Evaluate a decision

Asks a list of questions about some input and returns one typed answer per question, in order, with per-answer confidence and probabilities. Served by any decision model: Typesafe’s jev models (and Laya, Nimble, and Clef), or any chat model through structured output.

The map form (a state plus questions keyed by name) is deprecated but still served; a map-form request is answered in the map form.

Images sent by reference to Typesafe can also come as a multipart request: the JSON in a part named request, and each image in its own file part, named by the image part’s name.

POST /v1/decisions

Error

A valid request URL is required to generate request examples

200

{
  "model": "<string>",
  "answers": [
    {
      "type": "choice",
      "name": "<string>",
      "probability": 123,
      "choice": "<string>",
      "score": 123,
      "box": {
        "top": 123,
        "bottom": 123,
        "left": 123,
        "right": 123
      },
      "probabilities": [
        {
          "value": "<string>",
          "label": "<string>",
          "probability": 123
        }
      ],
      "confidence": 123,
      "legend": {},
      "answer_confidence": 123,
      "action": {},
      "abstention": "passed",
      "abstention_threshold": 123,
      "low_confidence": true
    }
  ],
  "id": "<string>",
  "usage": {
    "prompt_tokens": 123,
    "prompt_tokens_details": {
      "text_tokens": 123,
      "audio_tokens": 123,
      "image_tokens": 123,
      "cached_read_tokens": 123,
      "cached_write_tokens": 123
    },
    "completion_tokens": 123,
    "completion_tokens_details": {
      "text_tokens": 123,
      "accepted_prediction_tokens": 123,
      "audio_tokens": 123,
      "citation_tokens": 123,
      "num_search_queries": 123,
      "reasoning_tokens": 123,
      "image_tokens": 123,
      "rejected_prediction_tokens": 123
    },
    "total_tokens": 123,
    "tool_usage": {
      "web_search": {
        "num_requests": 123
      }
    },
    "cost": {
      "input_tokens_cost": 123,
      "output_tokens_cost": 123,
      "reasoning_tokens_cost": 123,
      "citation_tokens_cost": 123,
      "search_queries_cost": 123,
      "request_cost": 123,
      "total_cost": 123
    }
  },
  "extra_fields": {
    "request_type": "<string>",
    "pricing_request_type": "<string>",
    "routing_info": {
      "provider": "anthropic",
      "model": "<string>",
      "key": "<string>",
      "resolved_key_alias": {
        "model_id": "<string>",
        "model_name": "<string>",
        "model_family": "<string>"
      },
      "is_fallback": true,
      "primary_provider": "anthropic",
      "primary_model": "<string>",
      "requested_provider": "anthropic",
      "requested_model": "<string>",
      "server_side_fallback_model": "<string>"
    },
    "provider": "anthropic",
    "original_model_requested": "<string>",
    "resolved_model_used": "<string>",
    "latency": 123,
    "upstream_latency": 123,
    "chunk_index": 123,
    "raw_request": "<unknown>",
    "raw_response": "<unknown>",
    "cache_debug": {
      "cache_hit": true,
      "cache_id": "<string>",
      "hit_type": "<string>",
      "requested_provider": "<string>",
      "requested_model": "<string>",
      "provider_used": "<string>",
      "model_used": "<string>",
      "input_tokens": 123,
      "threshold": 123,
      "similarity": 123,
      "cache_hit_latency": 123
    },
    "routing_metadata": {
      "calls": [
        {
          "request_type": "<string>",
          "provider_used": "<string>",
          "model_used": "<string>",
          "input_tokens": 123,
          "output_tokens": 123,
          "count_toward_budgets": true
        }
      ]
    },
    "guardrail_debug": {
      "judge_calls": [
        {
          "phase": "<string>",
          "rule_id": 123,
          "rule_name": "<string>",
          "guardrail_name": "<string>",
          "guardrail_provider": "<string>",
          "action": "<string>",
          "reason": "<string>",
          "judge_provider": "<string>",
          "judge_model": "<string>",
          "judge_request_type": "<string>",
          "prompt_tokens": 123,
          "prompt_tokens_details": {},
          "completion_tokens": 123,
          "completion_tokens_details": {},
          "total_tokens": 123
        }
      ],
      "detections": [
        {
          "phase": "<string>",
          "rule_id": 123,
          "rule_name": "<string>",
          "guardrail_name": "<string>",
          "guardrail_provider": "<string>",
          "reason": "<string>",
          "assessments": [
            "<string>"
          ],
          "action": "detect",
          "enforced": true
        }
      ]
    },
    "parse_errors": [
      {
        "code": "<string>",
        "message": "<string>",
        "param": "<string>",
        "line": 123
      }
    ],
    "converted_request_type": "<string>",
    "dropped_compat_plugin_params": [
      "<string>"
    ],
    "dropped_unsupported_tools": [
      "<string>"
    ],
    "provider_response_headers": {},
    "passthrough_path": "<string>"
  },
  "routing": {}
}

Authorizations

BearerAuthVirtualKeyAuthApiKeyAuth

Authorization · string · header · required

Bearer token authentication. Use your provider API key or Bifrost authentication token. Virtual keys (prefixed with sk-bf-) can also be passed here.

Body

application/jsonmultipart/form-data

Option 1

Option 2

model · string · required

Model in provider/model format

Example:

"typesafe/jev-1.13.0"

input · string object[] object · required

The evidence. A string of text; a conversation, as an array of messages (with text, JSON, and image parts; images are read only by providers that accept them), function and custom tool calls, and their outputs; an object of structured data (passed to Typesafe-based models as their state); or null.

questions · object[] · required

Questions in the order asked. Answers follow the same order.

Minimum array length: 1

Show child attributes

safety_identifier · string

Stable end-user identifier, forwarded to providers that accept one (OpenAI)

prompt_cache_key · string

Partitions prompt-cache reuse by a caller-chosen key. Forwarded to OpenAI; other providers drop it.

prompt_cache_options · object

Prompt-caching options, forwarded to OpenAI; other providers drop them.

Show child attributes

fallbacks · string[]

Fallback models in provider/model format

Response

200

application/json

Successful response

Option 1

Option 2

model · string · required

Model that performed the decision

answers · object[] · required

One answer per question, in question order

Show child attributes

id · string

Unique identifier for the decision response

usage · object

Token usage information

Show child attributes

extra_fields · object

Additional fields included in responses

Show child attributes

routing · object

Laya only. The checkpoint that answered and why, passed through untouched