Research Domain
Agentic Intelligence
D-CSIL LabRats — production terminal agents with a deterministic policy engine, reversible git-backed checkpoints, and durable memory. Remy leads on a cloud model; Justin assists from a second instance on a local one. Together they replaced the earlier Hermes and OpenClaw agents.
Research Question
Can an autonomous engineering agent be safe, auditable, and reversible by construction rather than by model goodwill?
The LabRats design pushes safety out of the model and into the system: allow/deny policy rules the model cannot rewrite, git-backed snapshots for every file edit, an inspectable tool timeline, and SQLite session persistence.
They route across OpenAI, Anthropic, and local model endpoints, and are reachable over Telegram, email, and Messenger. The 'Soul' layer gives each rat its identity, mood, and whisker nudges.