Skip to content

docker.md

Latest commit

 

History

History
106 lines (78 loc) · 2.39 KB

docker.md

File metadata and controls

106 lines (78 loc) · 2.39 KB
Raw

Docker Deployment

AutoGPT ships two Docker images: kevinrsdev/autogpt for the agent binary and kevinrsdev/orchgpt for the orchestrator. A docker-compose.yml in the repository root wires them together.

Pre-built Images

Image Tag Size Purpose
kevinrsdev/autogpt 0.4.5 ~12 MB autogpt agent CLI
kevinrsdev/orchgpt 0.4.5 ~12 MB orchgpt orchestrator

Both images are Alpine-based and stripped of debug symbols for minimal footprint.

Running a Single Agent Container

docker run -it \
  -e GEMINI_API_KEY=<your_key> \
  -e AUTOGPT_WORKSPACE=/workspace \
  -v $(pwd)/workspace:/workspace \
  --rm --name autogpt \
  kevinrsdev/autogpt

The -v flag mounts a local directory so generated files persist after the container exits.

Running the Orchestrator Container

docker run -it \
  -e GEMINI_API_KEY=<your_key> \
  -p 8443:8443 \
  --rm --name orchgpt \
  kevinrsdev/orchgpt

Docker Compose (Recommended)

The docker-compose.yml in the repository root starts both services with shared networking:

services:
  autogpt:
    build:
      context: .
      dockerfile: Dockerfile.autogpt
    environment:
      - GEMINI_API_KEY=${GEMINI_API_KEY}
      - ORCHESTRATOR_ADDRESS=orchgpt:8443

  orchgpt:
    build:
      context: .
      dockerfile: Dockerfile.orchgpt
    environment:
      - GEMINI_API_KEY=${GEMINI_API_KEY}
    ports:
      - "8443:8443"

Start both:

docker compose up --build

Docker Compose sets up a bridge network so autogpt can reach orchgpt by container name. The ORCHESTRATOR_ADDRESS=orchgpt:8443 env var tells the agent where to connect.

Building Custom Images

To build with different feature flags, edit the Dockerfile and rebuild:

# Step 1: build autogpt image
docker build -f Dockerfile.autogpt -t my-autogpt .

# Step 2: build orchgpt image
docker build -f Dockerfile.orchgpt -t my-orchgpt .

# Step 3: run
docker run -i \
  -e GEMINI_API_KEY=<key> \
  -t my-autogpt

Accessing the Container Workspace

After starting a container:

# Find running container ID
docker ps

# Attach shell
docker exec -it <container_id> /bin/sh

# Explore workspace
ls workspace/
# architect/  backend/  designer/  frontend/

Stop all running containers:

docker stop $(docker ps -q)