Concepts
Parameters

Parameters reference

This page summarizes key request parameters supported by KushRouter across the two API surfaces you can call:

  • OpenAI-compatible Chat Completions: POST /api/openai/v1/chat/completions
  • OpenAI-compatible Responses: POST /api/openai/v1/responses
  • Anthropic-compatible Messages: POST /api/anthropic/v1/messages

Parameters must match the surface syntax you call. If a parameter is not supported by that surface, the server returns 400 with a descriptive error.

You can use OpenAI models on the Anthropic surface and Claude models on the OpenAI surfaces. The parameter rules still follow the surface you call. See cross‑provider compatibility.

Common

  • model (string, required)
  • messages (array)
  • stream (boolean)

OpenAI Chat Completions

  • Tools: tools, tool_choice, legacy function_call and functions
  • Generation controls: temperature, top_p, max_tokens, max_completion_tokens, stop
  • Penalties: frequency_penalty, presence_penalty, logit_bias
  • Sampling: n, seed, user
  • Reasoning/output: parallel_tool_calls, response_format, reasoning_effort
  • Service & metadata: service_tier, store, metadata
  • Streaming options: stream_options.include_usage
  • Retrieval/web: web_search_options

OpenAI Responses

  • Input: input or messages
  • Tools: tools, tool_choice, legacy functions, function_call
  • Generation: temperature, top_p, max_output_tokens
  • Reasoning/output: response_format, reasoning_effort, reasoning_summary
  • Sampling: n, seed, user, logit_bias, frequency_penalty, presence_penalty
  • Streaming: stream: true yields named Responses events
  • Follow‑ups: tool_outputs, previous_response_id
  • Additional: include, web_search_options, prompt_cache_retention, truncation, max_tool_calls, audio, modalities, prediction, strict, verbosity, conversation

See model-specific reasoning summary support in OpenAI Responses → Reasoning summaries.

Anthropic Messages

  • Content: messages (Anthropic blocks)
  • Generation: max_tokens (required), temperature, top_p, top_k
  • Stops: stop_sequences
  • Service & metadata: service_tier, metadata, cache_control
  • Reasoning: thinking, reasoning_effort
  • Tools: tools, tool_choice, parallel_tool_calls
  • Streaming: stream: true yields Claude‑style SSE events
  • Headers: anthropic-version

Validation rules

  • The surface determines accepted parameters.
  • Incompatible combinations are rejected with 400.
  • Payloads that exceed limits return 413.

See also

IndexRequest schemas