AI Integration

RustAPI offers native support for building AI-friendly APIs using the rustapi-toon crate. This allows you to serve optimized content for Large Language Models (LLMs) while maintaining standard JSON responses for traditional clients.

The Problem: Token Costs

LLMs like GPT-4, Claude, and Gemini charge by the token. Standard JSON is verbose, containing many structural characters (", :, {, }) that count towards this limit.

JSON (55 tokens):

[
  {"id": 1, "role": "admin", "active": true},
  {"id": 2, "role": "user",  "active": true}
]

TOON (32 tokens):

users[2]{id,role,active}:
  1,admin,true
  2,user,true

The Solution: Content Negotiation

RustAPI uses the Accept header to decide which format to return.

Accept: application/json -> Returns JSON.
Accept: application/toon -> Returns TOON.
Accept: application/llm (custom) -> Returns TOON.

This is handled automatically by the LlmResponse<T> type.

Dependencies

[dependencies]
rustapi-rs = { version = "0.1.550", features = ["toon"] }
serde = { version = "1.0", features = ["derive"] }

Implementation

use rustapi_rs::prelude::*;
use rustapi_toon::LlmResponse; // Handles negotiation
use serde::Serialize;

#[derive(Serialize)]
struct User {
    id: u32,
    username: String,
    role: String,
}

// Simple handler returning a list of users
#[rustapi_rs::get("/users")]
async fn get_users() -> LlmResponse<Vec<User>> {
    let users = vec![
        User { id: 1, username: "Alice".into(), role: "admin".into() },
        User { id: 2, username: "Bob".into(), role: "editor".into() },
    ];

    // LlmResponse automatically serializes to JSON or TOON
    LlmResponse(users)
}

#[tokio::main]
async fn main() {
    let app = RustApi::new().route("/users", get(get_users));

    println!("Server running on http://127.0.0.1:3000");
    app.run("127.0.0.1:3000").await.unwrap();
}

Testing

Standard Browser / Client:

curl http://localhost:3000/users
# Returns: [{"id":1,"username":"Alice",...}]

AI Agent / LLM:

curl -H "Accept: application/toon" http://localhost:3000/users
# Returns:
# users[2]{id,username,role}:
#   1,Alice,admin
#   2,Bob,editor

Providing Context to AI

When building an MCP (Model Context Protocol) server or simply feeding data to an LLM, use the TOON format to maximize the context window. See the MCP Integration (Agent Tools) recipe for exposing your endpoints as callable tools for Claude, Cursor and other agents.

// Example: Generating a prompt with TOON data
let data = get_system_status().await;
let toon_string = rustapi_toon::to_string(&data).unwrap();

let prompt = format!(
    "Analyze the following system status and report anomalies:\n\n{}",
    toon_string
);

// Send `prompt` to OpenAI API...

Keyboard shortcuts