auto-agents
auto-agents is an open, local-first workflow runtime for repository agents. It turns multiple agent CLIs into repeatable software-factory workflows that can run from desktop, web, terminal, and automation contexts.
What auto-agents is for
Use auto-agents when you want to:
Standardize multi-step software workflows instead of running ad-hoc prompts.
Reuse the same workflows across multiple AI agent providers.
Keep workflow assets and skill files versioned in your repository.
Run in either interactive mode (draft + feedback + finalize) or unattended mode.
Supported interaction surfaces
auto-agents is designed to operate as:
A desktop app (
auto-agentsorauto-agents gui).A local web server UI (
auto-agents serve).A command-line workflow runner (
auto-agents <command>).A GitHub-oriented automation runtime for issue and PR operations.
Core concepts
Agent adapters
auto-agents uses provider-specific adapter crates to map one shared command model to different external CLIs. --agent selects which adapter to use per run.
Workflow presets
Workflows are grouped into presets and loaded from assets/workflows/<preset>/. You can list presets with auto-agents presets and override the active preset for a single invocation with --preset <name>.
Two-phase workflow pattern
Most thinking workflows follow a draft/finalize pattern:
Generate a draft artifact.
Collect human feedback.
Produce a finalized artifact.
With --auto, auto-agents synthesizes stand-in feedback for supported two-phase flows so they proceed end-to-end without manual pauses.
Start here
For installation and first-run setup, see Getting started.
For build and release command details, see Installation and build.
For command behavior and flags, see CLI reference.
For architecture and internal module map, see Architecture.
Copyright and license
Copyright © 2026 Geoff Seemueller.
Documentation and source materials in this project are provided under the repository license.