ai gateway
Bring your own model. Route through one API.
One OpenAI-compatible endpoint that tries your own hardware first — and tells you exactly how often it succeeded.
Optional. Teemesh's product layer needs none of it — if you already run Claude Code, Factory or Codex, keep them and start there.
POST /v1/complete
{ "messages": [{ "role": "user", "content": "…" }] }
→ 200
{ "teemesh": { "tier_used": "genome_local", "was_sovereign": true } }
bring your own model
Point Teemesh at hardware you already run — Ollama, a home GPU, a mesh node. One OpenAI-compatible endpoint in front of all of it.
sovereign-first cascade
Every request tries genome_local, then your tenant mesh, then cloud BYOK last — a dead tier fails over in ~2s, never silently forever.
published, not promised
GET /v1/usage/summary returns the real offload ratio for your account, flagged honestly when the sample is still thin — never a marketing number.