ts

@@ -37,13 +37,13 @@ import type { LLMAdapter } from '../types.js'
  * Additional providers can be integrated by implementing {@link LLMAdapter}
  * directly and bypassing this factory.
  */
-export type SupportedProvider = 'anthropic' | 'openai'
+export type SupportedProvider = 'anthropic' | 'openai' | 'ollama'
 
 /**
  * Instantiate the appropriate {@link LLMAdapter} for the given provider.
  *
  * API keys fall back to the standard environment variables
- * (`ANTHROPIC_API_KEY` / `OPENAI_API_KEY`) when not supplied explicitly.
+ * (`ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_API_KEY`) when not supplied explicitly.
  *
  * Adapters are imported lazily so that projects using only one provider
  * are not forced to install the SDK for the other.
@@ -65,6 +65,10 @@ export async function createAdapter(
       const { OpenAIAdapter } = await import('./openai.js')
       return new OpenAIAdapter(apiKey)
     }
+    case 'ollama': {
+      const { OllamaAdapter } = await import('./ollama.js')
+      return new OllamaAdapter(apiKey)
+    }
     default: {
       // The `never` cast here makes TypeScript enforce exhaustiveness.
       const _exhaustive: never = provider

src/llm/ollama.test.ts

@@ -0,0 +1,141 @@
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { OllamaAdapter } from './ollama.js'
+import { createAdapter } from './adapter.js'
+
+const encoder = new TextEncoder()
+
+function createFetchMock(response: unknown): ReturnType<typeof vi.fn> {
+  return vi.fn(async () => ({
+    ok: true,
+    status: 200,
+    json: async () => response,
+    text: async () => JSON.stringify(response),
+    body: new ReadableStream<Uint8Array>({
+      start(controller) {
+        controller.close()
+      },
+    }),
+  }))
+}
+
+describe('OllamaAdapter', () => {
+  afterEach(() => {
+    vi.restoreAllMocks()
+  })
+
+  it('creates an Ollama adapter through createAdapter()', async () => {
+    const adapter = await createAdapter('ollama')
+    expect(adapter.name).toBe('ollama')
+  })
+
+  it('sends chat requests to the local Ollama endpoint', async () => {
+    const adapter = new OllamaAdapter(undefined, 'http://localhost:11434')
+    const mockResponse = {
+      id: 'abc123',
+      model: 'qwen',
+      choices: [
+        {
+          message: {
+            role: 'assistant',
+            content: [{ type: 'text', text: 'hello' }],
+          },
+          finish_reason: 'stop',
+        },
+      ],
+      usage: {
+        prompt_tokens: 1,
+        completion_tokens: 2,
+      },
+    }
+
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => ({
+      ok: true,
+      status: 200,
+      json: async () => mockResponse,
+      text: async () => JSON.stringify(mockResponse),
+      body: new ReadableStream<Uint8Array>({
+        start(controller) {
+          controller.close()
+        },
+      }),
+    }))
+
+    vi.stubGlobal('fetch', fetchMock)
+
+    const result = await adapter.chat(
+      [{ role: 'user', content: [{ type: 'text', text: 'Hello' }] }],
+      { model: 'qwen' },
+    )
+
+    expect(fetchMock).toHaveBeenCalledOnce()
+    const [, init] = fetchMock.mock.calls[0] as [string, RequestInit]
+    expect(init.method).toBe('POST')
+    expect(init?.headers).toEqual({ 'Content-Type': 'application/json' })
+    expect(typeof init?.body).toBe('string')
+    expect(JSON.parse(String(init?.body))).toMatchObject({
+      model: 'qwen',
+    })
+    expect(result.content).toEqual([{ type: 'text', text: 'hello' }])
+    expect(result.model).toBe('qwen')
+  })
+
+  it('parses streaming-style chat responses and ignores final empty chunks', async () => {
+    const adapter = new OllamaAdapter(undefined, 'http://localhost:11434')
+    const chunk = '{"model":"llama2","message":{"role":"assistant","content":"Hello"}}\n'
+      + '{"model":"llama2","message":{"role":"assistant","content":""}}\n'
+
+    const fetchMock = vi.fn(async () => ({
+      ok: true,
+      status: 200,
+      json: async () => ({}),
+      text: async () => chunk,
+      body: new ReadableStream<Uint8Array>({
+        start(controller) {
+          controller.close()
+        },
+      }),
+    }))
+
+    vi.stubGlobal('fetch', fetchMock)
+
+    const result = await adapter.chat(
+      [{ role: 'user', content: [{ type: 'text', text: 'Hello' }] }],
+      { model: 'llama2' },
+    )
+
+    expect(result.content).toEqual([{ type: 'text', text: 'Hello' }])
+    expect(result.model).toBe('llama2')
+  })
+
+  it('streams SSE events from Ollama and emits done', async () => {
+    const adapter = new OllamaAdapter(undefined, 'http://localhost:11434')
+    const chunk = `data: {"choices":[{"delta":{"content":"hi"}}]}\n\n` +
+      `data: [DONE]\n\n`
+
+    const fetchMock = vi.fn(async () => ({
+      ok: true,
+      status: 200,
+      text: async () => chunk,
+      body: new ReadableStream<Uint8Array>({
+        start(controller) {
+          controller.enqueue(encoder.encode(chunk))
+          controller.close()
+        },
+      }),
+    }))
+
+    vi.stubGlobal('fetch', fetchMock)
+
+    const events = [] as Array<unknown>
+    for await (const event of adapter.stream(
+      [{ role: 'user', content: [{ type: 'text', text: 'Hello' }] }],
+      { model: 'qwen' },
+    )) {
+      events.push(event)
+    }
+
+    expect(events.length).toBeGreaterThanOrEqual(2)
+    expect(events[0]).toEqual({ type: 'text', data: 'hi' })
+    expect((events[events.length - 1] as any).type).toBe('done')
+  })
+})

src/llm/ollama.ts

@@ -0,0 +1,534 @@
+/**
+ * @fileoverview Ollama adapter implementing {@link LLMAdapter}.
+ *
+ * Supports local Ollama servers via `/api/chat` and handles function calling
+ * with Ollama's OpenAI-compatible tool definition format.
+ *
+ * The adapter is intentionally lightweight: it uses the native Fetch API and
+ * parses both regular JSON responses and SSE streams from Ollama.
+ */
+
+import type {
+  ContentBlock,
+  LLMAdapter,
+  LLMChatOptions,
+  LLMMessage,
+  LLMResponse,
+  LLMStreamOptions,
+  LLMToolDef,
+  StreamEvent,
+  TextBlock,
+  ToolUseBlock,
+  ToolResultBlock,
+} from '../types.js'
+
+const DEFAULT_BASE_URL = 'http://localhost:11434'
+
+function hasToolResults(msg: LLMMessage): boolean {
+  return msg.content.some((block) => block.type === 'tool_result')
+}
+
+function toOllamaTextContent(msg: LLMMessage): string {
+  return msg.content
+    .filter((block) => block.type === 'text')
+    .map((block) => block.text)
+    .join('')
+}
+
+function toOllamaUserMessage(msg: LLMMessage): Record<string, unknown> {
+  const text = toOllamaTextContent(msg)
+  return {
+    role: 'user',
+    content: text || undefined,
+  }
+}
+
+function toOllamaToolMessages(msg: LLMMessage): Array<Record<string, unknown>> {
+  const messages: Array<Record<string, unknown>> = []
+
+  for (const block of msg.content) {
+    if (block.type !== 'tool_result') continue
+    messages.push({
+      role: 'tool',
+      tool_call_id: block.tool_use_id,
+      content: block.content,
+    })
+  }
+
+  return messages
+}
+
+function toOllamaAssistantMessage(msg: LLMMessage): Record<string, unknown> {
+  const text = toOllamaTextContent(msg)
+  const assistantMessage: Record<string, unknown> = {
+    role: 'assistant',
+    content: text || undefined,
+  }
+
+  const toolCalls = msg.content
+    .filter((block) => block.type === 'tool_use')
+    .map((block) => ({
+      id: block.id,
+      type: 'function',
+      function: {
+        name: block.name,
+        arguments: JSON.stringify(block.input),
+      },
+    }))
+
+  if (toolCalls.length > 0) {
+    assistantMessage.tool_calls = toolCalls
+  }
+
+  return assistantMessage
+}
+
+function toOllamaMessages(messages: LLMMessage[], systemPrompt?: string): Record<string, unknown>[] {
+  const result: Record<string, unknown>[] = []
+
+  if (systemPrompt) {
+    result.push({ role: 'system', content: systemPrompt })
+  }
+
+  for (const msg of messages) {
+    if (msg.role === 'assistant') {
+      result.push(toOllamaAssistantMessage(msg))
+      continue
+    }
+
+    if (!hasToolResults(msg)) {
+      result.push(toOllamaUserMessage(msg))
+      continue
+    }
+
+    const text = toOllamaTextContent(msg)
+    if (text.length > 0) {
+      result.push({ role: 'user', content: text })
+    }
+
+    result.push(...toOllamaToolMessages(msg))
+  }
+
+  return result
+}
+
+function toOllamaFunction(tool: LLMToolDef): Record<string, unknown> {
+  return {
+    name: tool.name,
+    description: tool.description,
+    parameters: tool.inputSchema,
+  }
+}
+
+function normalizeFinishReason(reason: unknown): string {
+  if (typeof reason !== 'string') {
+    return 'end_turn'
+  }
+
+  switch (reason) {
+    case 'stop':
+    case 'end_turn':
+      return 'end_turn'
+    case 'length':
+      return 'max_tokens'
+    case 'tool_call':
+    case 'function_call':
+      return 'tool_use'
+    default:
+      return reason
+  }
+}
+
+function isMeaningfulOllamaMessage(message: any): boolean {
+  if (!message || typeof message !== 'object') {
+    return false
+  }
+
+  const content = message.content
+  if (typeof content === 'string') {
+    return content.trim().length > 0
+  }
+
+  if (Array.isArray(content)) {
+    return content.length > 0
+  }
+
+  return false
+}
+
+function chooseBestResponseObject(objects: any[]): any {
+  let best: any = null
+
+  for (const obj of objects) {
+    const message = obj.message ?? obj.choices?.[0]?.message
+    if (isMeaningfulOllamaMessage(message)) {
+      best = obj
+    }
+  }
+
+  return best ?? objects[objects.length - 1]
+}
+
+function parseTextAsJson(text: string): any {
+  const trimmed = text.trim()
+  if (!trimmed) {
+    return null
+  }
+
+  try {
+    return JSON.parse(trimmed)
+  } catch {
+    // Fall through to SSE / NDJSON parsing.
+  }
+
+  const objects: any[] = []
+
+  // Try server-sent events blocks.
+  for (const chunk of trimmed.split(/\r?\n\r?\n/)) {
+    let payload = ''
+    for (const line of chunk.split(/\r?\n/)) {
+      if (line.startsWith('data:')) {
+        const value = line.slice(5).trim()
+        if (value === '[DONE]') {
+          continue
+        }
+        payload += value
+      }
+    }
+
+    if (payload.length === 0) {
+      continue
+    }
+
+    try {
+      objects.push(JSON.parse(payload))
+    } catch {
+      // ignore non-JSON payloads
+    }
+  }
+
+  if (objects.length > 0) {
+    return chooseBestResponseObject(objects)
+  }
+
+  // Try line-delimited JSON fallback.
+  for (const line of trimmed.split(/\r?\n/)) {
+    const candidate = line.trim()
+    if (!candidate) continue
+    try {
+      objects.push(JSON.parse(candidate))
+    } catch {
+      // ignore
+    }
+  }
+
+  if (objects.length > 0) {
+    return chooseBestResponseObject(objects)
+  }
+
+  return null
+}
+
+function parseToolCall(message: any): ToolUseBlock | null {
+  const toolCall = message?.tool_call ?? message?.tool_call?.arguments ? message?.tool_call : undefined
+  if (!toolCall || typeof toolCall.name !== 'string') {
+    return null
+  }
+
+  let input: Record<string, unknown> = {}
+  if (typeof toolCall.arguments === 'string') {
+    try {
+      const parsed = JSON.parse(toolCall.arguments)
+      if (parsed && typeof parsed === 'object' && !Array.isArray(parsed)) {
+        input = parsed as Record<string, unknown>
+      }
+    } catch {
+      // ignore malformed JSON
+    }
+  } else if (toolCall.arguments && typeof toolCall.arguments === 'object') {
+    input = toolCall.arguments as Record<string, unknown>
+  }
+
+  return {
+    type: 'tool_use',
+    id: toolCall.id ?? `${toolCall.name}:${Math.random().toString(16).slice(2)}`,
+    name: toolCall.name,
+    input,
+  }
+}
+
+function parseOllamaContent(message: any): ContentBlock[] {
+  if (typeof message === 'string') {
+    return [{ type: 'text', text: message }]
+  }
+
+  const content: ContentBlock[] = []
+  const items = Array.isArray(message.content) ? message.content : []
+
+  for (const item of items) {
+    if (item?.type === 'text' && typeof item.text === 'string') {
+      content.push({ type: 'text', text: item.text })
+    } else if (item?.type === 'tool_use' && typeof item.id === 'string' && typeof item.name === 'string') {
+      content.push({
+        type: 'tool_use',
+        id: item.id,
+        name: item.name,
+        input: item.input ?? {},
+      })
+    } else if (item?.type === 'tool_result' && typeof item.tool_use_id === 'string') {
+      content.push({
+        type: 'tool_result',
+        tool_use_id: item.tool_use_id,
+        content: typeof item.content === 'string' ? item.content : String(item.content ?? ''),
+        is_error: Boolean(item.is_error),
+      })
+    } else if (item?.type === 'image' && item.source) {
+      content.push({
+        type: 'image',
+        source: {
+          type: 'base64',
+          media_type: item.source.media_type ?? 'image/png',
+          data: item.source.data ?? '',
+        },
+      })
+    } else if (typeof item === 'string') {
+      content.push({ type: 'text', text: item })
+    }
+  }
+
+  return content.length > 0
+    ? content
+    : [{ type: 'text', text: String(message.content ?? '') }]
+}
+
+function buildOllamaResponse(body: any): LLMResponse {
+  const choice = Array.isArray(body.choices) && body.choices.length > 0
+    ? body.choices[0]
+    : null
+
+  const message = choice?.message ?? body.message ?? body
+  const content = parseOllamaContent(message)
+
+  const toolUse = parseToolCall(message)
+  if (toolUse) {
+    content.push(toolUse)
+  }
+
+  return {
+    id: body.id ?? choice?.id ?? '',
+    content,
+    model: body.model ?? choice?.model ?? 'ollama',
+    stop_reason: normalizeFinishReason(choice?.finish_reason ?? body.finish_reason),
+    usage: {
+      input_tokens: body.usage?.prompt_tokens ?? body.usage?.input_tokens ?? 0,
+      output_tokens: body.usage?.completion_tokens ?? body.usage?.output_tokens ?? 0,
+    },
+  }
+}
+
+function stripTrailingSlash(url: string): string {
+  return url.replace(/\/+$|$/, '')
+}
+
+/**
+ * Lightweight adapter for Ollama's local `/api/chat` endpoint.
+ */
+export class OllamaAdapter implements LLMAdapter {
+  readonly name = 'ollama'
+  readonly #baseUrl: string
+  readonly #apiKey?: string
+
+  constructor(apiKey?: string, baseUrl = DEFAULT_BASE_URL) {
+    const envApiKey = (globalThis as any).process?.env?.OLLAMA_API_KEY
+    this.#apiKey = apiKey ?? (typeof envApiKey === 'string' ? envApiKey : undefined)
+    this.#baseUrl = stripTrailingSlash(baseUrl)
+  }
+
+  private buildRequestBody(messages: LLMMessage[], options: LLMChatOptions): Record<string, unknown> {
+    const body: Record<string, unknown> = {
+      model: options.model,
+      messages: toOllamaMessages(messages, options.systemPrompt),
+      temperature: options.temperature ?? 1,
+      max_tokens: options.maxTokens,
+    }
+
+    if (options.tools) {
+      body.functions = options.tools.map(toOllamaFunction)
+      body.function_call = 'auto'
+    }
+
+    return body
+  }
+
+  private buildHeaders(): Record<string, string> {
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+    }
+
+    if (this.#apiKey) {
+      headers.Authorization = `Bearer ${this.#apiKey}`
+    }
+
+    return headers
+  }
+
+  // -------------------------------------------------------------------------
+  // chat()
+  // -------------------------------------------------------------------------
+
+  async chat(messages: LLMMessage[], options: LLMChatOptions): Promise<LLMResponse> {
+    const url = `${this.#baseUrl}/api/chat`
+    const requestBody = this.buildRequestBody(messages, options)
+    requestBody.stream = false
+
+    const response = await fetch(url, {
+      method: 'POST',
+      headers: this.buildHeaders(),
+      body: JSON.stringify(requestBody),
+      signal: options.abortSignal,
+    })
+
+    if (!response.ok) {
+      const text = await response.text()
+      throw new Error(`Ollama API request failed (${response.status}): ${text}`)
+    }
+
+    const text = await response.text()
+    const body = parseTextAsJson(text)
+    if (body === null) {
+      throw new Error(
+        `Ollama API returned invalid JSON response: ${text.slice(0, 200)}`,
+      )
+    }
+
+    return buildOllamaResponse(body)
+  }
+
+  // -------------------------------------------------------------------------
+  // stream()
+  // -------------------------------------------------------------------------
+
+  async *stream(messages: LLMMessage[], options: LLMStreamOptions): AsyncIterable<StreamEvent> {
+    const url = `${this.#baseUrl}/api/chat`
+    const requestBody = this.buildRequestBody(messages, options)
+    requestBody.stream = true
+
+    const response = await fetch(url, {
+      method: 'POST',
+      headers: this.buildHeaders(),
+      body: JSON.stringify(requestBody),
+      signal: options.abortSignal,
+    })
+
+    if (!response.ok) {
+      const text = await response.text()
+      throw new Error(`Ollama API request failed (${response.status}): ${text}`)
+    }
+
+    const reader = response.body?.getReader()
+    if (!reader) {
+      throw new Error('Ollama stream response has no body')
+    }
+
+    const decoder = new TextDecoder()
+    let buffer = ''
+    let lastMessage: any = null
+    let accumulatedText = ''
+
+    try {
+      while (true) {
+        const { value, done } = await reader.read()
+        if (done) {
+          break
+        }
+
+        buffer += decoder.decode(value, { stream: true })
+
+        while (true) {
+          const boundary = buffer.indexOf('\n\n')
+          if (boundary === -1) {
+            break
+          }
+
+          const packet = buffer.slice(0, boundary)
+          buffer = buffer.slice(boundary + 2)
+          const lines = packet.split(/\r?\n/)
+          let data = ''
+
+          for (const line of lines) {
+            if (line.startsWith('data:')) {
+              data += line.slice(5).trim()
+            }
+          }
+
+          if (data === '[DONE]') {
+            buffer = ''
+            break
+          }
+
+          if (!data) {
+            continue
+          }
+
+          try {
+            const payload = JSON.parse(data)
+            const choice = Array.isArray(payload.choices) && payload.choices.length > 0
+              ? payload.choices[0]
+              : null
+            const delta = choice?.delta ?? payload.delta
+
+            if (delta?.content) {
+              const text = String(delta.content)
+              accumulatedText += text
+              yield { type: 'text', data: text }
+            }
+
+            const toolDelta = delta?.tool_call ?? payload.tool_call
+            if (toolDelta || choice?.message?.tool_call || payload.message?.tool_call) {
+              const toolCallMessage = toolDelta ?? choice?.message?.tool_call ?? payload.message?.tool_call
+              const toolUse = parseToolCall({ tool_call: toolCallMessage })
+              if (toolUse) {
+                yield { type: 'tool_use', data: toolUse }
+              }
+            }
+
+            if (payload.message) {
+              lastMessage = payload.message
+            }
+            if (choice?.message) {
+              lastMessage = choice.message
+            }
+          } catch {
+            // Ignore malformed SSE payloads.
+          }
+        }
+      }
+
+      let finalMessage: any = lastMessage ?? {
+        id: '',
+        model: options.model,
+        message: { content: [{ type: 'text', text: accumulatedText }] },
+        usage: { prompt_tokens: 0, completion_tokens: 0 },
+        finish_reason: 'stop',
+      }
+
+      const isEmptyMessage = finalMessage?.message &&
+        ((typeof finalMessage.message.content === 'string' && finalMessage.message.content.trim() === '') ||
+          (Array.isArray(finalMessage.message.content) && finalMessage.message.content.length === 0))
+
+      if (isEmptyMessage && accumulatedText.length > 0) {
+        finalMessage = {
+          ...finalMessage,
+          message: { content: [{ type: 'text', text: accumulatedText }] },
+        }
+      }
+
+      const finalResponse: LLMResponse = buildOllamaResponse(finalMessage)
+
+      yield { type: 'done', data: finalResponse }
+    } catch (err) {
+      const error = err instanceof Error ? err : new Error(String(err))
+      yield { type: 'error', data: error }
+    }
+  }
+}

src/types.ts

@@ -186,7 +186,7 @@ export interface ToolDefinition<TInput = Record<string, unknown>> {
 export interface AgentConfig {
   readonly name: string
   readonly model: string
-  readonly provider?: 'anthropic' | 'openai'
+  readonly provider?: 'anthropic' | 'openai' | 'ollama'
   readonly systemPrompt?: string
   /** Names of tools (from the tool registry) available to this agent. */
   readonly tools?: readonly string[]
@@ -285,7 +285,7 @@ export interface OrchestratorEvent {
 export interface OrchestratorConfig {
   readonly maxConcurrency?: number
   readonly defaultModel?: string
-  readonly defaultProvider?: 'anthropic' | 'openai'
+  readonly defaultProvider?: 'anthropic' | 'openai' | 'ollama'
   onProgress?: (event: OrchestratorEvent) => void
 }
 

tsconfig.json

@@ -3,7 +3,8 @@
     "target": "ES2022",
     "module": "ESNext",
     "moduleResolution": "bundler",
-    "lib": ["ES2022"],
+    "lib": ["ES2022", "DOM"],
+    "types": ["node"],
     "outDir": "dist",
     "rootDir": "src",
     "declaration": true,