Reference

Gateway Provider Reference

OpenRouter and Requesty configuration, attribution headers, reasoning controls, usage telemetry, and shared OpenAI tool encoding.

OpenRouter and Requesty are OpenAI-compatible gateway providers. Both are built on the shared OpenAI-compatible client and add gateway-specific defaults for reasoning, usage, and telemetry.

Factories

import { openrouterModelClient } from '@inbrowser/model/providers/openrouter';
import { requestyModelClient } from '@inbrowser/model/providers/requesty';

Factory	Endpoint	Client id
`openrouterModelClient(config)`	`https://openrouter.ai/api/v1/chat/completions`	`openrouter:${config.model}`
`requestyModelClient(config)`	`https://router.requesty.ai/v1/chat/completions`	`requesty:${config.model}`

Both factories return a ModelClient.

Config

Both providers accept the shared cloud-provider fields:

Field	Description
`apiKey`	Bearer token sent to the gateway.
`model`	Gateway model id.
`reasoningEffort`	`off`, `low`, `medium`, or `high`, when the gateway supports reasoning controls.

They also accept optional app attribution:

const client = openrouterModelClient({
  apiKey,
  model: 'openai/gpt-4.1-mini',
  appAttribution: {
    referer: 'https://example.com',
    title: 'Example Builder',
  },
});

appAttribution.referer becomes HTTP-Referer. appAttribution.title becomes X-Title. No attribution headers are sent unless you pass them.

Reasoning controls

Gateway providers use OpenRouter-compatible reasoning fields:

`reasoningEffort`	Request fields
`off`	`reasoning: { enabled: false }`
`low`, `medium`, `high`	`reasoning: { effort, summary: 'auto' }` and `include_reasoning: true`

Reasoning stream deltas are surfaced as ModelEvent values with kind: 'thinking' when the gateway sends reasoning or reasoning_content delta fields.

Usage telemetry

Both gateway providers request included usage details and preserve provider telemetry in ModelUsage:

Usage field	Source
`promptTokens`	`usage.prompt_tokens`
`outputTokens`	`usage.completion_tokens`
`cachedTokens`	`usage.prompt_tokens_details.cached_tokens`
`reasoningTokens`	`usage.completion_tokens_details.reasoning_tokens`
`costUsd`	`usage.cost`

Fields are omitted when the provider does not report them.

Tool encoding

OpenRouter and Requesty share the same nested OpenAI tool encoding. A ToolSpec becomes:

{
  type: 'function',
  function: {
    name: tool.name,
    description: tool.description,
    parameters: tool.parameters,
  },
}

The shared encoding keeps gateway providers aligned when new tool metadata is added to the model contract.