Claims before edits
Four layers keep agents apart: a check before every edit, a watcher on every run, a landing check, and ask-to-release when one agent needs what another holds.
See it in the demoTermy runs your team’s coding agents side by side on one repo, warns before two touch the same file, and merges nothing until your own checks pass.
A session as it plays out in Termy: an edit refused before it collides, checks run on the exact commit, and nothing lands until a person approves.
An interactive demo with example data. No agents run in your browser.
Run several coding agents on one repository and the same four things go wrong. Here is each one, and what Termy does about it.
Both rewrite the same function, and the last save wins.
Each agent claims what it is changing. An edit to code someone else holds is refused.
The agent says the tests pass. Nobody ran them.
Your own checks run in a sandbox, on the exact commit.
Usage piles up across accounts, with no task to pin it on.
Each run has a cost against its task, and a budget stops new work at the cap.
A cleanup resets the branch, and a day’s work is gone.
Termy writes a backup before it moves a branch, and never force-pushes.
Say what you want built, or start from a GitHub or Linear issue. Termy turns it into tasks with acceptance criteria.
Planning runs on the agent you’re signed in to. You approve the plan before any code is written.
Claude Code, Codex and Termy’s own agent work at once, each in its own copy of the repo, each holding a claim on what it is changing.
Your project’s own checks run in a sandbox on the exact commit, and browser QA screenshots the pages it changed.
A second model can review the diff. A person who didn’t run the agent approves, then Termy lands it or opens a pull request that closes the issue.
Deploy to Docker with health checks and automatic rollback. Cloud targets are planned.
No queue and no turns. Each teammate’s agent works in its own copy of the repo, under their own account, and overlap shows up before the work happens, not at merge.
Four layers keep agents apart: a check before every edit, a watcher on every run, a landing check, and ask-to-release when one agent needs what another holds.
See it in the demoShare a file and edit it together with live cursors, watch each other’s agent runs, talk by voice, camera or screen share, and answer your teammates’ agents’ approval requests from your inbox.
See it in the demoWhen a task fails its checks, autopilot hands it to another agent with the failure attached, within your budget. Anything it can’t decide waits for you.
How checks workEach one shows its owner, branch, agent, files and what it is waiting on. Join as a participant or watch read-only.
See it in the demoInvites with roles (owner, admin, developer, viewer), an audit log of every claim and approval, and a switch that stops every agent at once. Single sign-on is planned.
Read about securityThe CLIs you already have, on your own accounts. Termy never resells tokens.
An illustration with example output. Agents run in Termy’s own terminals, under your own accounts.
From a preview of every task to a deploy you can roll back, with your own Mac doing the work.
Agents run on your Mac under your own accounts. Your team’s hub sees coordination data, such as who holds which file, and a file’s contents only while you share it live.
Termy’s server answers on 127.0.0.1 and nothing else, and every request needs the session it issued at launch.
Checks, previews and merges run your repository’s code in Docker or the macOS sandbox. With neither available, Termy refuses rather than run it bare.
No resets, rebases or force-pushes. Termy writes a backup before it moves a branch, and only ever fast-forwards.
Edits stay inside the project, .git is off limits, and risky shell commands wait for your say-so.
Provider keys live in the system’s credential store, not in config files. Termy never resells or meters tokens.
When Termy can’t establish that something is safe, it stops and tells you why instead of carrying on.
Didn’t see yours? Ask us anything when you join the beta.
A macOS app where your team’s coding agents work on one repository at the same time. It keeps them from colliding, runs your own checks on every change, and lands nothing until they pass.
Sign in with Claude Code, Codex, Gemini CLI, OpenCode or Cursor Agent, or use keys for Anthropic, OpenAI, Groq, OpenRouter, Gemini or a local model. Copilot CLI, Aider and Goose run too.
Agents and checks run on your Mac. Your team’s hub sees coordination data, such as who holds which file and how runs are going, and a file’s contents only while you share it live. Your model provider sees what your agent sends it, as it would without Termy.
Model usage is billed by your providers, on your own accounts. Termy doesn’t resell or mark up tokens. Pricing for Termy itself after the beta hasn’t been announced.
macOS 13 or later today. Projects in JavaScript and TypeScript, Python, Go and Rust. Windows and Linux builds are planned.
No. Checks are your project’s own, agents can’t edit the tests that judge them, and the person who ran an agent can’t approve that agent’s change.
Deploying to cloud providers, Windows and Linux builds, single sign-on, and Jira. They are marked planned wherever they appear, and nothing on this page shows them working.