concai

concai is the command-line companion for running Concordance locally. It bootstraps the Python SDK, installs the inference engine, and uploads mods to a running server.

Installation

cargo install concai

This installs the concai binary into Cargo's bin directory. You can also run it from a checkout with cargo run -- <command>.

Requirements

• uv must be available on PATH. concai uses it to create the virtual environment and install Python wheels.
• macOS and Linux users can install uv with brew install uv or the official install script. Windows users can run winget install astral-sh.uv.

Quickstart

1. Bootstrap the project

   concai init
   

   This installs the Concordance SDK into the managed virtualenv, writes a .env file (use --env-file to change the location), and creates a starter mod at mods/hello_world.py. Re-run with --force to overwrite existing files.

2. Install the engine

   concai engine install
   

   The engine is installed into ./.venv using the wheel specified in the embedded versions.toml. Provide a different artifact with concai engine install --wheel <path-or-url>.

3. Set up credentials

   Update the generated .env with any required values (for example, set HF_TOKEN for gated Hugging Face downloads).

4. Run the server

   concai engine serve
   

   Runs the OpenAI-compatible server on 0.0.0.0:8000. Change the host, port, or env file with --host, --port, and --env-file. The command exits early if the engine environment is missing; install it with the previous step.

5. Upload mods

   concai mod upload --file-name mods/hello_world.py
   # or bundle a directory
   concai mod upload --dir mods/my_project
   

   Entry points are detected automatically from @mod-decorated functions. Successful registrations are printed, along with instructions for enabling a mod via /v1/chat/completions.

Command Overview

• concai version — prints a JSON blob containing the CLI version and the embedded release metadata.
• concai init — installs the SDK, scaffolds .env, and writes a sample mod.
• concai engine install — creates the engine virtualenv (if needed) and installs the engine wheel.
• concai engine serve — launches the local inference server using the managed virtualenv.
• concai mod upload — submits single-file or directory-based mods to a running server.

Use concai <command> --help for the complete list of flags. Advanced users can point the CLI at a different versions.toml by setting CONCORD_VERSIONS_PATH before building or running.