DockerREPL

Docker container execution environment

DockerREPL executes Python code in a Docker container running on the same host machine as the RLM process. Each code execution runs in an isolated container environment with its own filesystem, network namespace, and process tree, providing better security and reproducibility than LocalREPL. The container requests LM calls from the host’s LM Handler when code executes llm_query(). This environment is ideal for CI/CD pipelines, reproducible execution environments, and scenarios requiring stronger isolation than LocalREPL while maintaining the convenience of local execution.

Prerequisite: Docker must be installed and running.

rlm = RLM(
    backend="openai",
    backend_kwargs={"model_name": "gpt-5-mini"},
    environment="docker",
    environment_kwargs={
        "image": "python:3.11-slim",
    },
)

Arguments

Argument	Type	Default	Description
image	str	"python:3.11-slim"	Docker image to use
setup_code	str	None	Code to run at initialization
context_payload	str | dict | list	Auto	Initial context (set by RLM)
lm_handler_address	tuple	Auto	Socket address (set by RLM)

How It Works

1. Starts Docker container with volume mount to temp directory
2. Installs dill and requests in container
3. Host runs HTTP proxy server on random port
4. Container calls proxy via host.docker.internal
5. Proxy forwards llm_query() to LM Handler via socket
6. State persisted via dill to /workspace/state.dill

┌────────────────────────────────────────┐
│ Host                                   │
│  ┌────────────┐ Socket ┌────────────┐ │
│  │ HTTP Proxy │◄──────►│ LM Handler │ │
│  └─────┬──────┘        └────────────┘ │
└────────┼───────────────────────────────┘
         │ HTTP
┌────────┼───────────────────────────────┐
│ Docker │ Container                     │
│  ┌─────▼──────┐                        │
│  │   Python   │ llm_query() → proxy    │
│  │   exec()   │                        │
│  └────────────┘                        │
└────────────────────────────────────────┘

Custom Image

You can use your own custom Docker images or update the given image. Pre-install dependencies:

FROM python:3.11-slim
RUN pip install numpy pandas dill requests

environment_kwargs={"image": "my-rlm-image"}