Sandbox quickstart

RAGFlow's CodeExec agent component needs a sandbox provider to run Python and JavaScript code.

The simplest setup flow is:

Start the required sandbox services.
Open the RAGFlow admin page.
Go to Admin > Sandbox Settings.
Choose a provider and save the configuration.
Test the connection in the same page.

Admin page

Configure sandbox providers from the admin page:

self_managed: Uses the executor manager service.
ssh: Runs code on a remote machine over SSH.
aliyun_codeinterpreter, e2b, and tenki: Cloud providers.

Provider options

RAGFlow supports multiple sandbox providers. Configure the active provider in Admin > Sandbox Settings after the services are up.

self_managed: Runs code inside Docker-managed sandbox containers. This is the default provider.
ssh: Runs code on a remote machine over SSH.
aliyun_codeinterpreter and e2b: Cloud-hosted providers that remain available in the admin provider list.
tenki: Cloud-hosted provider that runs each execution in a disposable Tenki microVM. See Tenki below.

Tenki

tenki runs each code execution in a fresh Tenki microVM and destroys it afterwards. It is cloud-hosted, so it needs no local sandbox services, gVisor, or Docker base images — only outbound network access and an API key.

Configure it in Admin > Sandbox Settings:

api_key (required): Tenki API key. Create one at app.tenki.cloud under API Keys.
project_id (required): the Tenki project that sandboxes are created under.
base_url (optional): override the Tenki API endpoint.
image (optional): sandbox base image. Leave empty to use the Tenki default image, which includes python3 and node.
allow_outbound (optional, security-relevant): whether the sandbox may make outbound network connections. Defaults to false so sandboxed code has no network access; set it to true when code needs the network (for example, to install packages).
timeout, max_lifetime, cpu_cores, memory_mb, disk_size_gb, and the output/artifact limits have sensible defaults and can be tuned in the same page.

Notes:

Supported languages are Python and JavaScript.
Files written to the artifacts/ directory of the working directory are returned as run artifacts.
The provider uses only Tenki's create/exec/destroy operations; it does not use volumes or snapshots.

Self-Managed prerequisites

The following prerequisites apply only when you use the self_managed provider:

Linux distribution compatible with gVisor.
gVisor installed and configured.
Docker version 25.0 or higher (API 1.44+). Ensure your executor manager image ships with Docker CLI 29.1.0 or higher to stay compatible with the latest Docker daemons.
Docker Compose version 2.26.1 or higher (similar to RAGFlow requirements).
uv package and project manager installed.
(Optional) GNU Make for simplified command-line management.

NOTE

The error message client version 1.43 is too old. Minimum supported API version is 1.44 indicates that your executor manager image's built-in Docker CLI version is lower than 29.1.0 required by the Docker daemon in use.

Build Docker base images

The sandbox uses isolated base images for secure containerized execution environments.

Option 1: Build from source

Build the runtime base images:

docker build -t sandbox-base-python:latest ./sandbox_base_image/python
docker build -t sandbox-base-nodejs:latest ./sandbox_base_image/nodejs

Alternatively, build all base images at once using the Makefile:

make build

Build the executor manager image:

docker build -t sandbox-executor-manager:latest ./executor_manager

Option 2: Pull base images from Docker Hub

If you do not need to customize runtime dependencies, pull the published base images and tag them with the names used by standalone Docker Compose:

docker pull infiniflow/sandbox-base-python:latest
docker pull infiniflow/sandbox-base-nodejs:latest

docker tag infiniflow/sandbox-base-python:latest sandbox-base-python:latest
docker tag infiniflow/sandbox-base-nodejs:latest sandbox-base-nodejs:latest

Then restart the standalone sandbox services:

# Stop the standalone Sandbox services before applying the changes.
docker compose -f docker-compose.yml down
# Start the standalone Sandbox services in the background.
docker compose -f docker-compose.yml up -d

Running with RAGFlow

If you use the self_managed provider, verify that gVisor is installed and operational.
Configure docker/.env:

Include sandbox in COMPOSE_PROFILES if you want the default self_managed executor-manager service.
If you use the ssh provider, set RAGFLOW_SSH_KNOWN_HOSTS to the host known_hosts file and keep SSH_KNOWN_HOSTS as the path used inside the Go container. The default values are ${HOME}/.ssh/known_hosts and /etc/ragflow/ssh_known_hosts.
Keep the self-managed deployment defaults in .env if you need to change the sandbox-executor-manager image, pool size, base images, seccomp, memory, or timeout.

Add the following entry to your /etc/hosts file to resolve the executor manager service:
```
127.0.0.1 es01 infinity mysql minio kvrocks sandbox-executor-manager
```
Start the RAGFlow service as usual.
Open Admin > Sandbox Settings.
Select a provider.
Fill in the required fields.
Click Save.
Click Test Connection if needed.

Environment variables

The variables in docker/.env are grouped by scope.

System-level variables

These variables apply to the sandbox deployment:

COMPOSE_PROFILES: Include sandbox to start the default self-managed executor-manager service.
SANDBOX_ARTIFACT_BUCKET: MinIO bucket used for files generated by sandbox code.

Self-managed deployment defaults

These variables are shown in Admin as deployment defaults for self_managed. Changing them requires restarting sandbox-executor-manager.

SANDBOX_EXECUTOR_MANAGER_IMAGE: Docker image for the executor manager service. Defaults to infiniflow/sandbox-executor-manager:latest.
SANDBOX_EXECUTOR_MANAGER_POOL_SIZE: Number of Python and Node.js sandbox containers kept in each runtime pool. Defaults to 3.
SANDBOX_BASE_PYTHON_IMAGE: Python runtime image used by executor-managed containers. Defaults to infiniflow/sandbox-base-python:latest.
SANDBOX_BASE_NODEJS_IMAGE: Node.js runtime image used by executor-managed containers. Defaults to infiniflow/sandbox-base-nodejs:latest.
SANDBOX_EXECUTOR_MANAGER_PORT: Host port exposed by the executor manager. Defaults to 9385 and binds to 127.0.0.1 in the provided Compose configuration.
SANDBOX_ENABLE_SECCOMP: Enables the optional seccomp profile for sandbox containers. Defaults to false.
SANDBOX_MAX_MEMORY: Memory limit for each sandbox runtime container. Defaults to 256m; it is not the total memory limit for all sandbox containers.
SANDBOX_CONTAINER_NETWORK: Docker network attached to each sandbox runtime container. Defaults to none, which prevents external network access. Use bridge only when sandboxed code needs outbound network access.
SANDBOX_TIMEOUT: Default execution timeout. Defaults to 10s.
SANDBOX_EXECUTOR_MANAGER_API_TOKEN: Shared secret used to authenticate requests from RAGFlow to the executor manager. Set a strong value for network-accessible deployments; the same value is passed to both services by the provided Compose configuration.

Admin-managed runtime settings

Provider selection and runtime settings are configured in Admin > Sandbox Settings.

Examples:

Choose the active provider
Configure self_managed runtime settings
Configure all ssh settings

For self_managed:

Runtime settings are editable in Admin
Deployment defaults come from .env and are shown as read-only values

Running standalone

Manual setup

Initialize the environment variables:
```
cp .env.example .env
```
Launch the sandbox services with Docker Compose:
```
docker compose -f docker-compose.yml up
```

Test the sandbox setup:

source .venv/bin/activate
export PYTHONPATH=$(pwd)
uv pip install -r executor_manager/requirements.txt
uv run tests/sandbox_security_tests_full.py

Using Makefile

Run all setup, build, launch, and tests with a single command:

make

Monitoring

To follow logs of the executor manager container:

docker logs -f sandbox-executor-manager

Or use the Makefile shortcut:

make logs