AIHubMix text and streaming pilot

This local source pilot selects AIHubMix's paid gpt-4o-mini model through https://aihubmix.com/v1/chat/completions. Its declared scope is text generation and streaming. Source preparation, public contract acceptance, installed-package acceptance, current account/API availability and delivery are separate gates. This draft gives no delivery or provider-target credit.

Exact source facts

AIHubMix's own model page publishes a model-specific server record at loaderData.model.record. It binds model: "gpt-4o-mini" to numeric context_window: 128000 and max_output: 16384. The same record publishes retire_stage: "active". These exact fields support the pilot's context and output ceilings. The displayed 128K and 16.4K labels are rounded; no K/M conversion or upstream-model substitution was used. An active documentation label does not certify access on an account or a successful live API call.

The model's own guide documents POST /v1/chat/completions, Authorization: Bearer $AIHUBMIX_API_KEY, ordinary messages/choices responses, and SSE data: JSON chunks ending in [DONE]. Its tools, structured-output, vision and reasoning capability flags are explicitly unconfirmed. The other capability booleans in this profile limit the pilot's declared scope; these flags are not evidence that the vendor lacks those features. Optional prices are omitted from the catalog draft.

Billing and retired models

The default gpt-4o-mini is usage-priced and requires the caller to establish funding, permissions and model eligibility. The existing schema's free-tier value describes the vendor-level free-model programme, following the existing Mancer profile, whose default is also paid. AIHubMix's separate programme is model-limited. Its published no-card and allowance claims do not grant free access to this paid default or certify current eligibility on any account.

Current model-specific guides mark gpt-4o-free and gpt-5.5-free retired and describe HTTP 404 model_retired. Both are excluded from the catalog and its default/fallback list. Older free-model marketing and readable static parameter schemas cannot establish current availability of these IDs.

AIHubMix documents remote mapping, routing and fallback. The SDK's disableInternalFallback option controls the SDK path; acceptance of gateway identity, routing and billing remains a separate live/provider question.

Local public contract

test/continuous-test-suite-aihubmix-pilot.ts prepares 16 mandatory cases through the built public SDK and built CLI, with optional --package-root support for a later installed consumer:

Cases	Count	Required observation
SDK/CLI generate/stream default	4	Actual catalog default, route, Bearer header, unique user prompt, max_tokens: 32, stream mode, no tools/tool choice, and fresh backend answer
SDK/CLI generate/stream invalid explicit model	4	All wire preconditions pass before HTTP 400 rejection; no fallback or successful reply
SDK/CLI generate/stream each retired explicit ID	8	Same wire preconditions, HTTP 404 retirement cause, preserved caller rejection, no fallback or reply

The fixture strips real credentials before imports, uses temporary homes and sets provider routes to its owned random-port listener. Public MODEL_CONFIG_URL uses that same listener's HEAD /health and GET /api/v1/models. A schema-valid empty registry keeps the provider's actual catalog default in control. Positive metadata initialization requires the exact ordered request pair; explicit CLI model paths require no metadata request. Successful SDK metadata checks run after shutdown, and preserve original SDK errors when generation fails. CLI commands keep a 60-second deadline.

The standalone component now passes all 16 cases with zero failures, skips or unowned network attempts and known process cleanup. The fixture remains byte-identical to the original source freeze. Runtime receipt AIHUB-20261009T171326Z-11 binds these results to the fresh standalone build. Integrated and installed-consumer acceptance remain separate.

The exact validated Q04 source invocation and public-suite command are:

pnpm exec tsx tools/verify-provider-onboarding.ts --catalog-stage source --provider aihubmix
pnpm exec tsx test/continuous-test-suite-aihubmix-pilot.ts

The verification lane executes the public command through the existing credential-free owned-command runner with RECOVERY_PROOF_TEST_GUARD=true, separate process groups and recorded cleanup. The guard must record zero unowned attempts. The source invocation has no extra forwarded -- argument.

Provenance and remaining gates

The legacy evidence.rosterVerified slot contains the explicit PUBLIC_DOCS_ONLY_NOT_ROSTER_VERIFIED marker and documentation methods. It has no HTTP status, auth probe or billing probe. liveMatrix is null and the PR field is the canonical introduction PR URL (https://github.com/juspay/neurolink/pull/1966), set after the PR existed; it was PENDING_PR while the work was local. Q04 source admission accepts either the marker or a canonical introduction PR URL. Its source-stage report keeps roster, live, installed-package and delivery claims separate and gives no provider-count credit. Structural parsing or source-stage admission does not certify current account/API availability.

The Q04 successor starts from 08c3c4e4e2881874fdf0e453923a3507b759532d, which descends from the checked live release 8c78cd172a8b2036b75132488f9d24548f2a1e88. Its public fixture is byte-identical to the original source freeze. The profile's only semantic change is the PR marker required by Q04. Shared schema, runtime, helpers and original source receipts are preserved. Actual Q04 source admission passed and reports zero provider-count, roster, live, installed-package and publication credit.

Canonical catalog generation affects the index, provider union, provider/model enum regions and generated credential region. The standalone frozen install, canonical generation and fixed point, fresh SDK/CLI/browser build, normal lint, strict source/tools/CI types and all 16 guarded public cases passed. Source outside the generated regions is preserved. A command-delimiter failure before Q04 admission was retained and corrected by invoking the same production validator directly; no source or fixture repair was required.

Integrated source/build/required suites, normal hooks and commit, any separately assigned production sensitivity controls and clean installed-consumer acceptance remain pending. Accounts, funding, provider inference, external delivery and target credit are separate.

The retained source receipts were retrieved on 2026-10-09:

First-party source	Body SHA256
Paid default model page	d30240d57bde0caf42a89cd7c4f96fbbc32217835756321482d362c64586c12e
Paid default model guide	6cc75a6d9fece483ee7417c0e4f9a6272282ef317a57f6c4d664b3766b65de7e
Model field definitions	06c69590bec8c9b808718c541c6a5efc9bd0e54a86c5e81d287d2a2840231999
Model-limited free programme	27cd20549bb9b747c2768eacecf0a66fa58055708432529f3b52c587ad217a22
Retired gpt-4o-free guide	1e1c70b0cd06d56c47f36e73d34487802c7760bd0541073d5227d55ba4267fdf
Retired gpt-5.5-free guide	36e1d73d1e8a47d94b458981888495d3d23671089fb0386d4da431e093778d83