Free
$0
For a small fleet.
- Up to 3 agents
- Every provider, local models included
- Memory, skills and delegation
- Permissions and sandbox
- 2 connectors
Native agent harness for macOS
Each agent has its own computer, long-term memory and routines. They work in the background, hand tasks to each other and check in with you — on your Mac or on Telegram.
Every agent picks its own model
or any Anthropic- or OpenAI-compatible API
Start from a template — chief of staff, researcher, coder, writer, lookout — or write your own. Each agent gets a name, a role and a planet.
A persistent workspace with a real zsh terminal and a file editor, with an optional sandbox. Add Notion, GitHub or Linear in one click.
Routines run on schedule, agents delegate to whoever fits best, and every command waits for your OK until you choose to trust it.
Claude, Ollama, Gemini, OpenAI, Kimi, OpenRouter, LM Studio… Each agent picks its provider, and one fleet can mix them.
Agents pass tasks to each other. The delegation map shows your fleet as a solar system: who helps whom, live.
An agent can split a job across up to five parallel copies of itself — its moons.
Durable notes for each agent, plus a shared fleet memory with what everyone should know about you.
Scheduled instructions every few minutes or at a set time, even with the window closed.
Talk to the fleet from your phone through your own bot: replies, approvals with buttons, and files.
MCP servers for the fleet — Notion, Linear, GitHub, Sentry, Playwright and more — with OAuth sign-in, limited to the agents that need them.
With agent-desktop, agents read and use your Mac apps through accessibility, without taking over your mouse.
Folders with a SKILL.md, the same format as Claude Code and Codex. Agents can save what they learn, if you approve.
In Ask mode every command and edit waits for ⌘↩. “Always allow” saves precise rules, never global wildcards.
Allow or deny from a notification or the menu bar, without opening the app.
Pure SwiftUI with its own agent loop. No Electron, no Node, no Python. Keys stay in your Keychain.
Your agents, conversations, memory and files live on your Mac. No sign-up and no servers of ours in between: requests go straight from your Mac to the model you choose — or nowhere at all, with Ollama.
Every install starts with 14 days of Pro, no card needed. Then you keep the Free plan — or go Pro.
$0
For a small fleet.
$79/year
$9/month
For a fleet that works while you don’t.
Everything in Free, plus:
Yours forever: if you stop renewing, the last version from your year stays Pro.
The monthly plan doesn’t include the forever license.
Coming soon
Shared templates, skills and connectors, with central billing.
Need several licenses today? We offer volume licenses.
Moonsmith doesn’t include model usage: run Ollama for free, or use your own API key or plan.
A Mac with macOS 14 or later, and a model. Ollama runs models locally for free; or bring a key for Claude, Gemini, OpenAI or any compatible API.
Moonsmith switches to Free. Nothing is deleted — agents, memory and conversations stay. Pro features pause until you upgrade.
The yearly plan includes every update released during your year. If you don’t renew, the last version from that year keeps Pro, forever; renew any time to jump to the latest. One caveat: AI providers change their APIs, so a very old version may lose some providers over time.
A license is for one person, on all of their own Macs.
No. There is no Moonsmith account or server between you and your model, and the app sends no analytics. What leaves your Mac goes to the provider you picked, or to Telegram if you turn it on.
App Store apps must run inside Apple’s App Sandbox, which rules out a real terminal for your agents. Moonsmith is signed with a Developer ID and notarized by Apple instead, so it opens without warnings.