AutoGen Code Execution

Opt in to sandboxed local code execution for framework: autogen with Docker on by default

Enable sandboxed local code execution for framework: autogen runs via a single YAML flag — off by default so LLM-generated code never runs on your host without consent.

Quick Start

1. Simple opt-in (Docker on)

   Set code_execution: true — the wrapper defaults work_dir to coding and use_docker to True.

   framework: autogen
   topic: Plot a sine wave and save it as sine.png
   config:
     autogen:
       code_execution: true

   export OPENAI_API_KEY=sk-...
   praisonai agents.yaml --framework autogen

2. Custom work_dir

   Pass a dict to override the sandbox directory — use_docker still defaults to True.

   framework: autogen
   topic: Analyse sales.csv and print totals
   config:
     autogen:
       code_execution:
         work_dir: "sandbox"
         use_docker: true

3. Local execution (unsafe, dev only)

   Setting use_docker: false runs generated code directly on the host and logs a warning.

   framework: autogen
   config:
     autogen:
       code_execution:
         use_docker: false   # LLM code runs on the host — dev only

How It Works

The wrapper resolves code_execution into AutoGen’s code_execution_config before it builds the UserProxyAgent.

The user_proxy is constructed with the resolved code_execution_config and the resolved human_input_mode. When the LLM emits a code block, the user_proxy runs it in the sandbox path; when nothing is opted in, code_execution_config is False and no code runs.

Configuration Options

Every option lives under config.autogen.* in your team YAML.

Precedence Ladder

code_execution accepts a bool for the common case and a dict for overrides.

# Level 1: Bool (simplest — off by default)
framework: autogen
config:
  autogen:
    code_execution: true   # Docker on, work_dir=coding

# Level 2: Dict (custom overrides)
framework: autogen
config:
  autogen:
    code_execution:
      work_dir: "sandbox"
      use_docker: true

# Level 3: Override human_input_mode
framework: autogen
config:
  autogen:
    code_execution: true
    human_input_mode: "ALWAYS"   # or NEVER, TERMINATE (default)

Failure Scenario

When use_docker: false the wrapper emits this warning before running code on the host:

framework=autogen: code_execution is enabled with use_docker=False;
LLM-generated code will run directly on the host.
Set config.autogen.code_execution.use_docker: true to sandbox.

Best Practices

Related