Escoffier Labs
Mise en place
for your agents.
Small, composable agent-ops tools that help AI agent runtimes work with real local environments. Memory that survives the session. Secrets that arrive sealed. Logs that become evidence. Every station, kept in order.
-
Runs the kitchen. AI agent memory, handoffs, and local guardrails for Codex, Claude Code, OpenCode, Hermes, and OpenClaw. Agents write handoff notes, you approve what sticks, every future session starts smarter.
-
Keeps the larder. Mirrors the browser sessions and secrets you choose from your daily driver to the machines your agents run on, sealed with AES-256-GCM the whole way.
-
Nothing leaves the kitchen unchecked. Policy-driven content scanning and publish gates: hostnames, private IPs, keys, and credentials caught at the boundary, not in the postmortem.
-
Every station keeps its log. Exports local agent-session history from Codex, Claude Code, OpenClaw, OpenCode, and Hermes to portable JSONL. Local-only, no network calls.
-
What comes in from beyond the kitchen. Exports notes, chat archives, crawler output, issues, and git history into the same portable adapter format StationTrail speaks.
-
The bound ledger. A local-first evidence graph over all of it: SQLite full-text search, Markdown export, and Brigade-ready evidence bundles from StationTrail and SourceHarvest imports.
-
Works every station. Maintenance CLI for the Claude Code and OpenClaw memory system in four verbs: status, lint, ingest, compact.
-
Keeps the mise en place tight. Audits oversize OpenClaw bootstrap files and trims them into reference cards with heuristics plus LLM judgment.
-
Calls the orders. Privacy-first push notifications from your coding agents to Discord, Telegram, and Signal. Zero telemetry, a single Go binary.
Solomon's Guide to Cookin' with Gas
The companion cookbook: how this kitchen actually runs in production, what broke, and how it was fixed.