-
feat(providers): keyless Kilo free models + Gemma 4 (Google AI Studio) (#181)
frostbyte_neo 发布于
2026-06-03 08:09:47 +00:00 Two genuinely-free additions, each verified by live API probing (cost:0 /
HTTP 200) and an end-to-end routing smoke test.Keyless provider support (Kilo's anonymous gateway):
- BaseProvider gains a
keylessflag; OpenAICompatProvider omits the
Authorization header when set (a bogus bearer could be rejected). - Keys route accepts an empty key for keyless providers, stores a sentinel
row so the router still sees the platform as configured, and de-dupes
(one sentinel row per keyless platform). Non-keyless providers still
require a key. - Keys page disables the key field for keyless providers and shows a
"No API key needed" note; the button reads "Enable". - Kilo registered
keyless: true; health check validates against
/api/gateway/models (the /v1/models path only accepts POST → 405).
Catalog:
- V20: 4 free Kilo models (200 req/hr per IP, anonymous) —
poolside/laguna-m.1, poolside/laguna-xs.2,
nvidia/nemotron-3-super-120b-a12b, stepfun/step-3.7-flash. - V19: Gemma 4 on Google AI Studio free tier — gemma-4-31b-it and
gemma-4-26b-a4b-it (reached with the existing Gemini key).
Note: laguna-m.1 and nemotron-3-super-120b already exist via OpenRouter
(keyed, ~20/day). The Kilo rows route the same models keyless at 200/hr and
add fallback redundancy.下载附件
- BaseProvider gains a