v1.803.0 v0.9.5
cheaper-inference/OmniRouteSep 3, 2026 → Sep 3, 20260 days between releases
What's New
Adds task-type-aware routing, SQLite caching, and quota forecasting alongside MCP toggling and security utilities
New Features
- Task-type-aware routing with configurable weight share for code vs chat requests
- SQLite persistence for the response cache to survive restarts
- New `/v1/quota-forecast` endpoint for remaining daily headroom
- On/off switch for the `/mcp` endpoint
- Encryption key rotation utility for stored secrets
Improvements
- Response cache now persists to SQLite for restart resilience
- Analytics dashboard shows cache hit rates and saved tokens
- Models endpoint now reports execution status using router checks
- Security improvements via encryption key rotation
- MCP endpoint behavior changes (403 on disabled verbs)
Upgrade Notes
Configure task routing weights using `GET/PUT /api/settings/task-weight-share` and toggle the MCP endpoint via `GET/PUT /api/settings/enable-mcp`. Opt-out of SQLite cache persistence by setting `RESPONSE_CACHE_PERSIST=0`. Use `npm run rotate-encryption-key` to re-encrypt stored secrets.
The Two Releases
v1.803.0
Sep 3, 2026· 2 months ago
Focuses on expanding MCP capabilities by exposing request headers to scripts and streamlining project imports by reusing existing workspace resources.
Full release notesv0.9.5
Sep 3, 2026· 2 months ago
Introduces task-type-aware routing, persistent SQLite caching, a quota forecast API, and an MCP toggle, alongside security and stability improvements.
Full release notes