BRAID
Your parallel agents,
now on autopilot.
Run Claude, Codex, and your shell side by side in isolated git worktrees — or hand a task to a relay and let it implement, review, open the PR, and fix CI until it's green. You decide how much it does on its own.
Mac · Local-first · Your code never leaves your machine
From devs running agents in parallel
Early users on what changed once Braid took over the juggling.
Hands down the best way to run multiple agents across multiple projects.
Alex K.
Senior engineer
Went from one agent at a time to four. The worktrees mean nothing collides — I stopped fighting git and shipped.
Maya R.
Indie developer
Coming back to a task two days later used to cost me an hour. Now I hit restore and pick up mid-thought.
Jordan T.
Staff engineer
Sound familiar?
Running multiple AI agents at once should make you faster. Most days, it just makes a mess.
Five Claude windows. Three branches. No idea what's running.
You started a refactor in one terminal, kicked off tests in another, asked Codex to fix a bug in a third. Now you're the human Kubernetes scheduler, and you're losing.
Agents finish while you're not looking. Or get stuck and just sit there.
No notifications. No status. You tab back twenty minutes later to find one done, one waiting on input, and one halfway through clobbering your branch.
Your worktrees, terminals, and PRs don't talk to each other.
Branch in one window. PR status in a browser tab. Notes in Notion. CI in Slack. Switching context costs you ten minutes every time, and the agents don't wait.
How Braid works
From juggling terminals by hand to handing the whole loop to a relay — same workspace, your call on how much it runs alone.
Spin up an isolated worktree
Name a task, pick a base branch, and Braid provisions a fresh git worktree on disk. No stashing, no branch confusion — just a clean room for the work.
# auto-created on task creation
/projects/myapp/.worktrees/auth-flow
✓ branch: feat/auth-flow
✓ worktree: ready
Send agents to work in parallel
Open a Claude terminal, a Codex terminal, or a plain shell — all inside the task's worktree. Each agent has its own room. None of them collide.
Or let a relay take it to green
Hand the task to a relay and check back when it's green. Claude implements, Codex reviews, Claude folds in the notes and opens the PR — and when CI goes red, Braid feeds the logs back to Claude to fix.
Hand the work to a relay
A relay chains your agents into a loop and runs it end-to-end for you. You set the guardrails and decide how much it does on its own.
Walk away. Come back to a green PR.
The Ship relay runs the whole loop: Claude implements your task, Codex reviews the diff, Claude folds in the findings, and the PR opens itself — then Braid watches CI and, when it goes red, feeds the failing logs straight back to Claude to fix. It keeps going until the checks are green.
Bounded by guardrails you set — a maximum number of iterations and a wall-clock limit — so it converges or stops, but never spins forever.
Compose Claude and Codex into a loop
Relays are just graphs: agent steps, shell steps, repeats, and conditionals. Chain Claude to build, Codex to review, a shell to run your tests — then branch on the result. Save it as a preset and reuse it on any task in the project.
Per-step model and effort overrides. Prompt templates that get your task's title and description injected automatically — so every step knows what it's working on.
You decide how much it decides
Set the dial to manual, tiered, or full-auto. On tiered, Braid answers the low-risk calls itself, following the agent's own recommendation. It stops to ask you only when a choice is genuinely risky — and logs every decision it makes in the run.
A safety denylist always pauses for the dangerous stuff — drops, deletes, deploys, production, secrets — no matter where the dial is set.
The fundamentals, done right
The same foundation holds whether you're driving or a relay is: real terminals, isolated worktrees, and GitHub that never blocks you.
Claude and Codex are first-class
Each task can host multiple terminals — and Braid knows the difference between a Claude session, a Codex session, and a plain shell. Named, recognized, and kept in their own panes. When an agent finishes (or stalls), you'll know.
Real PTYs under the hood, not fake terminal emulators. Anything you can run in your shell, you can run here — including the agent a relay just used, picked back up where it left off.
# Three tasks, three isolated checkouts
~/myapp/.worktrees/
├─ auth-flow → feat/auth-flow
├─ fix-ci → fix/ci-pipeline
└─ docs-update → chore/docs
# Each agent stays in its lane.
# No stashing. No clobbering.
# No "wait, which branch was I on?"
Every task gets its own checkout
Braid provisions a real git worktree per task — automatically. Three agents working at once means three branches, three directories, zero collisions.
Status badges tell you when a worktree is provisioning, ready, broken, or missing. Recovery is one click.
GitHub that never blocks you
PR refresh, CI checks, review status — every GitHub call runs as a background job, persisted to disk. The UI never freezes. The spinner is only ever in one panel, never the whole app.
Uses your local gh CLI auth. No tokens to paste. No OAuth app to install.
feat: implement OAuth flow
opened 2 hours ago · main ← feat/auth-flow
Branch
feat/auth-flow
↑ 3 ahead
Last activity
3 commits
2 days ago
Previous session
Pick up where you left off — even days later
Come back to a task and Braid shows you exactly where it stood: branch ahead/behind, last commits, prior terminals, the activity timeline of every command and PR event.
One click restores your terminals to the same shells, same working directories. No more "what was I doing?"
Your code stays on your machine
No SaaS lock-in. No cloud sync. No telemetry pings. Braid is a real desktop app — your tasks, terminals, notes, and history live in a SQLite file on your laptop. Close the lid, board a flight, keep working.
SQLite, on disk
Every project, task, command, and snapshot lives in one local database file. Back it up. Inspect it. Own it.
Works offline
No internet, no problem. Your terminals, worktrees, and notes don't depend on cloud latency. Even GitHub data is cached locally.
No telemetry
We don't watch you work. The only network call is auth and license check. Your code, prompts, and shell output never leave your machine.
What's coming next
A Pro license unlocks every update during your year — including everything below as it ships.
Linear integration
Pull issues straight into Braid as tasks. Status syncs both ways. Your Linear board and your worktrees stay in lockstep.
Phone & remote push
Desktop alerts already fire when a relay finishes or needs you. Next up: push to your phone — so you can kick off a relay and leave your desk entirely.
Deeper GitHub sync
Read PR review threads, inline comments, and check details inside Braid. Reply without leaving the task.
Slack integration
Pipe agent finishes, CI events, and PR updates into the Slack channel of your choice. Or pull team threads into the right task.
Built-in task board
Plan your work without leaving Braid. Backlog, in-flight, blocked, done — a lightweight board for solo devs who don't need Jira.
Jira integration
For teams who can't escape it. Pull tickets, transition status, log work — all without leaving Braid.
Notion integration
Sync task notes to a Notion database. Keep your team's source of truth without copy-paste.
Windows & Linux
Native builds for Windows and Linux are on the way. A Pro license covers every platform at no extra cost.
Simple pricing
Free to try on a real project. Upgrade to Pro when you're ready to run multiple.
Free
For trying Braid out
- 1 project, unlimited tasks
- Claude, Codex & shell terminals
- Relays, presets & the autonomy dial
- Auto worktrees + GitHub PR & CI
- Multiple projects
Pro
For shipping at full speed
- Unlimited projects
- Everything in Free — relays on every project
- All updates for the year
- Linear, Jira & Notion as they ship
- Priority support from the Braid team
Renew yearly. Cancel anytime.
What happens when my license expires?
You keep the latest version you had — forever. No lock, no downgrade, no subscription trap. (Full details in the FAQ below.)
Questions, answered
Things people ask before they buy.
Can Braid really run on its own — and is that safe?
What happens when my license expires?
Is my code uploaded anywhere?
gh CLI — with Braid never in the middle.
Why $50 — what's the catch?
Mac only? When are Windows and Linux coming?
Which AI agents does Braid actually support?
Do you offer refunds?
Stop juggling.
Start braiding.
Get the desktop app. Run your agents in parallel — or hand a task to a relay and let it ship on its own.
Free with 1 project · macOS · Windows & Linux on the roadmap