Supported paid providers
| Provider | Supported V0 path | Valuation / required configuration | Complete evidence |
|---|---|---|---|
| OpenAI / tested compatible APIs | POST /chat/completions and /completions; JSON or fully consumed SSE | Requested and returned model IDs configured; rates per million input/output tokens; cached rate when cache usage is nonzero; max_request_usd | Valid nonnegative consistent token counters, supported text usage/tier, final stream usage and terminal evidence |
| Firecrawl | v2 POST /batch/scrape followed by owned GET /batch/scrape/{id} | per_credit_usd and max_request_usd per created batch | Terminal settlement with valid credits/pages and completed observed responses; monotonic resource history |
| Exa | POST /search | per_request_usd, or explicit allow_estimate: true; max_request_usd | Valid consistent returned costDollars evidence; estimate provenance retained |
Each may set base_url and credential_env. Defaults are https://api.openai.com/v1, https://api.firecrawl.dev/v2 and https://api.exa.ai. Do not put credentials, query strings or fragments in base URLs. Application environment/SDK configuration must match the chosen endpoint/model; declaring a provider does not change the application.
OpenAI rates have no automatic pricing lookup. Configured rates carry configured_marginal_rate. Missing usage/model rates, cache usage without a cache rate, contradictory counters, nondefault tiers, unsupported paid tools/audio or unfinished streams cannot certify. SDKs must use supported fetch/http/https transport; older custom Agent/node-fetch routing may require the SDK’s documented global fetch option. No claim covers every SDK or model.
Firecrawl creation is unknown pending work until terminal settlement. Poll responses are observations, not duplicate spend. Unowned IDs, decreasing/contradictory counts, reopened terminal resources, unconsumed status responses and unresolved jobs make the comparison INCONCLUSIVE. Webhooks, unowned remote work and additional endpoints are excluded.
Exa costDollars remains a provider-reported estimate, not an invoice charge. A configured per-request rate replaces valuation but keeps returned estimate evidence. Missing, negative or contradictory economic evidence is rejected even with a configured rate. Additional search endpoints/modes are not automatically certified.
Live checks have covered a tested OpenAI-compatible endpoint with ordinary JSON and completed SSE, a one-page Firecrawl v2 batch through terminal settlement, and a one-result Exa search returning a consistent estimate. These checks do not establish live certification for OpenAI-hosted APIs, every compatible model, nonzero cache/reasoning usage, nondefault tiers, larger or paginated Firecrawl outcomes, all Exa modes, or invoices. Stay within the endpoint and usage requirements above; unknown required usage produces INCONCLUSIVE.