Runtimes · usage · cost
Every runtime on the device, one ledger
The native harness, Claude Code, Codex, and on-device models — each runtime's sessions, tokens, and cost roll up into cross-session usage you can actually read.
Usage rollups
Token counts and cost aggregate across sessions and runtimes, deduplicated and bucketed — what ran, on which model, at what spend.
A network of daemons (beta)
Lend idle resources to teammates and borrow theirs — with explicit permission, a session can run on a colleague's idle machine or a headless remote device, and still lands in the same record.
Resource-aware
Execution reacts to CPU, RAM, and battery. Heavy runs pause when you unplug; queued work resumes when capacity returns.
Overnight autonomy
Scheduling and auto-started sessions let agents continue while you're away — recorded like everything else, reviewable in the morning.
On-prem friendly
Local models via OpenBMB run fully on-device — the same architecture that makes private, on-prem enterprise deployment natural rather than a port.
GUI E2E self-testing
Agent-powered end-to-end testing of the app's own UI supports supervised self-evolution — ORG-2 is built and tested with ORG-2.