Deployment

Deployment models for behest.

behest is a library crate, not a standalone binary. You can deploy it in several ways.

Library

The most common model. Add behest to your Cargo.toml, construct an AgentRuntime in your application, and run it in-process:

let runtime = AgentConfig::builder()
    .with_file("behest.toml")?
    .build_runtime()
    .await?;
runtime.run(req).await?;

No network, no serialisation, no overhead. The runtime is just a Rust struct.

Sidecar

Run the behest runtime in a sidecar process alongside your main application. The main app submits runs via the library API; the sidecar handles model calls, tool execution, and persistence.

Resource requirements

• Memory: ~20 MB baseline (runtime + default memory stores). Grows with session history and embedding vectors.
• CPU: Model calls are I/O-bound. Tool execution is CPU-bound but usually short.
• Network: Outbound to model providers (HTTPS) and optional stores (Redis, Postgres, etc.).

See also

• Performance — throughput and latency.
• Feature Flags — the full feature and other flags.