Docker

Run AgenticCodingBench in Docker without installing Python.

Build the Image

docker build -t swarmone/agentic-coding-bench .

Run a Speed Benchmark

Mount a volume to save reports. Use host.docker.internal to reach services on your host machine.

docker run --rm -v $(pwd)/results:/results \
  swarmone/agentic-coding-bench speed \
  -e http://host.docker.internal:8000 \
  -m my-model \
  --suite full \
  -o /results/report.md

Recording Proxy

Run the recording proxy for agent mode:

docker-compose up proxy

Docker Compose

Set the endpoint and model as environment variables, then run:

export ACB_ENDPOINT=http://your-gpu-server:8000
export ACB_MODEL=your-model-name

docker-compose run agentic-coding-bench

Example docker-compose.yml

docker-compose.yml

version: "3.8"

services:
  agentic-coding-bench:
    image: swarmone/agentic-coding-bench
    environment:
      - ACB_ENDPOINT=${ACB_ENDPOINT}
      - ACB_MODEL=${ACB_MODEL}
      - ACB_API_KEY=${ACB_API_KEY:-}
    volumes:
      - ./results:/results
    command: >
      speed --suite full
      -o /results/report.md

  proxy:
    image: swarmone/agentic-coding-bench
    ports:
      - "19000:19000"
    environment:
      - ACB_ENDPOINT=${ACB_ENDPOINT}
      - ACB_MODEL=${ACB_MODEL}
    command: >
      record -P 19000
      -o /results/session.jsonl
    volumes:
      - ./results:/results

Volume Mounts

Host Path	Container Path	Purpose
$(pwd)/results	/results	Reports and JSON output
$(pwd)/workloads	/workloads	Recorded JSONL workloads
$(pwd)/config	/config	YAML configuration files

Networking Tips

Use host.docker.internal to reach services running on your host (macOS and Windows). On Linux, add --network host.
For the recording proxy, expose port 19000 and point your coding agent at http://localhost:19000.
GPU inference servers on the same Docker network can be reached by service name (e.g., http://vllm:8000).