ArcavenAE follows the Unix philosophy: small, composable tools for operating teams of AI agents. Each does one job, works on its own, and connects through inspectable seams: files, command lines, tmux, and signed artifacts. Use what you want.
🚧 Pardon the construction — we're just getting started. Some projects are useful today; others are designs or early builds. The tour below says which is which.
| Project | Function | State |
|---|---|---|
| marvel | Fleet orchestration for agent sessions: declarative teams, heartbeats, restart policies, shifts, and tmux runtime adapters. It can launch any console rather than requiring an Arcaven-specific agent. | usable, single-host |
| kos | A lossless knowledge record for decisions, evidence, and ruled-out approaches, organized by confidence instead of flattened into one undifferentiated memory. | usable |
| flyloft | Task-scoped retrieval across levels of detail. Automatic summaries remain linked to preserved source material, so a compact view never replaces the record. | skeleton |
| switchboard | Remote access to any tmux session across hosts and NAT. SSH remains end to end; the relay routes sessions it cannot read. | protocol proven; rebuild in progress |
| sideshow | Content-pack distribution for skills, commands, rules, and hooks. Packs can be built as frozen, signed compositions with traceable provenance. | usable, narrow; release verification in progress |
| critic | A planned registry and arena for comparing agent-generated variants, preserving lineage and accounting for the cost of the selected outcome. | designed, pre-code |
| beadle | Issue and PR triage against a repository's declared intent, informed by what maintainers actually act on. | running today |
| usher | Onboarding and installation help for people and machines, with opt-in configuration and usage reporting. | design phase, private |