[Provider] Finish remaining OpenAI-compatible provider batch: Chutes, Featherless, Novita

[olddonkey]

Area

Provider registry / model discovery / catalog / CLI / Dashboard / documentation

Goal

Finish the remaining providers from the accepted OpenAI-compatible provider-promotion batch.

The original umbrella contained twelve providers. The shared discovery/security contract and nine provider presets have already landed on dev.

This issue now tracks only the three remaining candidates:

• chutes
• featherless
• novita

Keep this umbrella open until each remaining candidate is either merged or explicitly deferred with the reason recorded.

Delivered work

Shared discovery/security contract

#652 landed the common provider-discovery contract, including:

• registry-owned discovery destination/query policy;
• bounded response bytes and accepted row counts;
• declarative eligibility filtering;
• redirect refusal;
• outbound destination-policy enforcement;
• fixed-preset host protection;
• same-named custom-provider destination preservation;
• fixture-based regression coverage rather than live-provider CI dependencies.

Provider presets already delivered

The following original candidates are now first-class providers on dev:

• baseten — feat(providers): add Baseten Model APIs preset #653
• deepinfra — feat(providers): add DeepInfra preset #747
• hyperbolic — feat(providers): add Hyperbolic preset #751
• sambanova — feat(providers): add SambaNova and Nebius presets #870
• nebius — feat(providers): add SambaNova and Nebius presets #870
• digitalocean — feat(providers): add DigitalOcean and Scaleway presets #872
• scaleway — feat(providers): add DigitalOcean and Scaleway presets #872
• nscale — feat(providers): add Nscale and Vultr presets #937
• vultr — feat(providers): add Nscale and Vultr presets #937

These providers are complete for this umbrella and should not be reimplemented here.

Remaining work

Chutes

Tracked by draft PR #1315.

Expected preset:

id: chutes
base URL: https://llm.chutes.ai/v1
adapter: openai-chat
auth: Bearer API key

The implementation must retain the established safety properties:

• registry-owned fixed destination;
• existing same-named custom destinations must not be silently retargeted;
• bounded public-model discovery;
• admit only models with provider evidence for the required chat/tool capability;
• provider-native model IDs preserved;
• no false claim that the public model catalog validates the supplied API key;
• fixture-only tests;
• explicit security review for the credential/destination contract.

Do not broaden this preset to arbitrary user-hosted Chutes endpoints or unrelated Chutes APIs.

Featherless

Tracked by draft PR #1317.

Expected preset:

id: featherless
base URL: https://api.featherless.ai/v1
adapter: openai-chat
auth: Bearer API key

The major catalog constraint remains important: Featherless exposes a very large model catalog.

Discovery must therefore remain explicitly bounded and fail closed rather than downloading or admitting the complete catalog.

The current implementation direction should preserve:

• bounded response size and row count;
• first-page / bounded discovery behavior;
• current-plan eligibility;
• chat-model eligibility;
• tool-use eligibility;
• exclusion of gated/ineligible rows;
• conservative capability metadata;
• authenticated key-validation boundary where supported;
• same-named custom-provider destination preservation;
• fixture-only tests;
• explicit maintainer security review.

Novita AI

Tracked by draft PR #1318.

Expected preset:

id: novita
base URL: https://api.novita.ai/openai/v1
adapter: openai-chat
auth: Bearer API key

Earlier review blocked canonical promotion on operating-entity/provider evidence rather than on the wire protocol itself.

The current follow-up contains updated evidence and should be judged on that record rather than the older blocker.

The implementation should preserve:

• fixed registry-owned destination;
• same-named custom-provider destination preservation;
• bounded model discovery;
• admission only for rows positively identified as compatible chat models;
• provider-native slash-containing model IDs;
• conservative capability claims;
• public model-list success must not be treated as proof that an API key is valid;
• fixture-only tests;
• explicit maintainer/security review of the canonical-provider evidence.

Common acceptance contract

Each remaining provider must:

• be an explicit canonical registry allowlist entry;
• use the documented fixed OpenAI-compatible endpoint;
• preserve an existing same-named custom provider's destination;
• bound discovery response bytes before unbounded parsing/caching;
• bound accepted raw model rows;
• exclude models that cannot serve the supported chat/tool path;
• preserve provider-native model IDs, including embedded /;
• avoid claiming unsupported reasoning, parallel-tool, multimodal, or structured-output capabilities;
• expose the preset consistently through CLI, Dashboard/provider workspace, and documentation;
• use fixture-based deterministic tests rather than live provider endpoints in CI;
• avoid logging or exposing API-key material;
• receive required security review for new credential destinations;
• merge into dev, or be explicitly deferred with the reason recorded here.

Active PRs

• feat(providers): add Chutes preset #1315 — Chutes
• feat(providers): add Featherless preset #1317 — Featherless
• feat(providers): add Novita AI preset #1318 — Novita AI

All three intentionally relate to this umbrella rather than auto-closing it individually.

Close condition

Close this issue as completed when all three remaining providers have either:

1. landed on dev; or
2. been explicitly declined/deferred with a documented reason and no remaining implementation is expected under this umbrella.

Do not reopen completed providers from the original batch here. Future providers with materially different authentication, transport, or discovery contracts should use separate issues.

[github-actions]

_{Automated translation bookkeeping — detected language: English.}

[Ingwannu]

Thanks for the unusually thorough proposal and endpoint verification. The direction is useful, but this should remain an umbrella issue rather than become one review-sized change containing all twelve providers.

Please proceed in phases against dev:

1. Land the shared discovery contract first: registry-derived discovery URL/query/filtering, bounded response bytes and accepted row count, chat-model eligibility filtering, representative metadata parsing, and same-named custom-provider destination preservation. Keep tests fixture-based; CI must not depend on live provider endpoints.
2. Add providers in small reviewable batches (roughly 2–4 with the same discovery shape), with registry/CLI/dashboard/docs parity and provider-specific fixtures in each PR.
3. Keep providers requiring a different transport, authentication scheme, or arbitrary prediction schema out of this issue. Together AI remains the existing baseline and must not be re-added.

The safety requirements in the report are acceptance criteria, especially preventing an existing custom provider/key from being silently retargeted and preventing oversized mixed catalogs from entering the Codex catalog. A provider should not claim streaming, tools, structured output, or multimodal support beyond what its selected upstream models and official documentation actually establish.

I am assigning this to you. You do not need to implement every provider in one PR; reference this umbrella issue from each phased PR and target dev.

[lidge-jun]

Phase 1 landed: #652 merged as 48f2e8362.

That is the shared discovery contract you were asked to land first — registry-owned endpoint/query policy, declarative eligibility filters, bounded response bytes and row count, redirect: "error" so credentials cannot follow a redirect, destination-policy checks before fetch, OAuth host pinning for fixed presets, and same-named custom-provider collision preservation.

Verified on the merged dev tip rather than from the PR branch: bun x tsc --noEmit clean, and 151 tests pass across the discovery-contract, provider-registry-parity, and codex-catalog suites.

Keeping this umbrella open for the remaining provider batches. Next up is #653 (Baseten pilot), which no longer has the contract as a blocker. Its windows-latest job is currently failing on the Test step — I confirmed that is a genuine failure rather than a cancelled run, and I have dispatched a rerun now that the prerequisite is on dev. Worth noting for the follow-on batches: the same PR's focused suites pass locally here (38 pass, typecheck clean), so if the Windows failure persists it is more likely a runner-level problem than a defect in the preset itself — the earlier #610 investigation hit EEXIST: epoll_ctl from the Bun test runner on Linux CI, and I could not find a matching upstream Bun issue to pin it to.

Thanks for structuring this as phases. Landing the contract before the twelve presets is what made #652 reviewable as one security-boundary change instead of twelve.