@avasis-ai/agent-runtime

Runtime engine for Avasis terminal agents. Run persistent background agents locally with small models via Ollama.

Install

npm install -g @avasis-ai/agent-runtime

Quick Start

# 1. Install Ollama and pull a model
ollama pull phi4-mini:3.8b

# 2. Create an agent config (or use https://avasis.ai/builder)
avasis-agent run avasis.config.json

# 3. Run in background
avasis-agent run avasis.config.json --daemon

# 4. Check status
avasis-agent status

Commands

| Command | Description | |---------|-------------| | run [config] | Run an agent from a JSON config file | | status | Show agent state and recent runs | | validate [config] | Validate a config file without running | | list-models | List available Ollama models | | pull <model> | Pull a model from Ollama |

Flags

| Flag | Description | |------|-------------| | --verbose | Show tool calls and internal logs | | --daemon | Run in background (detached process) | | --ollama-url <url> | Custom Ollama base URL |

Config Format

{
  "id": "my-agent",
  "version": "1.0.0",
  "type": "single",
  "ollamaBaseUrl": "http://localhost:11434",
  "agent": {
    "name": "file-organizer",
    "model": "phi4-mini:3.8b",
    "systemPrompt": "Organize files by type into subdirectories.",
    "tasks": [
      {
        "name": "organize",
        "prompt": "List files in /tmp/workspace and sort them into images/, docs/, code/ subdirectories.",
        "tools": ["fs_list", "fs_move"],
        "maxTokens": 2048
      }
    ],
    "trigger": { "type": "interval", "value": "300000" }
  }
}

Built-in Tools

| Tool | Description | |------|-------------| | fs_read | Read file contents | | fs_write | Write content to a file | | fs_list | List directory contents (supports recursive) | | fs_move | Move or rename files | | fs_delete | Delete a file | | shell_exec | Execute shell commands | | git_status | Get git repository status | | git_diff | Get git diff of changes | | git_commit | Stage all changes and commit |

How It Works

1. Loads agent config (JSON file)
2. Connects to local Ollama instance
3. Runs tasks with native tool calling (model decides which tools to use)
4. Tool results are fed back to the model for reasoning
5. State is persisted to .avasis/state.json across runs
6. For interval/daemon mode: runs the task loop forever in the background

License

MIT