Ashlr AO Documentation

Everything you need to orchestrate AI coding agents from a single command center.

What is Ashlr AO?

Ashlr AO is a local-first agent orchestration platform. It gives a single developer (or team) the ability to spawn, monitor, and manage multiple AI coding agents across multiple repositories from one dashboard.

Supported backends include Claude Code, OpenAI Codex, Aider, and Goose. Each agent runs in an isolated tmux session with real-time terminal capture, status detection, and output parsing.

Current version: v1.6.1 — 22 Python modules, 1,926 tests, production-ready.

Quick Links

Core Concepts

Agents

An agent is a running instance of an AI coding tool (Claude Code, Codex, etc.) managed by Ashlr. Each agent gets its own tmux session, working directory, role, and task. Terminal output is captured every second, parsed for status (planning, working, waiting, error, idle), and broadcast to the dashboard in real time via WebSocket.

Projects

Projects map to git repositories on your local machine. When you spawn an agent, you assign it to a project. This enables per-project filtering, focus mode, branch tracking, and fleet templates. Projects auto-detect git remotes, branches, and metadata.

Workflows

Workflows define a DAG (directed acyclic graph) of agent tasks with dependency ordering. You specify agent specs with depends_on relationships, and Ashlr executes them in the correct order. Available on the Pro tier.

Backends

A backend is the CLI tool that powers an agent. Ashlr ships with support for four backends:

BackendCommandNotes
claude-codeclaudePlan mode, model selection, tool restriction, stream-json output
codexcodexOpenAI Codex CLI
aideraiderAI pair programming
goosegooseAutonomous coding agent by Block

Roles

Roles provide a visual label and color for agents in the dashboard. Nine built-in roles cover common development tasks:

Architecture Overview

Ashlr AO is a modular Python package (ashlr_ao) with 22 focused modules totaling approximately 15,500 lines of code. The dashboard is a single HTML file with inline CSS and JS — no build step required.

LayerTechnologyPurpose
ServerPython 3.11+ / aiohttpAsync HTTP + WebSocket, process management via psutil
FrontendVanilla JS (ES2022+)Zero dependencies, no build step
PersistenceSQLite via aiosqliteZero-config single-file database
Process mgmttmuxSession isolation, output capture
LicensingPyJWT + Ed25519Offline-first signed JWT, no phone-home
IntelligencexAI Grok (optional)Summaries, NLU parsing, fleet analysis
DesktopTauri v2Native macOS app, system tray, sidecar server

Licensing

Ashlr AO uses an open-core licensing model:

FeatureCommunity (Free)Pro (Paid)
Concurrent agentsUp to 5Up to 100
Users / seats1Up to 50
Core orchestrationFullFull
Intelligence (LLM)GatedIncluded
WorkflowsGatedIncluded
Fleet presetsGatedIncluded
Multi-user authGatedIncluded
Edit this page on GitHub