Diri – Managing Multiple Coding Agents

What it is

• A native macOS (and beta‑Linux) application, written in Rust with the GPUI framework, that lets you run many AI coding agents (Claude Code, Codex, Cursor, Gemini, etc.) side‑by‑side.
• Each agent gets its own Git worktree/branch so their changes never clash.
• An internal MCP (multi‑agent coordination) server lets an agent spawn and talk to other agents, enabling simple “swarm” workflows.

Key capabilities

Feature	What you get
Parallel agent UI	Live status pane shows which agents need your input, which are working, and which have finished – all in one window instead of dozens of terminals.
Worktree isolation	Every agent runs in its own Git worktree, so you can review, stage, commit, and run CI checks for each agent’s changes independently.
Agent‑to‑agent spawning	An agent can launch other agents, hand them subtasks, read their output, and answer their questions via the built‑in MCP server.
Session persistence	Each terminal is owned by its own process; closing the UI or restarting the engine does not kill running agents.
Local‑first	Works with the CLIs you already have installed; no need for tmux, sudo, a Diri‑hosted relay, or a Diri account.
SSH support	Agents can run on any SSH‑accessible machine you control, letting you leverage remote compute.
Extensible agent list	Comes pre‑configured for 22 agents; you can add more via simple agent manifests.

Installation

• macOS (15+): brew install --cask cristicretu/diri/diri – signed and notarized DMG available.
• Linux (beta): x86_64 Ubuntu 22.04/24.04, X11 or Wayland, Vulkan 1.3. Packages are optional; see diri/LINUX.md for building from source.

Getting started

1. Install the app.
2. Open Diri and add the agents you want (the UI can auto‑detect CLIs on your system).
3. Launch agents; each appears in its own pane with a status badge.
4. Use the built‑in Git worktree view to inspect diffs, stage, commit, and run PR checks.
5. If an agent needs a decision, a notification pops up in the sidebar.

Typical workflow

• Write a PRD in a Diri note.
• Break it into to‑dos and assign each to an agent.
• Agents work in parallel, writing code into isolated worktrees.
• You review the changes in context, merge what you like, and let the remaining agents continue.

Extending the tool

• Add new agents by creating an agent manifest (see docs/AGENT-MANIFESTS.md).
• Contribute small, well‑tested changes; the project welcomes bug fixes, documentation improvements, and manifest additions.

Links

• Download latest release
• Quick‑start guide
• Full documentation
• Roadmap

License

• AGPL‑3.0 (see LICENSE).

---

Diri aims to remove the human bottleneck when working with dozens of AI coding agents, scaling the workflow from a handful to potentially hundreds of concurrent agents.