Agent work control plane
Coordinate coding agents without losing the thread.
TeamLoop turns local Codex or Claude runs into a shared loop: define a task, assign it to an authorized runner, collect test/diff evidence, then approve or request another pass.
Focused loopReal local repoEvidence before review
Product
The control plane around every local agent loop.
Shared project state
Members, tasks, runners, attempts, evidence, and reviews live in one D1-backed workspace.
Authorized local agents
Each runner records owner, repo path, capabilities, online status, and the command needed to reconnect.
Delegation without meetings
Route a task to the best eligible runner, including a teammate's agent once they authorize it.
Proof before opinion
Test output, diff summaries, attempt status, and agent notes are attached to the run before review.
Approve or request changes
Human decisions are recorded as the handoff gate before downstream PR work starts.
Collaborative control
Invite teammates to review, comment, approve, or contribute their own authorized runner layer.
Workflow
The browser is the control room. The repo stays local.
1 runner online
Connect from the repo root
Judges can start with `npx --yes teamloop@latest connect`, then run once or keep a daemon polling.
3 evidence items
Work becomes auditable
Every attempt records what ran, what changed, and whether the result is ready for human judgment.
explicit assignment
Send work to the best runner
Assign a task to your own runner or to a teammate's authorized agent when their local context is the better fit.
safe local execution
Keep code changes on the machine
The browser coordinates the loop, while the actual implementation, tests, and diff stay inside the local repo.
review gate
Approve, retry, or hand off
Human review happens after evidence lands, so the next step is a judgment call instead of another status meeting.
team-ready
Collaborate after the loop is proven
Invite teammates once there is a real trail to inspect, or let their runner pick up a follow-up task directly.
Teams
Invite teammates when collaboration helps.
A teammate can review, comment, approve, request changes, or authorize a runner so you can assign suitable tasks directly to their agent layer.