Quickstart

Install

Add the application-facing crate:

cargo add everruns

The default features include typed tool macros and the offline simulator. They do not select a network provider.

Run one turn

use everruns::{Agent, Model};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let agent = Agent::builder()
        .instructions("You are a concise assistant.")
        .model(Model::simulated("Everruns is ready."))
        .build()?;

    let session = agent.session();
    let turn = session.send_and_wait("Are you ready?").await?;
    println!("{}", turn.response);
    Ok(())
}

Model::simulated follows the same model/provider path as a live provider, but returns the configured response deterministically. This makes the smallest Framework program useful in tests, examples, and disconnected development. send_and_wait is the request/response convenience; use send when the application needs to stream output or add steering input while a turn runs.

Use OpenAI

Enable the provider feature and set the credential in the host environment:

cargo add everruns --features openai
export OPENAI_API_KEY=sk-...

Replace the simulated model:

use everruns::{Agent, OpenAI};

let agent = Agent::builder()
    .instructions("You are a concise assistant.")
    .provider(OpenAI::from_env()?)
    .model("gpt-5.6-terra")
    .build()?;
# Ok::<(), Box<dyn std::error::Error>>(())

The model id stays credential-free. OpenAI::from_env configures the separate provider at the application boundary and redacts the key from debug output.

Continue with Agents, Tools and macros, and Sessions.