Internal orchestration layer for TechSafe Industries. Not a SaaS product. Not publicly exposed.
- On-demand only — no persistent services, no always-on gateway
- SQLite ledger — single-file datastore for all sessions, decisions, and audit trails; no Redis, no BigQuery
- Deterministic routing — agent dispatch is rule-based and auditable, no probabilistic black-box routing
- Governance-first gating — every agent action is logged and gated before execution
- No SaaS / no public exposure — internal use by TechSafe operators only; no VPS scaling, no public endpoints
- Structured outputs only — all agent responses return typed, schema-validated JSON; no freeform prose passed downstream
- Kimi is LLM engine only — OpenClaw controls agent lifecycle, logging, and routing; Kimi provides inference only
| Folder | Purpose |
|---|---|
app/ |
Core application logic — orchestrator, session manager, agent dispatcher |
db/ |
SQLite database files (gitignored) and migration runner |
schema/ |
SQL migration files applied in order (001_init.sql, 002_…, etc.) |
agents/ |
Agent registry JSON and individual agent definition files |
docs/ |
Architecture documents, decision records, and governance specs |
tests/ |
Unit, integration, and E2E test suites |
src/ |
TypeScript gateway source (existing Node layer) |
config/ |
Environment-specific configuration (non-secret values only) |
docker/ |
Dockerfile variants and compose configs for local dev only |
v1 scaffolding in progress. No agent logic implemented yet.