Request lifecycle
Admission, routing and accounting in a corporate integration.
Send an inference request to https://api.b2b-llm.com using a Model API credential. A request is admitted only after credential state, company activation, project permissions, IP rules and applicable budgets permit it. The configured model route then determines which upstream capacity can execute the request.
Keep these values distinct
| Value | Purpose |
|---|---|
| Canonical model ID or alias | Selects a configured model |
| Credential | Establishes identity and project context |
| Company balance | Pays the actual settled charge |
Key/project usage | Enforces resettable usage budgets |
| Official catalog rate | Reference price, not necessarily your charge |
Model-route eligibility remains authoritative even when a key or project prefers one upstream. An only policy does not create missing model routes.
Before sending production traffic
Read /v1/models, use the protocol supported by the chosen model and verify a small request. Pass only parameters accepted by that model. Preserve request IDs from error/success responses for investigation. Do not implement an unbounded retry loop: a connection failure does not establish that no work occurred upstream.
Examples below and in the endpoint pages illustrate payload shape rather than current availability or pricing. Decimal financial values must be handled with exact decimal arithmetic. Consult company pricing for the separation between actual charge and budget valuation.
Images and configured capabilities
Text rates and capability badges do not determine image charges. Use your authorized project key to request the available Images offers and a quote for the actual operation. The endpoint index and Images workspace explain the client workflow. A recorded charge is not an immediate company balance or project-usage update. For recordings, use Playground Audio and the transcription API. Audio is separate from text chat and Images.