SDK quickstart

This path embeds the complete MicroClaw Agent Engine and keeps configuration in your application. It requires Rust 1.93 or newer and a Tokio runtime.

1. Add dependencies

Use the published SDK from crates.io:

Cargo.toml

[dependencies]
microclaw-sdk = { version = "0.6.1", features = ["full"] }
tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

2. Configure and run an Agent

src/main.rs

use microclaw_sdk::{FullRuntimeConfig, MicroClaw};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = FullRuntimeConfig::new(
        "openai",
        "gpt-5",
        std::env::var("OPENAI_API_KEY")?,
    );
    let microclaw = MicroClaw::configure(config)
        .workspace(std::env::current_dir()?)
        .caller_channel("my-rust-app")
        .max_concurrent_runs(4)
        .build()
        .await?;

    let agent = microclaw
        .agent("assistant")
        .system_prompt("Be concise and cite the files you inspect.")
        .build()?;

    let result = agent.run("Summarize this project").result().await?;
    println!("{}", result.final_text);
    microclaw.runtime().shutdown().await;
    Ok(())
}

Set OPENAI_API_KEY, then run the application with cargo run. The API key stays owned by the host application; it does not need to be written to a MicroClaw configuration file.

Integration checklist

A first integration is complete when the host application can:

• build one MicroClaw value and reuse it rather than rebuilding per request;
• create an Agent profile with an explicit name and tool policy;
• consume ordered events for live progress and use RunResult for completion;
• classify failures through RuntimeErrorCode instead of matching text;
• cancel an active run and shut the Runtime down cleanly.

Start with one local Agent. Add managed Skills only after the basic run works, and add a Worker only when the application needs queueing or isolation. This keeps the first integration small while preserving the same API as it grows.

3. Stream progress

Keep the RunHandle mutable and read events before taking its terminal result:

let mut run = agent.run("Review this repository");
while let Some(envelope) = run.next_event().await {
    println!("{}: {:?}", envelope.sequence, envelope.event);
}
let result = run.result().await?;

Events are ordered by sequence. Treat RunResult as the authoritative terminal outcome; events are the live projection for a UI, log, or protocol.

YAML configuration instead

Applications that already use MicroClaw configuration can load it directly:

let microclaw = MicroClaw::builder("microclaw.config.yaml")
    .caller_channel("my-rust-app")
    .build()
    .await?;

Use builder_from_environment() to follow MicroClaw's normal configuration discovery. Continue with SDK concepts before wiring events and controls into your UI.

Common failures

Symptom	What to check
Unavailable	Provider reachability, credentials, or Worker health; retry only when retryable is true.
InvalidRequest	Agent profile, selected Skill names, protocol version, and run identity.
A Skill is listed but unavailable	Read its catalog diagnostic; install its dependency or enable it before starting the run.
Events stop after a disconnect	Keep the same run ID and last sequence; remote Workers resume and suppress duplicates automatically.
Shutdown does not finish immediately	Stop accepting work, cancel or drain active runs, then await Runtime::shutdown().