Skip to content

feat(providers): add Unifically as an OpenAI-compatible gateway - #14124

Closed
iliadg wants to merge 1 commit into
cheaper-inference:release/v3.8.51from
iliadg:feat/unifically-provider
Closed

iliadg wants to merge 1 commit into
cheaper-inference:release/v3.8.51from
iliadg:feat/unifically-provider

Conversation

@iliadg

@iliadg iliadg commented Sep 18, 2026 •

Copy link
Copy Markdown

Adds Unifically as an OpenAI-compatible API-key gateway. Disclosure: I run Unifically.

Registry entry in the eurouter / greenpt shape (#13025, #13024): default executor, no translator, passthroughModels: true, models: [].

Checked live, 2026-09-18

GET  https://api.unifically.com/v1/models                -> 200, OpenAI-shaped list, 155 models
GET  https://api.unifically.com/v1/models?category=llm   -> 200, 30 models
GET  https://api.unifically.com/v1/models  (no key)      -> 401
POST https://api.unifically.com/v1/chat/completions      -> 200, chat.completion
POST .../chat/completions  stream:true                   -> SSE chat.completion.chunk, final usage chunk, data: [DONE]

Auth is Authorization: Bearer <key>. Streaming was exercised with google/gemini-3.5-flash.

Why modelsUrl carries ?category=llm

The catalog is one list for text, image, video and audio (category: 30 llm, 41 image, 50 video, 34 audio). Only the llm models are chat-completion targets. The media models run through an async task API (POST /v1/tasks, then poll or webhook), which the default executor does not speak. An unscoped listing would offer 125 models that fail on every request, so discovery is scoped and there is a test pinning it.

What is not claimed

  • No tool or vision capability is declared. Neither was exercised.
  • hasFree is false. New accounts get a one-time starting balance, not a recurring free tier, so no badge.
  • serviceKinds is ["llm"] only. Media support would need an executor in the shape of open-sse/executors/kie.ts plus videoRegistry / audioRegistry entries. I can send that as a follow-up if it is wanted.
  • One known gap: in the streamed response the final usage chunk reported completion_tokens: 0 while the non-streamed call reported 3. That is on our side and I am fixing it upstream; it does not affect routing.

Added to AGGREGATOR_PROVIDER_IDS, since it routes to third-party model providers.

Verification

tests/unit/unifically-provider.test.ts: 7 passed. tests/unit/eurouter-provider.test.ts still 7 passed. Prettier and ESLint clean on the touched files. Changelog fragment included. gateways.ts grows by 19 lines of catalog data, so file-size-baseline.json will want the usual rebaseline note on merge.

Registry entry in the eurouter / greenpt shape: default executor, no
translator, passthroughModels with discovery scoped to
/v1/models?category=llm so the image, video and audio models in the same
catalog are not offered as chat-completion targets.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@iliadg

iliadg commented Sep 19, 2026

Copy link
Copy Markdown
Author

Closing this. I will resubmit it from the project's own account.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant