A terminal-native coding agent with a rich TUI for humans and a plain CLI that AI agents can invoke directly — two interfaces, one tool, zero friction.
OpenMarv speaks two languages. Humans get a full interactive TUI — built for focus and flow. AI agents get a clean, scriptable CLI — built for automation and pipelines. Same coding engine underneath, zero configuration switching.
A beautiful, interactive terminal interface. Multi-line editing, syntax-aware diffs, streaming output, and session history — designed for developers who live in the terminal.
Agents get a clean, structured interface: parseable JSON output, non-interactive commands, stdin support. Built to be called by machines — no prompts, no noise.
OpenMarv reads your project structure, understands file dependencies, and tracks what it has already changed — so every response is grounded in the actual state of your code.
OpenAI, Anthropic, Ollama, or any OpenAI-compatible endpoint. Switch models per-session without touching config files.
Sessions survive reboots. Full context, file history, and conversation intact. Your work and your agent's work are never lost — pick up exactly where you left off.
If you've used Claude Code or OpenCode, you already know the mental model. Same file-aware coding experience — just with multi-agent superpowers layered on top.
OpenMarv's CLI is designed to be called by AI agents. Any agent that can run shell commands can invoke OpenMarv to handle coding tasks — no SDK, no custom integration required.
Hermes can invoke OpenMarv directly from its terminal to delegate complex coding tasks — refactors, test generation, debugging — and get structured results back.
OpenClaw skills can call OpenMarv for any coding task. Build a skill once and let Claw delegate implementation work to OpenMarv automatically.
If it can run a shell command, it can use OpenMarv. The CLI is minimal, composable, and outputs structured JSON — easy to parse in any pipeline.
Develop interactively in OpenMarv's TUI while Hermes or OpenClaw invoke the same tool via CLI to handle delegated tasks. Same coding engine — whichever interface fits the caller.
Set up OpenMarv in a project, tell Hermes to use it for test generation or refactoring. Your agent does the gruntwork via CLI while you focus on architecture in the TUI.
The --json flag makes OpenMarv's output machine-readable. Wire it into any automation pipeline — agents get clean, parseable results from every coding task.
OpenMarv is under active development. See the releases page on GitHub for the full and up-to-date changelog.
Install OpenMarv in one line. Give your agents a coding agent they can actually work with.