When one AI agent or account runs out, another picks up where it stopped. When you hand the work to a teammate, nothing stays behind in a chat. The project travels in its taskbag: plain files in your own repository.
Vibe coding runs on many sessions: Claude Code, Codex, several accounts, several people. A quota can end a session without warning, and the next one starts cold. In one real multi-agent project:
Shift the project between AI agents and people without losing anything. The same alm commands for every agent; handoff and resume in one step.
Learn what each agent is good at, what to hand over at a switch, and a simple process: one task per outcome, proven steps, your approval before anything leaves your machine.
Progress by area, what waits on you, and a burndown with a forecast finish date from your real pace. (planned)
Tasks, proven steps and handoff notes live as readable files in .allomos/taskbag/, versioned by your own Git — not inside one agent's memory or one account.
# Claude Code · account 1 · quota nearly used
alm label A01-03 "implemented"
alm release A01-03 --note "form done; tests next"
# Codex · account 2 · a minute later
alm next
A01-03 Sign-in page ready · note: tests next
alm claim A01-03
An example of the forecast to come, not real data.
No account, no card, no trial clock. Your tasks stay on your machine and in your repository.
Live sharing, email invitations and roles for teams. Pricing to be announced.
v0.1, the macOS local tracker, is out. v0.2 (macOS and Windows, editing in the app) is in progress. Follow along or try it on GitHub.