ojob ojob.io/ai/miniagent goal=aValueExample mcp="(url: 'http://127.0.0.1:17879/mcp', type: remote, debug: false)" verbose=true maxsteps=25 rtm=20 readwrite=false checkall=true useshell=true outfile=/path/to/output.txt knowledge="You have access to a Linux shell and can run commands like ls, cat, grep, wc, curl, etc." conversation=/path/to/conversation.json libs=lib1,lib2 debug=false A mini autonomous agent that uses an LLM (use OAF_MODEL environment variable) and shell commands to achieve a specified goal in small steps.
Examples:
# Single MCP connection
ojob ojob.io/ai/miniagent goal="…" mcp="(cmd: 'docker run --rm -i mcp/dockerhub', timeout: 5000)" rtm=20
ojob ojob.io/ai/miniagent goal="…" mcp="(cmd: 'docker run --rm -i mcp/wikipedia-mcp', timeout: 5000)" rtm=20 knowledge="give final answer in markdown"
# Multiple MCP connections
ojob ojob.io/ai/miniagent goal="…" mcp="[(cmd: 'docker run --rm -i mcp/dockerhub', timeout: 5000), (cmd: 'docker run --rm -i mcp/wikipedia-mcp', timeout: 5000)]" rtm=20
| Name | Description |
|---|---|
| goal | Goal for the agent to achieve |
| mcp | A MCP connection object (or array of objects) using OpenAF's $mcp() function in SLON or JSON format |
| verbose | Whether to print detailed logs |
| maxsteps | Maximum number of steps to take |
| rtm | Rate of messages per minute |
| readwrite | Allow read-write commands without confirmation (dangerous) |
| checkall | Check all commands for banned operations (may ask for confirmation) |
| useshell | Whether to allow shell commands (default: true) |
| outfile | Output file path to save the final answer (if not provided, prints to console) |
| knowledge | Additional knowledge or context to provide to the agent (in the form of a string or a path to a text file) |
| conversation | Conversation history file to provide to the agent |
| libs | Comma-separated list of additional OJob libraries to load (e.g., "lib1,lib2") |
| debug | Enable debug mode |