Magnus

Magnus

[](https://melpa.org/#/magnus) [](https://github.com/hrishikeshs/magnus/actions/workflows/ci.yml)

A magit-inspired interface for managing multiple Claude Code and Codex instances within Emacs.

Run multiple AI agents in parallel, let them communicate to avoid conflicts, and handle their permission requests one at a time.

[Website](https://hrishikeshs.github.io/magnus/) · [Getting started](docs/getting-started.md) · [Command reference](docs/reference.md) · [Architecture](docs/architecture.md)

Screenshots

**Three agents working side by side in their own `vterm` buffers, with Magnus keeping the team visible:**


**Named agents, active work, attention, and coordination in one buffer:**


Why Magnus?

When working with Claude Code, you often want multiple agents working simultaneously:

• One agent refactoring the auth module
• Another writing tests
• A third updating documentation

But this creates problems:

• File conflicts: Agents might edit the same files
• Context sharing: How do agents know what others are doing?
• Permission chaos: Multiple agents asking for input at once
• Independent review: How do you get one model to review another model's work?

Magnus solves all of this.

Features

Native Claude Code and Codex agents

Each agent runs in its provider's full native TUI inside a `vterm` buffer. Streaming output, approval prompts, slash commands, and the terminal composer keep working as they do outside Emacs. Magnus adds orchestration around those interfaces instead of replacing them with another chat UI.

Other coding agents

Magnus ships with Claude Code and Codex support, but the provider laye