a-agmon/rs-graph-llm

What it solves

graph-flow is a stateful graph workflow framework designed for AI agents. It allows developers to build complex, interactive, and resumable workflows where AI agents can perform tasks, make decisions, and pause for human input, all while maintaining state across process restarts.

How it works

The framework uses a graph-based execution engine where nodes are defined as Task traits. Each task reads and writes to a shared, thread-safe Context that is serialized and persisted in a session. The flow of execution is controlled by NextAction returns from tasks, which determine if the graph should advance to the next node, jump to a specific task, wait for user input, or end the workflow.

It integrates natively with the Rig LLM library for chat history and agent runtime, and supports both in-memory and PostgreSQL storage backends for session persistence.

Who it’s for

Rust developers building AI agents and agentic workflows that require complex routing, state management, and human-in-the-loop interactions.

Highlights

• Stateful and Resumable: Sessions can be paused and resumed across restarts using persistent storage.
• Flexible Execution: Supports step-by-step (interactive) and fire-and-forget (continuous) execution models.
• Conditional Routing: Ability to branch the workflow at runtime based on data within the Context.
• Human-in-the-Loop: Built-in WaitForInput action to park workflows until external input is provided.
• Parallel Execution: FanOutTask allows running multiple child tasks concurrently and aggregating their results.
• Type-Safe: Leverages Rust's type system for performance and safety.