Opencognit

OpenCognit

Every coding agent. One window.

Run Claude Code, Codex, Gemini CLI, Kimi and Hermes side by side, let one of them orchestrate the others, and have every result checked before it reaches your project. Locally, with the subscriptions you already have.

⬇ Download for Linux · ⬇ Download for Windows · .deb · All releases · Deutsch
This repository used to host the open-source OpenCognit agent platform. That project is discontinued. Its code stays available under the MIT license in the legacy branch. OpenCognit continues as the desktop app described here, which is a commercial product and not open source. This repository is now its public home: releases, issues, and discussion.

What it does

- Many agents, one place. Real PTY sessions for each CLI, in a focus view or a grid. Several logins per CLI (work and private) stay separate.
- An orchestrator that hands out work. One agent starts others, gives each a numbered task and its own set of files, and is told when they report back. Agents cannot write outside their files, and two open tasks cannot overlap.
- Guardrails before anything runs. Every shell command an agent wants to run is checked against your rules: block, ask you, or warn. Commits are scanned for secrets first.
- See what is happening. A live graph shows which agent reads or changes which file and warns when two work in the same place.
- Work that survives. Sessions resume after a restart, checkpoints let you jump back, scheduled and overnight tasks run in their own worktree and leave a report.
- Your usage, added up. Tokens of all CLIs from their local logs, per day, model and project.

A look inside

Neurons: every agent is a ring at the folder it works in. Pulses run along the branches in the agent's colour when it reads or changes a file.

The…