Remuda – agent orchestration tool
Remuda is a command line application that makes many agentic programming workflows easier:
- create disposable working copies of Git repositories
- launch agents in them
- containerize your agents for safer unattended workflows
- reuse prompt snippets and inject context from tools like Jira and Slack
- broadcast messages to swarms of agents
Why Remuda?
I would frequently hear coworkers voice some problem or feature idea and think "(current frontier model) could oneshot that", but then I knew I'd have to clone the repo, assemble the prompt, configure all the flags, manage the terminal window, etc. There was enough activation energy still there that I just wouldn't go through with it most of the time.
Most agentic coding tools take care of managing single agents with the human in the loop, but not a lot of them expose a good interface for managing lots of agents working on unrelated tasks. The ones that do are often ecosystem locked to only one agent harness, despite there being at this point few meaningful differences between how they fundamentally work.
I wanted one that could do all that and fit with my own tooling preferences:
- session management with a terminal multiplexer (e.g. tmux)
- little to no config required
- strong integration with my other CLI tools
- portable binary
I believe coding agents are strongest when managed asynchronously. For the busy developer, they are a powerful weapon against codebase rot, allowing you to delegate smaller tasks while you focus on big ones. Remuda attempts to make this workflow as frictionless as possible.
Remuda was the answer. I started with the goal of being able to kick something
off with just vibe "please do this thing I don't have time for" and it grew
from there.
Installation
Requirements:
- Required to build/install (via
go install):- Go 1.24+ (any platform supported by the Go toolchain)
- Git
- Strongly recommended:
tmuxfor the default detached-session flow (zellijis a supported alternative when configured)- Agent harness (at least one is required to actually run an agent). Currently supported:
- Codex
- OpenCode
- Claude Code
- You can also provide a custom agent command
- Optional (feature-dependent):
- fzf for interactive selection flows
Build from source
go install github.com/yendo-eng/remuda/cmd/remuda@latest
# Verify installation
remuda --helpQuick Start
By default, remuda stores workspaces under ~/.remuda/repos. You can override
the root via the REMUDA_REPOS_BASE_DIR environment variable.
Tip — Interactive wizard: Add --wizard to most commands for a guided, TUI-based flow.
Here's the most generally useful workflow for feature development:
# Use a published image for container sessions IMAGE=ghcr.io/acme/vibe-dev:latest remuda vibe \ --name '<your-session-name-goes-here>' \ --container \ --container-name "$IMAGE" \ --yolo \ --use make-pr \ --use small-commits \ "please build the feature"
Documentation
For detailed usage information, see:
- Commands – vibe, clone, vibe-check
- Session Management – managing tmux sessions
- Container Mode – running agents in Docker
- Configuration – environment variables and config file
Session Merge Flags
remuda session kill --merge uses gh pr merge under the hood.
- Configure default merge flags in config:
defaults.merge.gh_flags - Override per run with repeatable
--merge-flag(replaces config list) - If neither is set, Remuda keeps the current default behavior:
--rebase
License
Remuda is licensed under the MIT License.