Streaming agent chat
Reasoning steps, tool calls, and edits render live, with real unified diffs inline. Transcripts are cached in SQLite, so they survive restarts and context compaction.
Early access · Free while in development
Run coding agents in parallel.
Without the terminal juggling.
BranchLab is a macOS app that runs an opencode agent per task, each in its own isolated git worktree, and gives you one window to chat, review diffs, and ship pull requests. It even keeps fixing CI until your checks go green.
Universal .dmg · auto-updates built in · needs git and the opencode CLI
git worktree per task
Every session gets its own branch and checkout. Agents can't step on each other, or on your working copy. No stashing, no branch juggling.
One window for N agents
A live sidebar shows who's working, who's waiting for you, diff stats, and PR/CI state for every session, even the ones off-screen.
Ship without leaving
Review diffs inline, reply with PR-style comments, then commit, push, and open the pull request, all from the same window as the chat.
The autofix loop
Green checks, while you review something else
BranchLab's Rust backend watches every open PR's checks for all sessions, not just the one on screen. When CI fails, it dispatches the agent with the failing logs and a mandate: reproduce, make the minimal fix, commit.
Off Autofix ✦ Superfix
- Off: just monitor the pipeline and surface failures.
- Autofix: the agent diagnoses and fixes locally, then commits. You review and push.
- Superfix: fix, commit, and push automatically. It loops until checks pass.
Everything in one place
Built for how agentic coding actually feels
Review like a PR
Unified or split diffs scoped to the last turn or all changes. Comment on specific lines, batch your notes, and send them to the agent as one structured message.
A review inbox on Home
PRs waiting on you, across every connected GitHub account, including Enterprise. One click checks a PR out into a fresh worktree session.
Your models, your dials
Pick the model per session, switch build/plan modes, tune thinking effort from low to max. Curate which providers show up in the picker.
MCP servers & permissions
Connect and disconnect MCP servers per session, watch language-server status, and answer the agent's permission requests inline in the transcript.
Feels native, stays yours
A real macOS app: Tauri + Rust, keyboard-driven, 15 themes, auto-updates. Your code never leaves your machine: agents run through your own opencode install and your own keys.
The review loop
Comment on the diff. The agent does the rest.
Review the agent's work the way you review a pull request: click a line, leave a note, batch your feedback. One click sends every comment (with its file, line number, and exact code) straight back into the conversation.
Comment on any line
PR-style inline comments, right in the changes panel.
Send it to the agent
Feedback lands as one structured message, and the fix comes back.
Start your day here
Describe a task. Get a worktree, a branch, and an agent.
The home screen is a launcher and a review queue in one: type a prompt to spin up an isolated session, or pick up a pull request that's waiting on you.
How it works
Three steps from idea to merged
- 1
Add a repo
Point BranchLab at any local git repository. It probes for
gitandopencode, binds a GitHub account if you have one, and that's the setup. - 2
Describe the task
Each prompt spins up a worktree on a fresh branch with its own agent. Start five tasks; they run side by side while the sidebar tracks every one.
- 3
Review & ship
Read the diff, push back with inline comments, then commit, push, and open the PR. Flip on Superfix and the checks take care of themselves.
FAQ
Questions developers ask first
What do I need to run it?
macOS, git, and the opencode CLI on your PATH (curl -fsSL https://opencode.ai/install | bash). GitHub features use the gh CLI. BranchLab checks for these on launch and walks you through anything missing.
Which models can I use?
Anything your opencode install supports: Anthropic, OpenAI, and the rest. Pick a model per session, set a project default, and tune mode and thinking effort from the composer.
Does my code leave my machine?
BranchLab has no backend service. Worktrees live on disk, transcripts live in a local SQLite database, and agents talk to model providers through your own opencode configuration and your own API keys or subscriptions. The app reports anonymous usage stats (screen views and feature counts, never code, paths, or prompts), and you can switch that off in Settings.
What does it cost?
BranchLab is free while it's in early development. You pay only for the model usage you already pay for through opencode.
Windows or Linux?
macOS-only today: the editor, terminal, and Finder integrations are Mac-native. A portability pass for Windows and Linux is planned.
Stop babysitting one agent.
Start reviewing five.
Free during early access · auto-updates keep you current