Everything on this page is optional reading โ€” Setup covers the normal path.

Host-managed setup

Run everything directly on your host instead of inside the devcontainer. You need Git, Node.js 24, Python 3.13, and Docker (for Postgres, Redis, and MinIO).
  1. Run the contributor bootstrap. It configures origin as your fork and upstream as eesi-ai/eesi, initializes the pipecat submodule, creates the Python venv, and copies the .env templates:
  1. Activate the virtual environment:
  1. Install Python requirements:
  1. Install UI dependencies (make sure node --version reports 24, e.g. via nvm use 24):
  1. Start Postgres, Redis, MinIO, Ollama, and Speaches, then the app:
Or step by step: make infra, make infra-init, make wait, make backend, make ui. See make help.
  1. Open http://localhost:3000.

Devcontainer internals

The checked-in .devcontainer/ follows the Dev Containers specification, so it should also work in Cursor, JetBrains IDEs, and other compatible editors. Only the VS Code path is actively tested โ€” if something breaks elsewhere, please open a GitHub issue.

What the bootstrap does

While the container starts, the devcontainer will:
  • Initialize the pipecat git submodule on the host through initializeCommand
  • Bring up postgres, redis, minio, ollama, and speaches through Docker Compose and wait for them to be healthy
  • Run one-shot ollama-init and speaches-init jobs to pull the default LLM, STT, and TTS models before the workspace starts
  • Seed the venv named volume from the image-baked Python 3.13 venv
  • Reinstall pipecat as editable from the bind-mounted submodule so source edits take effect
  • Create api/.env, api/.env.test, and ui/.env from their *.example templates if they do not already exist
  • Run npm ci for ui/ and api/mcp_server/ts_validator/
In the API env files, localhost for Postgres, Redis, and MinIO is rewritten to the docker network aliases postgres, redis, and minio. MINIO_PUBLIC_ENDPOINT deliberately stays on localhost because the browser on your host loads from it.
If you already had an api/.env or api/.env.test from host-managed development with localhost hosts for Postgres, Redis, or MinIO, the bootstrap leaves it untouched. Edit those URLs to the docker network aliases before starting the backend inside the devcontainer, or delete the file and let the bootstrap recreate it on the next rebuild.

When to rebuild the container

The workspace bind mount and the venv / node_modules named volumes persist across container restarts, so you rarely need to rebuild. Rebuild only when one of these changes:
  • .devcontainer/Dockerfile or .devcontainer/devcontainer.json
  • api/pyproject.toml or api/uv.lock
  • The pipecat/ submodule
Plain source edits, ui/package.json, and the .env.example templates do not require a rebuild. After a git pull, a quick check:
If the diff is empty, you can keep your current container.

Personal install hook

Anything you install inside the container outside the named volumes, notably under /home/vscode, is wiped on rebuild. This includes npm i -g packages and tools like the Claude or Codex CLIs. To reinstall personal tooling automatically on every rebuild, drop an executable script at .devcontainer/install.local.sh. It is gitignored and runs at the tail of the post-create bootstrap. Example:
Keep entries idempotent with command -v or a marker file so re-runs are cheap and safe.

Dev Container CLI

To use the devcontainer without VS Code, install the CLI once on your host (requires Node.js):
Start or rebuild the container from the repository root:
Open a shell in the running workspace container:
Run a one-off command from the host, e.g. starting the backend or the UI:

Fork and upstream remotes

The contributor bootstrap (scripts/setup_fork.sh) configures two remotes:
  • origin: your fork, where you push
  • upstream: eesi-ai/eesi, where new commits land
If you cloned eesi-ai/eesi directly instead of your fork, run it once (inside the devcontainer is fine):
To pull in upstream changes:
Check your remotes any time with git remote -v. You should see:
Always push feature branches to origin (your fork), then open a pull request against eesi-ai/eesi:main. Never push directly to upstream.