> ## Documentation Index > Fetch the complete documentation index at: https://docs.getbifrost.ai/llms.txt > Use this file to discover all available pages before exploring further. # Evaluate a decision (OpenAI format) > OpenAI's Decisions API: asks an ordered list of questions about an input and returns one answer per question, in order. Any decision model can answer: OpenAI's decisions models such as `gpt-6-luna`, Typesafe's jev models, or any chat model through structured output. The request is OpenAI's: a top-level `state` and a field a question, choice, or level does not define (such as `criteria`) are rejected rather than translated, while other top-level fields are forwarded unchanged. Answers come back in OpenAI's response shape, with `name: null` for a question sent without a name, plus Bifrost's `extra_fields` as on the other OpenAI routes; when raw responses are enabled and OpenAI served the request, OpenAI's own response body is returned instead. **Note:** This endpoint also works without the `/v1` prefix (e.g., `/openai/decisions`). ## OpenAPI ````yaml /openapi/openapi.json post /openai/v1/decisions openapi: 3.1.0 info: title: Bifrost API description: > Bifrost HTTP Transport API for AI model inference and gateway management. This API provides a unified interface for interacting with multiple AI providers including OpenAI, Anthropic, Bedrock, Gemini, and more through a single API, along with comprehensive management APIs for configuring and monitoring the gateway. ## API Structure ### Unified Inference API (`/v1/*`) The primary API using Bifrost's unified format. Model parameters use the format `provider/model` (e.g., `openai/gpt-4`, `anthropic/claude-3-opus`). ### Async Inference API (`/v1/async/*`) Submit inference requests for asynchronous execution. Returns a job ID immediately and allows polling for results. Supports all inference types except batches, files, and containers. ### Provider Integration APIs Native provider-format APIs for drop-in compatibility: - `/openai/*` - OpenAI-compatible API - `/anthropic/*` - Anthropic-compatible API - `/genai/*` - Google GenAI (Gemini) compatible API - `/bedrock/*` - AWS Bedrock compatible API - `/cohere/*` - Cohere compatible API ### Framework Integration APIs Multi-provider proxy endpoints for AI frameworks: - `/litellm/*` - LiteLLM proxy with all provider formats - `/langchain/*` - LangChain compatible endpoints - `/pydanticai/*` - PydanticAI compatible endpoints ### Management APIs (`/api/*`) APIs for managing and monitoring the Bifrost gateway: - `/api/config` - Configuration management - `/api/providers` - Provider and API key management - `/api/plugins` - Plugin management - `/api/governance/*` - Virtual keys, teams, customers, budgets, rate limits, routing rules, and pricing overrides - `/api/logs` - Log search and analytics - `/api/mcp/*` - MCP (Model Context Protocol) client management - `/api/session/*` - Authentication and session management - `/api/cache/*` - Cache management - `/health` - Health check endpoint ## Fallbacks Requests can include fallback models that will be tried if the primary model fails. version: 1.0.0 contact: name: Contact Us url: https://getmaxim.ai/bifrost license: name: Apache 2.0 url: https://opensource.org/licenses/Apache-2.0 servers: - url: '{baseUrl}' description: Your Bifrost instance variables: baseUrl: default: http://localhost:8080 description: >- Base URL of your Bifrost instance (e.g. https://bifrost.mycompany.com, or https://mycompany.com/bifrost when served under server.base_path) security: - BearerAuth: [] - BasicAuth: [] - ApiKeyAuth: [] tags: - name: Agent Gateway description: >- Agent registrations, A2A protocol calls, push callbacks, and operation history - name: Models description: Model listing and information - name: Chat Completions description: Chat-based text generation - name: Text Completions description: Text completion generation - name: Responses description: OpenAI Responses API compatible endpoints - name: OCR description: Optical character recognition for documents and images - name: Rerank description: Document reranking by relevance to a query - name: Decisions description: Structured decisions evaluated against annotated function-tool definitions - name: Embeddings description: Text embedding generation - name: Images description: Image generations, editing, and variations - name: Videos description: Video generation and management - name: Audio description: Speech synthesis and transcription - name: Count Tokens description: Token counting utilities - name: Batch description: Batch processing operations - name: Files description: File management operations - name: Containers description: Container management operations - name: Async Jobs description: Asynchronous job submission and retrieval endpoints - name: Realtime description: Realtime WebSocket and WebRTC endpoints - name: OpenAI Integration description: OpenAI-compatible API endpoints (/openai/*) - name: Azure Integration description: Azure OpenAI integration endpoints - name: Anthropic Integration description: Anthropic-compatible API endpoints (/anthropic/*) - name: GenAI Integration description: Google GenAI (Gemini) compatible API endpoints (/genai/*) - name: Bedrock Integration description: AWS Bedrock compatible API endpoints (/bedrock/*) - name: Cohere Integration description: Cohere compatible API endpoints (/cohere/*) - name: Typesafe Integration description: Typesafe compatible API endpoints (/typesafe/*) - name: LiteLLM Integration description: LiteLLM proxy endpoints with multi-provider support (/litellm/*) - name: LangChain Integration description: LangChain compatible endpoints with multi-provider support (/langchain/*) - name: PydanticAI Integration description: >- PydanticAI compatible endpoints with multi-provider support (/pydanticai/*) - name: Health description: Health check endpoints - name: Configuration description: Configuration management endpoints - name: Session description: Session and authentication endpoints - name: Providers description: Provider management endpoints - name: Plugins description: Plugin management endpoints - name: MCP description: Model Context Protocol endpoints - name: Governance description: Virtual keys, teams, and customers management - name: Routing description: Routing rules and complexity analyzer configuration - name: Logging description: Log search and management endpoints - name: Cache description: Cache management endpoints - name: Vault description: Vault secret management endpoints - name: Skills description: Skills Repository management, marketplace, and download endpoints - name: Audit Logs description: >- CADF-compliant audit log search, export, and signature verification endpoints - name: Webhooks description: Webhook endpoint management and signed async-job delivery history - name: Notifications description: >- Role-targeted dashboard notifications, delivered over the dashboard WebSocket - name: News Feeds description: >- RSS and Atom news sources shown in the News tab of the dashboard notification tray - name: Background Jobs description: >- Status, progress and cancellation of durable background jobs (cost recalculation, Warp log indexing, and others) - name: Warp description: >- Configuration for Warp, the dashboard agent that answers questions about the deployment's own telemetry - name: Relays description: >- Relays to non-LLM HTTP APIs - endpoint management, usage preview, Relay Logs and the /relay data plane paths: /openai/v1/decisions: post: tags: - OpenAI Integration summary: Evaluate a decision (OpenAI format) description: > OpenAI's Decisions API: asks an ordered list of questions about an input and returns one answer per question, in order. Any decision model can answer: OpenAI's decisions models such as `gpt-6-luna`, Typesafe's jev models, or any chat model through structured output. The request is OpenAI's: a top-level `state` and a field a question, choice, or level does not define (such as `criteria`) are rejected rather than translated, while other top-level fields are forwarded unchanged. Answers come back in OpenAI's response shape, with `name: null` for a question sent without a name, plus Bifrost's `extra_fields` as on the other OpenAI routes; when raw responses are enabled and OpenAI served the request, OpenAI's own response body is returned instead. **Note:** This endpoint also works without the `/v1` prefix (e.g., `/openai/decisions`). operationId: openaiCreateDecision requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/OpenAIDecisionRequest' responses: '200': description: Successful response content: application/json: schema: $ref: '#/components/schemas/OpenAIDecisionResponse' '400': description: Bad request content: application/json: schema: $ref: '#/components/schemas/BifrostError' '500': description: Internal server error content: application/json: schema: $ref: '#/components/schemas/BifrostError' security: - BearerAuth: [] - VirtualKeyAuth: [] - ApiKeyAuth: [] components: schemas: OpenAIDecisionRequest: type: object description: >- OpenAI's Decisions request. A field OpenAI's question, choice, or level does not define (such as Typesafe's `criteria`), instructions that are neither a string nor a list of parts, a top-level `state`, and a structured or null `input` are rejected with a 400; any other top-level field is forwarded to the provider unchanged, so a field a newer SDK sends still reaches it. required: - model - input - questions properties: model: type: string description: >- Model in provider/model format; a model without a provider prefix is OpenAI's example: gpt-6-luna input: description: >- A string of text, or a conversation of messages (with text, JSON, and image parts), function and custom tool calls, and their outputs oneOf: - type: string - type: array items: $ref: '#/components/schemas/DecisionInputItem' questions: type: array minItems: 1 description: Questions in the order asked. Answers follow the same order. items: $ref: '#/components/schemas/OpenAIDecisionQuestion' safety_identifier: type: - string - 'null' description: Stable end-user identifier, forwarded to OpenAI prompt_cache_key: type: - string - 'null' description: >- Partitions prompt-cache reuse by a caller-chosen key, forwarded to OpenAI prompt_cache_options: $ref: '#/components/schemas/DecisionPromptCacheOptions' fallbacks: type: array items: type: string description: >- Bifrost fallback models in provider/model format; never sent upstream OpenAIDecisionResponse: type: object description: >- OpenAI's Decision object, plus Bifrost's extra_fields as on the other OpenAI routes. Every model's answers are returned in this shape, and it carries no id. When raw responses are enabled and OpenAI served the request, OpenAI's own response body is returned instead, as on the other OpenAI routes. required: - model - answers properties: model: type: string description: Model that performed the decision answers: type: array description: One answer per question, in question order items: $ref: '#/components/schemas/DecisionAnswer' usage: type: object properties: input_tokens: type: integer input_tokens_details: type: object description: Present when the provider reports it properties: cached_tokens: type: integer cache_write_tokens: type: integer output_tokens: type: integer output_tokens_details: type: object description: Present when the provider reports it properties: reasoning_tokens: type: integer total_tokens: type: integer state_tokens: type: integer description: Laya only state_tokens_dropped: type: integer description: Laya only truncated: type: boolean description: Laya only truncated_questions: type: array items: type: string description: Laya only routing: type: object description: >- Laya only. The checkpoint that answered and why, passed through untouched extra_fields: $ref: '#/components/schemas/BifrostResponseExtraFields' description: >- Bifrost's response metadata (provider, latency, routing info, and the raw provider body when raw responses are on), as on the other OpenAI routes BifrostError: type: object description: Error response from Bifrost properties: event_id: type: string type: type: string is_bifrost_error: type: boolean status_code: type: integer error: $ref: '#/components/schemas/ErrorField' extra_fields: $ref: '#/components/schemas/BifrostErrorExtraFields' DecisionInputItem: description: One item of a conversation input oneOf: - $ref: '#/components/schemas/DecisionInputMessage' - $ref: '#/components/schemas/DecisionFunctionCall' - $ref: '#/components/schemas/DecisionFunctionCallOutput' - $ref: '#/components/schemas/DecisionCustomToolCall' - $ref: '#/components/schemas/DecisionCustomToolCallOutput' OpenAIDecisionQuestion: type: object additionalProperties: false required: - type - instructions properties: type: type: string enum: - predicate - choice - score name: type: string instructions: description: >- A string, or a list of text, image, and JSON parts that give the question its own images oneOf: - type: string - type: array minItems: 1 items: $ref: '#/components/schemas/DecisionInstructionPart' choices: type: array description: Choice only items: type: object additionalProperties: false required: - value properties: value: oneOf: - type: string - type: boolean description: type: string levels: type: array description: Score only, in order items: type: object additionalProperties: false required: - label properties: label: type: string description: type: string DecisionPromptCacheOptions: type: object description: Prompt-caching options, forwarded to OpenAI; other providers drop them. properties: mode: type: string enum: - implicit - explicit description: >- `implicit` (the default) adds a breakpoint at the end of the shared input; `explicit` uses only the parts marked with `prompt_cache_breakpoint` ttl: type: string enum: - 30m description: Minimum lifetime of every breakpoint the request writes DecisionAnswer: type: object required: - type properties: type: type: string description: The question's type, or `refusal` when the model declined to answer example: choice name: type: - string - 'null' description: >- The question's name, when it had one. OpenAI sends null for an unnamed question probability: type: number description: Predicate answers. Probability in [0,1] that the condition is true choice: description: Choice answers. The chosen value oneOf: - type: string - type: boolean score: type: number description: Score answers. Probability-weighted level index box: type: object description: >- Box answers. Each edge is a fraction from 0 to 1 of the image's height (`top`, `bottom`) or width (`left`, `right`), measured from its top-left required: - top - bottom - left - right properties: top: type: number bottom: type: number left: type: number right: type: number probabilities: type: array description: >- Distribution over choices or levels, in question order, when supplied items: type: object properties: value: description: The choice value, or the level index oneOf: - type: string - type: boolean - type: number label: type: string description: Score answers. The level's label probability: type: number confidence: type: number description: Model confidence for this answer, when supplied legend: type: object additionalProperties: {} description: Score answers. Level index to description, when supplied answer_confidence: type: number description: Laya only. Calibrated probability of the reported answer action: type: object description: Laya only. Action head output, passed through untouched abstention: type: string enum: - passed - abstained - unevaluated description: Laya only, when `min_confidence` is set abstention_threshold: type: number description: Laya only, when `min_confidence` is set low_confidence: type: boolean description: Laya only, when `min_confidence` is set BifrostResponseExtraFields: type: object description: Additional fields included in responses properties: request_type: type: string description: Type of request that was made pricing_request_type: type: string description: Request type used for pricing, when it differs from request_type routing_info: $ref: '#/components/schemas/RoutingInfo' provider: $ref: '#/components/schemas/ModelProvider' deprecated: true description: 'Deprecated: use routing_info.provider.' original_model_requested: type: string deprecated: true description: >- Deprecated: use routing_info.primary_model when routing_info.is_fallback is true, otherwise routing_info.model. For the model the caller asked for, read routing_info.requested_model. resolved_model_used: type: string deprecated: true description: >- Deprecated: use routing_info.resolved_key_alias.model_id when an alias matched, otherwise routing_info.model. latency: type: integer format: int64 description: >- Request latency in milliseconds. On a stream each chunk carries its own latency and the last chunk the total. upstream_latency: type: integer format: int64 description: >- Milliseconds spent waiting on the provider, summed over every attempt, retry and fallback. Absent when unknown. chunk_index: type: integer description: Index of the chunk for streaming responses, 0 otherwise raw_request: description: Raw provider request, when raw request capture is enabled raw_response: description: Raw provider response, when raw response capture is enabled cache_debug: $ref: '#/components/schemas/BifrostCacheDebug' routing_metadata: type: object description: >- Billable internal calls the routing plugin made while classifying the request's complexity properties: calls: type: array items: type: object properties: request_type: type: string provider_used: type: string model_used: type: string input_tokens: type: integer output_tokens: type: integer count_toward_budgets: type: boolean guardrail_debug: $ref: '#/components/schemas/BifrostGuardrailMetadata' parse_errors: type: array description: Errors hit while parsing JSONL batch results items: type: object properties: code: type: string message: type: string param: type: string line: type: integer converted_request_type: type: string dropped_compat_plugin_params: type: array description: >- Parameters the compat plugin dropped because the model does not support them items: type: string dropped_unsupported_tools: type: array description: Tool types removed because the target model does not support them items: type: string provider_response_headers: type: object description: >- The provider's HTTP response headers, without transport-level headers additionalProperties: type: string passthrough_path: type: string description: Provider path of a passthrough request ErrorField: type: object properties: type: type: string code: type: string message: type: string param: type: string event_id: type: string BifrostErrorExtraFields: type: object properties: routing_info: $ref: '#/components/schemas/RoutingInfo' provider: $ref: '#/components/schemas/ModelProvider' deprecated: true description: 'Deprecated: use routing_info.provider.' original_model_requested: type: string deprecated: true description: >- Deprecated: use routing_info.primary_model when routing_info.is_fallback is true, otherwise routing_info.model. For the model the caller asked for, read routing_info.requested_model. resolved_model_used: type: string deprecated: true description: >- Deprecated: use routing_info.resolved_key_alias.model_id when an alias matched, otherwise routing_info.model. request_type: type: string mcp_request_type: type: string a2a_request_type: type: string description: Agent Gateway operation, for A2A errors a2a_agent_name: type: string error_type: type: string description: >- Normalized, low-cardinality classification of why the request failed, declared by whichever component refused it. Prefixed by fault domain (caller_, policy_, provider_, bifrost_). Absent on failures that were not classified at source. retry_after_ms: type: integer format: int64 minimum: 1000 maximum: 300000 description: >- The provider's hint for how long to wait before retrying, in milliseconds, read from its retry-after-ms or Retry-After header or its google.rpc.RetryInfo error detail, and clamped to between 1000 and 300000. Absent when the provider gave no explicit hint. raw_request: description: Raw provider request, when raw request capture is enabled raw_response: description: Raw provider response, when raw response capture is enabled converted_request_type: type: string dropped_compat_plugin_params: type: array items: type: string latency: type: integer format: int64 description: Latency in milliseconds key_statuses: type: array description: >- Per-key results, for requests that ran on every key such as listing models items: type: object properties: key_id: type: string status: type: string enum: - success - list_models_failed provider: $ref: '#/components/schemas/ModelProvider' error: $ref: '#/components/schemas/BifrostError' mcp_auth_required: type: object description: >- Set when a per-user MCP tool needs the caller to complete an inline auth flow properties: kind: type: string mcp_client_id: type: string mcp_client_name: type: string message: type: string authorize_url: type: string session_id: type: string subject_token_missing: type: boolean description: >- token_exchange only. True when there was nothing to exchange, including when the caller's Bifrost sign-in has expired; message says which exchange_error: type: string description: >- token_exchange only. The identity provider's error code and description when it refused the exchange billed_usage: $ref: '#/components/schemas/BifrostLLMUsage' description: >- Usage the provider reported and billed even though the request failed DecisionInputMessage: type: object required: - role - content properties: type: type: string enum: - message description: Optional item type, as OpenAI's SDKs send it role: type: string example: user content: description: Text, or an array of parts oneOf: - type: string - type: array items: $ref: '#/components/schemas/DecisionInputPart' phase: type: string enum: - commentary - final_answer description: An assistant message's phase DecisionFunctionCall: type: object required: - type - call_id - name - arguments properties: type: type: string enum: - function_call call_id: type: string description: The call's ID, which its output names name: type: string arguments: type: string description: The call's arguments, as JSON text namespace: type: string id: type: string status: type: string enum: - in_progress - completed - incomplete DecisionFunctionCallOutput: type: object required: - type - call_id - output properties: type: type: string enum: - function_call_output call_id: type: string description: The call this output answers output: $ref: '#/components/schemas/DecisionToolOutput' id: type: string status: type: string enum: - in_progress - completed - incomplete DecisionCustomToolCall: type: object required: - type - call_id - name - input properties: type: type: string enum: - custom_tool_call call_id: type: string description: The call's ID, which its output names name: type: string input: type: string description: The call's freeform input namespace: type: string id: type: string status: type: string enum: - in_progress - completed - incomplete DecisionCustomToolCallOutput: type: object required: - type - call_id - output properties: type: type: string enum: - custom_tool_call_output call_id: type: string description: The call this output answers output: $ref: '#/components/schemas/DecisionToolOutput' id: type: string status: type: string enum: - in_progress - completed - incomplete DecisionInstructionPart: description: One part of a question's instructions oneOf: - type: object required: - type - text properties: type: type: string enum: - input_text text: type: string - type: object required: - type - image_url properties: type: type: string enum: - input_image image_url: type: string description: A base64 data URL or an HTTP(S) URL detail: type: string enum: - low - high - auto - original - $ref: '#/components/schemas/DecisionInputJSON' RoutingInfo: type: object description: >- Which provider, model and key handled the attempt that produced this response or error, and what the caller asked for. Also sent as x-bifrost-routing-info-* response headers. properties: provider: $ref: '#/components/schemas/ModelProvider' description: Provider that ran the attempt model: type: string description: Model sent to that provider key: type: string description: Name of the stored key used, never its ID or secret resolved_key_alias: $ref: '#/components/schemas/ResolvedKeyAlias' is_fallback: type: boolean description: >- True when a fallback served the request or returned the error. Omitted when false. primary_provider: $ref: '#/components/schemas/ModelProvider' description: The primary attempt's provider. Set only when a fallback ran. primary_model: type: string description: The primary attempt's model. Set only when a fallback ran. requested_provider: $ref: '#/components/schemas/ModelProvider' description: >- Provider the caller asked for, before routing rules, load balancing or session affinity changed it. Omitted when the caller sent a bare model. requested_model: type: string description: Model the caller asked for, before anything changed it server_side_fallback_model: type: string description: Model that answered when Anthropic switched models inside one call ModelProvider: type: string description: AI model provider identifier enum: - anthropic - azure - bedrock - bedrock_mantle - sagemaker - cerebras - cohere - deepseek - gemini - groq - mistral - ollama - opencode-go - opencode-zen - openai - parasail - perplexity - sgl - vertex - openrouter - elevenlabs - huggingface - nebius - xai - replicate - vllm - runway - runware - fireworks - sarvam - wafer - databricks - cloudflare - typesafe - minimax - byteplus BifrostCacheDebug: type: object properties: cache_hit: type: boolean cache_id: type: string hit_type: type: string requested_provider: type: string requested_model: type: string provider_used: type: string model_used: type: string input_tokens: type: integer threshold: type: number similarity: type: number cache_hit_latency: type: integer format: int64 description: Milliseconds spent serving a cache hit. Set only on a hit. BifrostGuardrailMetadata: type: object description: Guardrail judge calls and matches recorded for a request. properties: judge_calls: type: array description: >- Billable guardrail judge invocations (LLM-backed guardrail providers) items: $ref: '#/components/schemas/BifrostGuardrailJudgeCall' detections: type: array description: >- Guardrail matches. Enforced rules record detect-only matches; output-monitor rules record every match they see, with the action enforcement would have taken. items: $ref: '#/components/schemas/BifrostGuardrailDetection' BifrostLLMUsage: type: object description: Token usage information properties: prompt_tokens: type: integer description: > Total input tokens including any prompt-cache tokens (read + write). Subtract prompt_tokens_details.cached_read_tokens and prompt_tokens_details.cached_write_tokens to get the non-cached portion. prompt_tokens_details: $ref: '#/components/schemas/ChatPromptTokensDetails' completion_tokens: type: integer description: Number of output/completion tokens generated. completion_tokens_details: $ref: '#/components/schemas/ChatCompletionTokensDetails' total_tokens: type: integer tool_usage: type: object description: | Counts of provider-hosted (server-side) tool calls billed per call. properties: web_search: type: object properties: num_requests: type: integer description: Number of billable web search calls. cost: $ref: '#/components/schemas/BifrostCost' DecisionInputPart: description: >- One part of a message's content. User messages take text, JSON, and image parts; assistant messages take output text and refusals. oneOf: - $ref: '#/components/schemas/DecisionInputText' - $ref: '#/components/schemas/DecisionInputJSON' - $ref: '#/components/schemas/DecisionInputImage' - $ref: '#/components/schemas/DecisionInputImageAsset' - $ref: '#/components/schemas/DecisionOutputText' - $ref: '#/components/schemas/DecisionRefusal' DecisionToolOutput: description: A tool's output, as text or as text and image parts oneOf: - type: string - type: array items: oneOf: - $ref: '#/components/schemas/DecisionInputText' - $ref: '#/components/schemas/DecisionInputImage' DecisionInputJSON: type: object description: Structured evidence, read as JSON required: - type - data properties: type: type: string enum: - input_json data: description: A JSON object, or an array oneOf: - type: object additionalProperties: true - type: array items: {} ResolvedKeyAlias: type: object description: Set only when the requested model matched a model alias on the key required: - model_id properties: model_id: type: string description: Model ID sent to the provider model_name: type: string description: Canonical model name, used for pricing and logs model_family: type: string description: Resolved model family BifrostGuardrailJudgeCall: type: object properties: phase: type: string rule_id: type: integer rule_name: type: string guardrail_name: type: string guardrail_provider: type: string action: type: string reason: type: string judge_provider: type: string judge_model: type: string judge_request_type: type: string prompt_tokens: type: integer prompt_tokens_details: type: object additionalProperties: true completion_tokens: type: integer completion_tokens_details: type: object additionalProperties: true total_tokens: type: integer BifrostGuardrailDetection: type: object properties: phase: type: string description: Phase the match was found in (input, output, or mcp-) rule_id: type: integer rule_name: type: string guardrail_name: type: string description: Guardrail profile that matched guardrail_provider: type: string description: >- Guardrail provider type of the profile (for example regex or bedrock) reason: type: string assessments: type: array description: Sanitized policy categories; never the matched text items: type: string action: type: string enum: - detect - block - redact - transform description: >- What the guardrail did, or would have done when the rule only monitors enforced: type: boolean description: >- False when an output-monitor rule recorded the match without acting on it ChatPromptTokensDetails: type: object properties: text_tokens: type: integer audio_tokens: type: integer image_tokens: type: integer cached_read_tokens: type: integer description: > Tokens served from the prompt cache (cache hit). These tokens are already included in prompt_tokens and are billed at the reduced cache-read rate. Populated for all providers that support prompt caching (Anthropic, Bedrock, OpenAI, Gemini, xAI, etc.). cached_write_tokens: type: integer description: > Tokens written to the prompt cache on this request (cache creation / write). These tokens are already included in prompt_tokens and are billed at the cache-creation rate. Populated for providers that separately report cache write tokens (Anthropic, Bedrock). ChatCompletionTokensDetails: type: object properties: text_tokens: type: integer accepted_prediction_tokens: type: integer audio_tokens: type: integer citation_tokens: type: integer num_search_queries: type: integer deprecated: true description: >- Deprecated, use tool_usage.web_search.num_requests. Will be removed in 3.0.0. reasoning_tokens: type: integer image_tokens: type: integer rejected_prediction_tokens: type: integer BifrostCost: type: object description: Cost breakdown for the request properties: input_tokens_cost: type: number output_tokens_cost: type: number reasoning_tokens_cost: type: number description: Cost for reasoning/thinking tokens (reasoning models) citation_tokens_cost: type: number description: Cost for citation tokens search_queries_cost: type: number description: Cost for web search queries request_cost: type: number total_cost: type: number DecisionInputText: type: object required: - type - text properties: type: type: string enum: - input_text text: description: >- A string, or structured data that Typesafe-based models read as is and OpenAI receives as JSON text oneOf: - type: string - type: object additionalProperties: true - type: array items: {} prompt_cache_breakpoint: $ref: '#/components/schemas/DecisionPromptCacheBreakpoint' DecisionInputImage: type: object description: An inline image, read as evidence required: - type - image_url properties: type: type: string enum: - input_image image_url: type: string description: A base64 data URL or an HTTP(S) URL detail: type: string enum: - low - high - auto - original description: >- Image detail level, forwarded to models that read images (OpenAI's defaults to `auto`) prompt_cache_breakpoint: $ref: '#/components/schemas/DecisionPromptCacheBreakpoint' DecisionInputImageAsset: type: object description: >- Typesafe only. An image sent by reference, as one of Typesafe's `assets`, placed wherever its ID is written in the input, instructions, or criteria instead of as evidence itself. Other providers reject it. required: - type - id anyOf: - required: - data - required: - name properties: type: type: string enum: - input_image id: type: string description: The reference written where the image belongs data: type: string format: byte description: >- The image file, base64-encoded, without a `data:` prefix; its format is read from the bytes name: type: string description: >- In a multipart request, the name of the file part that holds the image. A JSON request cannot use it. DecisionOutputText: type: object description: An assistant message's text required: - type - text properties: type: type: string enum: - output_text text: type: string DecisionRefusal: type: object description: An assistant message's refusal required: - type - refusal properties: type: type: string enum: - refusal refusal: type: string DecisionPromptCacheBreakpoint: type: object description: >- Marks the end of a reusable prompt prefix on a text or image part of the input. Forwarded to OpenAI. properties: mode: type: string enum: - explicit securitySchemes: BearerAuth: type: http scheme: bearer description: > Bearer token authentication. Use your provider API key or Bifrost authentication token. Virtual keys (prefixed with `sk-bf-`) can also be passed here. BasicAuth: type: http scheme: basic description: > Basic authentication using the Bifrost admin username and password (`auth_config.admin_username` / `auth_config.admin_password`). Accepted on management APIs (`/api/*`, `/metrics`, `/ws`) only - the inference middleware never validates Basic credentials. ApiKeyAuth: type: apiKey in: header name: x-api-key description: | API key authentication via the `x-api-key` header. Virtual keys (prefixed with `sk-bf-`) can also be passed here. VirtualKeyAuth: type: apiKey in: header name: x-bf-vk description: > Bifrost Virtual Key for governance, routing, and access control. Supported on all inference endpoints (`/v1/*`, `/openai/*`, `/anthropic/*`, `/bedrock/*`, `/cohere/*`, `/genai/*`, `/langchain/*`, `/litellm/*`, `/pydanticai/*`, `/mcp`), not on management APIs (`/api/*`) - with the single exception of the self-service `GET /api/governance/virtual-keys/quota`, where the virtual key is the credential. Example: `sk-bf-*` prefixed keys. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.