lite-harness
Unified API for running OpenCode, Claude Code, Codex agents
A Docker server that lets teams deploy and run AI coding agents like Claude Code, Copilot, and Codex from one shared place, with scheduling, a secrets vault, sandboxed execution, and a human-approval inbox so each team member does not need their own agent setup.
Lite-harness is a server that lets teams run AI coding agents from a single shared place instead of having each person set them up individually. It wraps several popular AI coding tools, including Claude Code, OpenCode, GitHub Copilot, and Codex, under one consistent interface so they all behave the same way regardless of which tool is underneath.
The main idea is that you describe what you want an agent to do in plain language, and the server takes care of deploying it, scheduling it to run on a timer, storing any secret credentials it needs in a vault, and running it in an isolated sandbox environment so it cannot affect other things on your machine or server. If you want a human to approve the agent's actions before they happen, there is an Inbox interface built in for that. Agents can also be reached through a Slack integration or directly through an API.
Setting it up involves running a Docker container and connecting it to a LiteLLM gateway, which is a separate tool that routes AI requests to whichever language model provider you prefer. Once running, team members can install a command-line skill and use slash commands inside their AI coding tool to deploy agents without touching server configuration.
The project was built because the maintainers found running multiple AI coding servers separately was difficult: different APIs, different session handling, different ways to pass configuration. Combining them into one Docker image with shared session storage and shared tool access was the practical fix.
By default, agent session history is lost when the server restarts. Mounting a persistent storage volume solves that. The license is MIT.
Where it fits
- Give your whole team a shared server for deploying AI coding agents without each person setting up their own infrastructure.
- Schedule an AI agent to run on a timer and store its API credentials in a built-in vault rather than hardcoding them.
- Route agent task requests through Slack or a REST API instead of requiring everyone to use the command line.
- Require a human to approve an agent's proposed actions in an Inbox before they are carried out.