{
  "$comment": [
    "Ollama Agent — runs entirely locally with no GitHub Copilot auth required.",
    "",
    "This example shows how to point the wrapper at a locally-running Ollama instance",
    "instead of the GitHub Copilot API (BYOK: Bring Your Own Key).",
    "",
    "IMPORTANT — model requirements:",
    "  The agent runtime drives an agentic loop that depends on NATIVE tool calling",
    "  (function calling). The model MUST support structured tool calls, not just text.",
    "  Small/older models (e.g. qwen2.5-coder:7b) emit tool calls as plain-text JSON",
    "  and break the loop — you'll see a raw {\"name\":...} blob instead of an answer.",
    "  Use a model with robust tool-calling support. Verified working: qwen3.6.",
    "  See the 'Bring Your Own Model (BYOK)' section of the README for details.",
    "",
    "Prerequisites:",
    "  1. Install Ollama:  https://ollama.com",
    "  2. Pull a model:    ollama pull qwen3.6",
    "  3. Start Ollama:    ollama serve   (it listens on http://localhost:11434 by default)",
    "",
    "To run:",
    "  ./agents/ollama/start.sh start",
    "",
    "To use a different model, change copilot.model and make sure you have pulled it,",
    "and that it supports tool calling:",
    "  ollama pull llama3.3",
    "  # then set \"model\": \"llama3.3\" below",
    "",
    "You can also override model and provider entirely via environment variables,",
    "using the same vars as the gh-copilot CLI:",
    "  export COPILOT_PROVIDER_BASE_URL=http://localhost:11434/v1",
    "  export COPILOT_MODEL=qwen3.6",
    "  ./agents/ollama/start.sh start"
  ],

  "agentCard": {
    "name": "Ollama Agent",
    "description": "A locally-running A2A agent powered by Ollama. No GitHub Copilot account required.",
    "version": "1.0.0",
    "protocolVersion": "0.3.0",

    "defaultInputModes": ["text"],
    "defaultOutputModes": ["text"],
    "streaming": true,

    "skills": [
      {
        "id": "general-assistant",
        "name": "General Assistant",
        "description": "Answer questions, reason through problems, and assist with general tasks using a locally-running model",
        "tags": ["general", "qa", "reasoning", "local", "ollama"]
      }
    ]
  },

  "server": {
    "$comment": "Using port 3002 so this agent can run alongside the example (3000) and filesystem-assistant (3001).",
    "port": 3002,
    "hostname": "0.0.0.0",
    "advertiseHost": "localhost",
    "advertiseProtocol": "http"
  },

  "copilot": {
    "$comment": [
      "provider — points the SDK at your local Ollama instance instead of GitHub Copilot.",
      "",
      "  type: 'openai'  — Ollama exposes an OpenAI-compatible API.",
      "  baseUrl         — Ollama's default endpoint. Append /v1 for the OpenAI-compatible path.",
      "  wireApi         — Use 'completions' (Chat Completions API) for broadest Ollama model",
      "                    compatibility. Avoid 'responses' for Ollama: many models reject it",
      "                    with a 'does not support thinking' error.",
      "",
      "  No apiKey is needed — Ollama does not require authentication.",
      "",
      "model — must match a model you pulled with 'ollama pull' (run 'ollama list').",
      "        It MUST support native tool calling (see the $comment block at the top",
      "        of this file and the README BYOK section). Verified working: qwen3.6."
    ],
    "model": "qwen3.6:latest",
    "streaming": true,
    "systemPromptMode": "replace",
    "systemPrompt": "You are a helpful local AI assistant. Answer questions clearly and concisely. You run entirely on the user's local machine via Ollama.",
    "provider": {
      "type": "openai",
      "baseUrl": "http://localhost:11434/v1",
      "wireApi": "completions"
    }
  },

  "features": {
    "streamArtifactChunks": false
  },

  "timeouts": {
    "$comment": "Local models can be slower than cloud APIs, especially on first run. 300 s gives headroom.",
    "prompt": 300000
  },

  "logging": {
    "$comment": "Use 'debug' if you need to trace SDK-level events during development.",
    "level": "info"
  }
}