Build a squad of AI agents. Give each one a clear job, then watch every message move.
How Austy works
No overlap, by design.
Every agent has a role, exclusively owned responsibilities and an out-of-scope list. Saving an agent is blocked if a responsibility duplicates another member's.
Tasks have one owner. Agents delegate out-of-lane work to the owner, and auto-routing matches tasks to responsibilities.
Squad templates
Marketing, Product and engineering, Business operations, Online auctions ops, or start blank.
One engine per agent.
Put your frontend agent on Claude Code and your backend agent on Codex. Switch an agent's engine from its card.
| Engine | How it runs |
|---|---|
| Claude API | Your Anthropic key in Settings |
| OpenAI-compatible API | OpenAI, OpenRouter, Groq, Together, LM Studio, Ollama |
| Claude Code CLI | Runs in the agent's working folder with your Claude login |
| Codex CLI | Runs with your ChatGPT or OpenAI login |
| Custom CLI | Any command, acting through simple directive lines |
A shared Brain.
Every project has a Brain: markdown pages with a compiled truth that improves over time, and an append-only timeline of evidence.
Agents search before they work and write back what they learn. Point it at an Obsidian vault if you already keep one.
- Connections to email, Drive, Jira, the browser and automations
- Approvals for write actions, as macOS notifications
- Skills installed per agent
- Quick Capture: press ⌥⌘Space, highlight the screen, send it to an agent
- Keys and connections encrypted with the macOS keychain
A team, not a chatbot.
Squads
A team of agents with a mission, a lead, tasks, comms and shared context.
Comms you can read
Agents message each other, the squad or you. Every message is visible, with live pulses in the network view.
Autopilot and automations
Agents with pending work wake up until a turn budget is spent. Standing orders run on a schedule.
Put a squad to work.
Austy is in early access for Apple Silicon and Intel Macs.