Rust Confessional: a durable AI agent demo

Rust Confessional is a stage-friendly demo built on the official Temporal Rust SDK (Public Preview). Audience members submit a programming confession. An AI agent (Ferris) plans a response, consults an approved Rust remedy catalog, and returns a dry but useful judgment with a 1 to 5 "Ferris Level" severity. A live dashboard makes every state transition visible.

The demo lands when you kill the Temporal Worker at a controlled checkpoint, release the pending replies while it is offline, then restart it and watch Temporal replay history and resume the agent exactly where it stopped.

The thesis: Rust makes the agent loop explicit. Temporal makes its progress survive.

What it shows#

One Temporal Workflow per confession, plus a toggle to an aggregate single-Workflow mode that makes durable state vivid on stage Typed Rust activities for planning, remedy lookup, composition, and delivery A bounded autonomous decide/act loop (plan, look up a remedy, self-critique, compose, finish) alongside a simpler linear pipeline Three failure beats (transient model rate-limit, network partition, full Worker crash), each contrasting Temporal's durable retries with Rust's typed, compiler-checked error handling

Running it#

Docker-first, so no local Rust, Cargo, or Temporal install needed Deterministic fixture mode for offline stage use, with an optional OpenAI backend and optional Twilio inbound SMS make up to start; dashboard at localhost:3000, Temporal Web at localhost:8233

Built on#

Temporal Rust SDK (Public Preview): github.com/temporalio/sdk-rust Rust SDK docs: docs.temporal.io/develop/rust

Authors#

Melanie Warrick, with Melissa Herrera, Spencer Judge, Chris Olszewski, Tom Wheeler, and Shy Ruparel.