Stop choosing models.
Start completing work.
ThreadGrid is the control plane for AI work. Describe the outcome and it threads the job across Claude, Codex and OpenCode on your own computers, based on what each does best and how much capacity it has left.
Free · bring your own subscriptions · macOS and Windows
Thread
Add organization-level RBAC
Migrate existing users, add tests and update the admin UI.
Plan · Grid Fusion
stage 3 of 5- Plan
Claude - Build
Codex - TestMacBook Pro
- ReviewClaude
- FixCodex
How it works
One request. Threaded across your whole Grid.
You describe the outcome. The Grid Router picks the best agent for the job, or splits it into plan, build, test and review stages across several, and brings the results back together as one finished Thread.
Claudeplan · review
OpenCodesecond opinionCodeximplement · fixClaudeplan · reviewOpenCodesecond opinionCodeximplement · fixCapacity-aware routing
Your best model is a finite resource. Spend it deliberately.
ThreadGrid tracks how much quota each of your subscriptions has left and when it resets. Routine work goes to whoever has room, so your scarcest model is still there for the hard problems.
- Honest numbers. Every reading says where it came from: exact, observed, estimated or unknown. We never invent a percentage.
- Capacity Protection. Reserve a provider for planning and review, preserve it below a threshold, or queue work until it resets.
- Explainable routes. Every automatic choice shows the factors behind it. You can always override.
Grid capacity
updated 12s agoClaude
5-hour window · resets ~2h 14m
observedLOWCodex
weekly window · resets Friday
exactHIGHOpenCode
no plan limits exposed
unknownUNKNOWN
Route explanation
implementation → Codex| factor | Codex | Claude |
|---|---|---|
| capability fit | 0.00 | 0.00 |
| capacity health | 0.00 | 0.00 |
| track record | 0.00 | 0.00 |
| protection | 0.00 | 0.00 |
| score | 0.00 | 0.00 |
Claude is the better fit on paper, but it's low on capacity and reserved for review. Codex takes the implementation.
One Grid
More than a chat window with a model picker.
A multi-agent frontend asks which agent you want. ThreadGrid asks what you want done, then plans, routes, verifies and recovers.
planbuildreviewfixGrid Fusion
One request becomes a plan. An architect plans, a builder implements in an isolated worktree, local tests verify, and an independent reviewer checks before fixes land.
Smart EscalationComing soon
Start with a cheaper model, check the result, and step up to a stronger one only when tests or review fail.
Codex+212 −40 · tests passKeep thisClaude+168 −22 · tests passyou pick · nothing merges until you do
Compare
For hard problems, give the same task to two or three agents, each in its own worktree. Read their work side by side and keep the one you want.
Computer to computer
Pair your Mac and a Windows PC on one account, then send work to either from the other. It runs where the repo and tools already are.
Local-first by design
Repositories, credentials and agent processes stay on your machines. The cloud only coordinates.
Bring your own AI
Use the subscriptions you already pay for.
ThreadGrid doesn't resell tokens. It drives each provider's official tools on your machine, signed in as you, and coordinates them as one Grid.
Claudeclaude auth loginCodex CLIcodex loginOpenCodeopencode auth
Local modelsollama serveplanned
Geminigemini authplanned
Cursor Agentcursor-agent loginplanned✓ Claude · signed in · capacity LOW (observed)✓ Codex CLI · signed in · capacity HIGH (exact)✓ OpenCode · signed in · capacity unknown✓ Computer windows-pc · pairedClaude reserved for planning and review until it resets.Codex wants to push and open a PR
git push origin feature/rbacReview finished · 2 comments4m agoFixes applied · re-running tests6m agoiOS companion
Your Grid, in your pocket.
Walk away from your desk. Follow jobs, read the changes, send a follow-up or approve a protected action from your iPhone. The iPhone app is in testing and not yet available to download.
- Approvals bound to the exact action. Each approval covers one specific push, deploy or command, and can't be replayed.
- Your phone never holds your keys. It sends scoped commands to your Mac, over your LAN or an end-to-end encrypted relay.
Local-first
Powerful automation, with you in control.
ThreadGrid can run real developer tools, so it's built to fail closed.
Secrets stay on your machine
Provider credentials never leave your machine or touch our servers.
Agents run locally
Provider CLIs run on your computers, each thread in its own git worktree.
A thin cloud
The cloud handles sign-in, device pairing and notifications. Your prompts and code aren't uploaded by default.
Encrypted relay
Remote traffic between your devices is end-to-end encrypted.
Protected actions
Pushes, deploys and destructive commands stop and wait for your approval.
Every decision explained
Routes, retries and approvals are recorded so you can see why something happened.
All your AI. One Grid.
ThreadGrid is free. Download the beta for Mac or Windows, connect the agents you already use, and let the Grid decide who does what.