Add Public AI (TEE) inference provider

[sulijia]

What this adds

Registers publicai-tee as a new inference provider, backed by an OpenAI-compatible, TEE-hosted endpoint (https://tee.publicai.io). It serves conversational (chat-completion) models.

The endpoint follows the OpenAI chat-completions spec, so the helper simply extends BaseConversationalTask — no custom route/payload/response handling is needed.

Changes

• src/providers/publicai-tee.ts — new PublicAITeeConversationalTask (extends BaseConversationalTask, base URL https://tee.publicai.io)
• src/lib/getProviderHelper.ts — register the provider for the conversational task
• src/types.ts — add publicai-tee to INFERENCE_PROVIDERS and PROVIDERS_HUB_ORGS (org: publicai-tee)
• src/providers/consts.ts — add the provider to HARDCODED_MODEL_INFERENCE_MAPPING
• README.md — list the provider and its supported-models endpoint
• test/InferenceClient.spec.ts — add PublicAI TEE conversational integration tests (gated on HF_PUBLICAI_TEE_KEY)

Verification

Against the package (packages/inference):

• tsc (check) — passes
• eslint (lint:check) — passes
• oxfmt --check (format) — passes

The endpoint was manually verified to return OpenAI-compatible responses for both non-streaming and streaming chat completions (id / created / model / choices / usage present; SSE chunks end with data: [DONE]).

🤖 Generated with Claude Code

---

Note

Low Risk
Additive provider registration with no changes to shared routing or auth logic; behavior matches existing OpenAI-compatible conversational providers like publicai.

Overview
Adds publicai-tee as a new inference provider for conversational (chat completion) calls against the OpenAI-compatible TEE endpoint at https://tee.publicai.io, via a thin PublicAITeeConversationalTask on BaseConversationalTask.

Wires the provider through INFERENCE_PROVIDERS, PROVIDERS_HUB_ORGS, getProviderHelper, and HARDCODED_MODEL_INFERENCE_MAPPING, and documents it in README.md (provider list and partners models API). Integration tests cover non-streaming and streaming chatCompletion when HF_PUBLICAI_TEE_KEY is set.

^{Reviewed by Cursor Bugbot for commit 88767d1. Bugbot is set up for automated code reviews on this repo. Configure here.}

[sulijia]

Hi @Wauplin @hanouticelina @SBrandeis @celinah 👋

This PR is the JS client integration step of registering Public AI (TEE) as a new inference provider, following the register-as-a-provider guide.

Details:

• Provider id: publicai-tee
• Owning Hub org (Team plan): publicai-tee
• Endpoint: https://tee.publicai.io — OpenAI-compatible, TEE-hosted; serves conversational (chat-completion) models, so the helper just extends BaseConversationalTask (no custom route/payload/response handling).

I manually verified the endpoint returns OpenAI-compatible responses for both non-streaming and streaming chat completions (id / created / model / choices / usage present; SSE chunks end with data: [DONE]). Locally, tsc, eslint, and oxfmt --check all pass on the package.

Could you please approve the CI run and take a look when you have a moment? Also happy to hear the next steps for server-side enablement of the provider. Thanks a lot! 🙏