Attractor

Attractor is StrongDM's implementation of a non-interactive coding agent, designed to operate end-to-end once work is fully specified. The GitHub repository published by StrongDM contains natural-language specs (NLSpecs) that coding agents can use to build your own version of Attractor for a software factory. The specs are released under the Apache License 2.0.

What It Is

Attractor composes models, prompts, and tools into a graph-structured pipeline, which StrongDM describes as a generative SDLC. The repository ships three specs: the Attractor Specification, a Coding Agent Loop Specification, and a Unified LLM Client Specification. An NLSpec is a human-readable spec intended to be directly usable by coding agents to implement and validate behavior.

How the Graph Works

Each node in the graph corresponds to a phase of work and is governed by a core prompt, such as implementing functionality, identifying a bottleneck, optimizing for performance, or validating behavioral correctness. Edges between nodes are written in natural language and evaluated by the LLM, for example proceeding once a bottleneck is identified, or taking a branch if standards are met. Execution traverses the graph until convergence or termination conditions are met. StrongDM lists these key properties: deterministic given the same inputs, observable at every node transition, resumable from any checkpoint, and composable with other graphs.

Building Your Own

The README suggests giving a modern coding agent (such as Claude Code, Codex, OpenCode, Amp, or Cursor) the prompt to implement Attractor as described by the repository. It recommends controlling your own agentic loop and unified LLM SDK for a strong foundation, though this is not required.

Community Implementations

The product page states the spec is open and lists community implementations in languages including Rust, Go, Python, Java, F#, TypeScript, Scala, Ruby, C, and C#. Examples include Fabro, Kilroy, and Forge.