Skip to main content

Interface: MaxCostRequest

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:193

What a request sends, for estimateMaxCallCostUSD.

Properties​

cacheMarkers?​

optional cacheMarkers: boolean

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:221

The request asks for prompt caching: cache_control on the request or a block (Anthropic), cache settings or a breakpoint (OpenAI's prompt_cache_options, prompt_cache_breakpoint, prompt_cache_retention). A cache_control or prompt_cache_breakpoint key in the messages, the system prompt or the tools counts as well.


maxOutputTokens?​

optional maxOutputTokens: number

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:213

The request's output cap: max_tokens, max_completion_tokens or max_output_tokens, as the request sends it. A call with none has no bound. A host that cuts a cap above the model's ceiling passes the cut cap: the ceiling a library table knows shapes a request, and is never taken as what the provider will bill.


messages​

messages: unknown

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:200

The messages (or a Responses API input) as the provider is sent them. A string counts as one message. The bound counts what the request carries: a request that draws in context stored with the provider (OpenAI's previous_response_id, conversation or a stored prompt) is billed for tokens its text does not show, so a host sends that context here or refuses the request.


n?​

optional n: number

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:215

Completions asked for (n, best_of): each is billed its own output, so a request for more than one has no bound.


nonTextParts?​

optional nonTextParts: boolean

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:226

The request carries a content part that is not text (an image, audio, a file, a document), or asks for output that is not text (OpenAI's modalities with audio): billed by its own measure.


priorityTier?​

optional priorityTier: boolean

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:234

The request asks for a faster tier at a higher price (OpenAI's service_tier: 'priority' or fast mode, Anthropic's speed: 'fast').


providerSideTools?​

optional providerSideTools: boolean

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:232

The request names a tool the provider runs or defines (Anthropic's server tools, the Responses API's built-in tools), or tools the provider fetches itself (Anthropic's mcp_servers), whose definitions the request's text does not hold.


regionalProcessing?​

optional regionalProcessing: boolean

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:236

The request is processed in one region at an uplift (Anthropic's inference_geo: 'us', OpenAI's regional endpoints).


system?​

optional system: unknown

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:205

The system prompt (or instructions), and anything else the request sends that the provider bills as input text, such as a structured-output schema: its text is counted with the system prompt's.


tools?​

optional tools: unknown

Defined in: packages/agentos/src/api/runtime/modelPricing.ts:207

The tool definitions as the provider is sent them.