Run a company of AI agents on a server you own. Spin up named agents (claude, codex, pi…), put them on an org chart with a shared backlog, let them hand off work and ping your phone only when a human must decide. MIT.
Executable operating contracts for AI agent loops: 7 typed terminal states, verification-gated completion, bounded repair, and repo-native state — so a loop proves "done" instead of pretending. Claude Code plugin + portable Python core.
Loop engineering with verifiable provenance: every iteration of a branching agent loop becomes a recorded step in a content-addressed opentine artifact you can diff, fork, resume, and attest.
