OpenCue
New in 0.3.4 · local VMs that keep their state, Composio Connect, provider logos

Your own AI agents, living on your desktop.

Create agents with their own instructions, memory, inbox and computer. Share files, put them in group chats and let automations wake them up. Free, local-first, and you bring the model.

Download OpenCue Read the docs
curl -fsSL https://lautaro005.github.io/opencue/install.sh | sh

macOS · Linux · Windows — then run opencue

Assistant 2:05 PMHere's the hearings sheet.
Reminders 1:33 PMThree deadlines this week.
AI Research 10:28 AMI'd test this idea before writing…
Dev Helper 10:27 AMToday's notes are in your projects folder.
Finance YesterdayMy two-year plan would keep…
Research Assistant Oct 2The email wasn't sent: you declined.
Assistant
I emailed Court 9 about the October 9 hearing, 1:00–1:30 PM, asking whether you and your two colleagues can attend.
CSVcourts.csv2 KB
send a reminder to every court in this sheet
✉️ Sent 4 emails courts 15, 2, 1 and 9
✉️ Received an email ›
Court 9 replied: the October 9 hearing is cancelled. I updated the sheet.
XLSXhearings.xlsx14 KB
Message Assistant
Agents

Everything an agent needs to get work done

Each agent is a little coworker with a face, a job and the tools to do it.

Instructions & memory

Tell each agent who it is. It remembers facts about you in plain Markdown files you can read and edit, plus a shared memory for all agents.

Group chats

Put agents in a group, @mention who should answer and let them hand work to each other — with a turn limit so they don't run forever.

Files, both ways

Drop files into any chat. Agents see pictures, read documents from their workspace and send back reports and spreadsheets.

Automations

Wake an agent on a schedule, when new email arrives, from a local webhook or when someone writes in a group.

Its own inbox

Connect any IMAP/SMTP account or Gmail through Composio. Agents read, search and send — sending waits for your OK.

A computer of its own

A local VM you can watch live and that keeps its state in a folder you pick, Perplexity SPACE microVMs, VMPal machines, WSL or — with your approval — your own machine.

Apps & MCP

Plug in any MCP server, import your mcp.json, or connect Gmail, Slack, Notion and 1000+ apps through Composio from inside the app.

Skills

Reusable know-how in the Claude Code SKILL.md format. Write them, import them from a URL, or let agents save their own.

Voice

Press the mic and talk. ElevenLabs turns your voice into text and reads the reply back in the voice you pick.

Models

Bring your own model

Use API keys, local models, or the Codex, Claude Code and Grok CLIs you're already signed in to — now with OpenCue's tools through MCP.

Install

One line, then opencue

Installs for your user — no admin rights — and keeps itself up to date with signed updates.

1Open Terminal and run:

curl -fsSL https://lautaro005.github.io/opencue/install.sh | sh

2Start it from Applications or type opencue.

Or download the .dmg (Apple silicon and Intel). It isn't notarized yet, so the first time macOS blocks it: open System Settings → Privacy & Security and choose Open Anyway. The terminal install doesn't need this.

Latest release on GitHub · all install options

Privacy

Local-first, on purpose

No account, no telemetry. Your chats, files, memory and settings stay on your computer; keys live in the system keychain.

Ask before acting

Emails, commands on your machine, deletions and data-changing tools wait for your approval unless you make an agent autonomous.

Plain files

Instructions, memory and skills are Markdown. Shared files are regular files. Everything else is one SQLite database in your data folder.

Source available

The code is on GitHub under Apache-2.0 with the Commons Clause: free to use, modify and share — just not to sell.

Meet your agents.

macOS, Linux and Windows. Free forever.

Get OpenCue View on GitHub