Start with an open-source AI coding assistant on your computer

Install and open the workbench

Use the installer below, or download the client for your operating system from GitHub Releases. The installer verifies SHA-256, keeps the complete release in your user directory, starts the service and opens your browser.

Prebuilt releases need no Rust or Node.js. File search requires ripgrep; restricted command execution on Linux also requires bubblewrap. The installer does not install these system dependencies.

Connect your project and model

1. Select a project folder and create a session.
2. In Settings → Models, add your Provider endpoint, protocol and any required API key, then add a real model ID.
3. Select the model beneath the session composer. Start with “List the main project directories and suggest where to begin reading.”
4. Check the result, then request a specific change and the checks already defined by your project.

Local use versus a self-hosted Server

The local ternilo serve service works on the current computer. Add a Ternilo Server when you need phone access, several machines or shared sessions. The connected computer continues to run its own file operations and tasks.

Ternilo Server provides the official Docker image ghcr.io/mosttt/ternilo-server. Pull and deploy it with the official Compose configuration, without building locally. See the Docker Compose guide from the latest release below for image versions, initial setup, persistent data and HTTPS configuration.

Where do the work and data live?

Closing the browser does not stop a running local service. Keep the terminal open for a manually started service; Ctrl-C or ternilo stop stops it. Project files and Ternilo session data are stored separately and need separate backups.

Model requests go to your selected Provider. Self-hosting the assistant does not mean inference runs locally. Local inference requires an available local model endpoint with a compatible protocol.