OpenRouter

OpenRouter is a unified gateway in front of hundreds of models; the model slug selects which one, and it is vendor-prefixed (e.g. anthropic/claude-sonnet-4, openai/gpt-4o). It exposes two OpenAI-compatible wires. By default chat-openrouter builds on the Responses wire; you can opt into Completions with .with_completions().

Install

chat-rs = { version = "0.5.1", features = ["openrouter"] }

OpenRouterBuilder::new() reads OPENROUTER_API_KEY at build time and defaults to the base URL https://openrouter.ai/api/v1.

Minimal usage

use chat_rs::{ChatBuilder, ChatOutcome, openrouter::OpenRouterBuilder, parts, types::messages};

// OPENROUTER_API_KEY is read automatically. Default wire: Responses.
let client = OpenRouterBuilder::new()
    .with_model("anthropic/claude-sonnet-4")
    .build();
let mut chat = ChatBuilder::new().with_model(client).build();

let mut messages = messages::from_user(parts!["Say hello in one sentence."]);
if let ChatOutcome::Complete(res) = chat.complete(&mut messages).await.map_err(|e| e.err)? {
    println!("{:#?}", res.content);
}

Picking a wire

The default Responses wire is stateless here, so the full conversation is sent each turn. To target Chat Completions instead, add .with_completions() before .build():

use chat_rs::openrouter::OpenRouterBuilder;

let client = OpenRouterBuilder::new()
    .with_completions()
    .with_model("openai/gpt-4o")
    .build();

Configuration

with_model(impl Into<String>) takes a vendor-prefixed slug. A few more helpers:

• with_api_key(impl Into<String>) overrides the env var when you want to pass the key explicitly.
• with_base_url(impl Into<String>) points at a different host.
• with_reasoning_effort(impl Into<String>) sets "low", "medium", or "high" for reasoning-capable models. This is Responses-wire only; the Completions wire does not carry the field.

Capabilities

OpenRouter supports completion, streaming, structured output, and tools. Streaming needs the stream feature and runs as SSE over HTTP on either wire.