FreeLLMAPI no longer sends any key-check request to SerpSearch. This release also adds the onomeo provider, per-key spend in /v1/usage, GET /v1/models/:id, and per-key monthly caps across rerank, moderation and the router's key checks.
Providers
- SerpSearch keys are never checked with a request. v0.16.6 stopped the 5-minute polling but still sent one empty search when a key was added or on "Check now". At SerpSearch's request, nothing is sent at all now. A new key shows as unchecked until its first real search succeeds, and a key that gets 401/403 on a real search is benched (#1635, [#1599]).
- onomeo. New adapter for onomeo's free OpenAI-compatible gateway (30 calls per 5 hours, shared across its free models). The "— via onomeo.com" footer onomeo adds to free-tier answers is stripped, including from streams. Models arrive through the catalog: Premium now, free tier on 20 November (#1610, [#1460]).
- ModelScope daily cap updated to 100, matching its new credit-based free quota (#1045).
- Free.ai keys are no longer health-probed, since no endpoint can tell a real key from a fake one (#1540).
- Model discovery no longer labels paid models as free (#1528).
Routing and caps
- Per-key monthly caps are now enforced for rerank (#1627) and moderation (#1625), and the router's usable-key checks respect them too (#1626).
- A served search, embedding, rerank, decision, moderation or media request now marks its key as working (#1612, [#1620], [#1625]).
- Rerank traffic shows up in the requests log (#1621).
- Proxy fail-closed (opt-in). With
PROXY_REQUIRED_PLATFORMS, listed providers refuse to send traffic when their proxy is unavailable (#1539).
API
GET /v1/usagenow reports spend per key alongside its monthly cap (#1614, [#1568]). Documented in the API docs (#1615).GET /v1/models/:id(OpenAImodels.retrieve) is served from the same catalog as/v1/models(#1632).- The listed Claude aliases now work on the OpenAI-format routes (#1613).
- Model probes (opt-in). Set
MODEL_PROBE_ENABLED=1to probe each model on a schedule and label its status.POST /api/health/probe-modelsruns a pass on demand either way (#1492, [#1438]).
CLI
- Setup notes detect an installed Codex CLI or ChatGPT desktop app (#1443).
Updating
Desktop installers below cover Windows, Apple Silicon and Intel macOS, and Linux. Docker: ghcr.io/tashfeenahmed/freellmapi:v0.16.7 (latest also points here). Existing keys and profiles are retained. One additive migration (a new model_probe_status table) runs automatically on start.
Thanks to everyone who contributed to this release; you're credited in the README (#1633).
Full changelog: https://github.com/tashfeenahmed/freellmapi/compare/v0.16.6...v0.16.7
⭐ Like the free router? Go Premium — the live signed catalog, $19/yr, cancel anytime.
