Enterprise AI Delivery Engineering
AI systems in the customer's own environment: status you can verify, data you can restore, changes on record.
# 网关地址与 Key 由你自己的部署提供
curl https://<your-gateway>/v1/chat/completions \
-H "Authorization: Bearer $GATEWAY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"<model-name>","messages":[{"role":"user","content":"你好"}]}'Docs by component
- Lurus WitnessEarly pilot
Read-only collection, external probing, a real restore drill every month and a monthly evidence report; every reading states how strong its evidence is. Docs are in Chinese.
- KovaEarly pilot
Embedded durable execution engine: write-ahead-log crash recovery, execution records, replay.
- MemXEarly pilot
AI memory engine: extraction, deduplication, decay, hybrid retrieval; CLI / REST / MCP.
- Gateway & APIIn internal production use
Privately deployed multi-tenant LLM gateway, based on the open-source New API (AGPLv3). Interface details are in the API reference.
- Platform baseIn internal production use
Accounts and sign-in shared by every component. It is a base layer, not offered on its own.
Where to start
- Operators deploying Witness
- Read Deployment (Chinese), then Core concepts (Chinese), to see where each reading comes from and how certain it is.
- Developers integrating the gateway
- Make one call with the Quickstart; look up fields and error codes in the API reference.
- Decision-makers evaluating
- Read "What it is not" and "Where it stands" in the Witness overview (Chinese), then Restore drills and monthly reports (Chinese).
Boundaries
We do not resell data-center space or compute, and we do not promise numbers we cannot prove. Maturity uses four labels only (early pilot, in internal production use, open source, in design), and when unsure we pick the more conservative one.