For developers
Governed inference in one environment variable.
Cache, audit, tier-aware routing, and HITL on dangerous tool calls without rewriting your agent. Anthropic-compatible, OpenAI-compatible, OpenClaw-compatible.
What this covers
Change one variable and you get caching, audit receipts, tiered routing, and human approval on dangerous calls, with no rewrite. The Quickstart walks through it in five minutes. Inference is one use case: the same gate() call from @synoi/sdk governs any action, an API call, a payment, a device, an integration, not only LLM traffic. See the Gateway product page for the general path.
- One env var · ANTHROPIC_BASE_URL or OPENAI_API_BASE
- BYO keys · provider keys forwarded once, never persisted
- Receipts · hybrid-signed (Ed25519 + ML-DSA-65), public verifier, no SynOI dependency to verify
- HITL · Slack, SMS, desktop, mobile (coming): pick per-tenant
- Cache · significant hit-rate reduction on repeated and near-identical queries; P99 latency overhead under 2 ms (measured internally)
- OpenClaw plugins · native, your LLM keys never leave the machine