LongCat text and streaming source pilot

This source-level catalog entry selects Meituan's LongCat-2.5-Preview model through https://api.longcat.chat/openai/v1/chat/completions with Authorization: Bearer $LONGCAT_API_KEY. Its declared scope is text generation and streaming only. Source preparation, public contract acceptance, installed-package acceptance, current account/API availability and delivery are separate gates. This entry gives no provider-count, roster, live, package or delivery credit.

What the entry declares

Field	Value
Provider id	longcat (display name LongCat, operator Meituan)
Base URL	https://api.longcat.chat/openai/v1
Route	POST /chat/completions
Credential	LONGCAT_API_KEY as a Bearer token
Default model	LongCat-2.5-Preview (the only model listed, so no other fallback)
Context window	1,048,576 tokens
Maximum output	131,072 tokens
Capabilities	text and streaming; tools, structured output, thinking, vision off
Billing policy	no-free-tier, usage-priced pay-as-you-go

LongCat-2.0 is served on the same route but is outside this pilot's catalog scope, so it is deliberately not a default and not a fallback. The capability booleans bound this pilot's declared scope; they are not evidence that the vendor lacks those features.

Billing

The default model is usage-priced. Promotional credits, when an account has any, are consumed first, but no free allowance is stated, so the entry claims no free tier. The vendor's Pay-As-You-Go page says account verification is required before a top-up, and its FAQ scopes that requirement to users in Mainland China; it was not tried on an account.

Evidence caveats

• The context integer 1048576 rests on the vendor's own Retrieve Model response sample for the exact model id plus the 1M prose; no normative sentence states it. It has not been confirmed with GET /openai/v1/models/LongCat-2.5-Preview because that needs a key.
• The vendor SDK snippet omits /v1 although the exact HTTP route includes it; the catalog base URL https://api.longcat.chat/openai/v1 is taken from the curl example and is proven only against the owned fixture.
• The vendor's Pay-As-You-Go page says account verification is required before top-up; its FAQ scopes that to users in Mainland China. Both were read directly; it was not tried on an account. No live call was made.

Local public contract

test/continuous-test-suite-longcat-pilot.ts runs eight 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, /openai/v1/chat/completions, Bearer header, unique user prompt, max_tokens: 32, stream mode, no tools/tool choice, answer
SDK/CLI generate/stream invalid explicit model	4	All wire preconditions pass before HTTP 400 rejection; no fallback or successful reply

The fixture strips real credentials before imports, uses temporary homes and points the provider at an 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. The listener is the only server involved: no vendor host is contacted.

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

Provenance and remaining gates

The legacy evidence.rosterVerified slot carries the explicit VENDOR_DOCS_ONLY_NOT_ROSTER_VERIFIED marker. It has no HTTP status, auth probe or billing probe; liveMatrix is null. evidence.addedInPR is the introduction pull request URL (https://github.com/juspay/neurolink/pull/1980), set after the PR existed; it was PENDING_PR while the work was local. Source-stage admission does not certify current account or API availability.

Still pending and separate: an authenticated GET /openai/v1/models roster, a funded-account live matrix, installed-consumer acceptance and any delivery claim.