A free, open-source disk-space visualizer. Squarified treemap, duplicate hunting, disk-full forecasting, and cleanup that only ever moves files to the Trash.
Ayakaleaf Pro is a free, open-source, self-hosted Overleaf alternative: Overleaf Community Edition with almost all Server Pro features — track changes, project history, admin panel, SSO, Git-bridge, Z
Then start it with opencue. Installs for your user, no admin rights, and updates itself with signed releases.
Prefer an installer? Grab the .dmg, .AppImage, .deb, .rpm, .exe or .msi from
Releases.
First run: a short walkthrough asks your name, connects a provider (or start Ollama / LM Studio), picks a model and
creates your first agent. Replay it any time from Settings → General.
[!NOTE]
The apps aren't notarized or code-signed by Apple or Microsoft yet. A downloaded .dmg needs System Settings →
Privacy & Security → Open Anyway the first time (the terminal install doesn't); Windows SmartScreen may ask you to
confirm.
What agents can do
Bring your own model: OpenAI, Anthropic, xAI (Grok), Google Gemini, Meta Muse, Mistral, Groq, OpenRouter, CheaperInference,
Ollama (local and cloud), LM Studio, any OpenAI-compatible API — or the Codex, Claude Code, Grok, Muse Code,
Mistral Vibe and Antigravity CLIs you're already signed in to.
Safe by default: in Ask first mode, sending email, commands on your machine, deletions and data-changing tools wait
for your OK. No account, no telemetry; secrets stay in a private file on your computer.
Not planned: phone numbers, payments/wallet, QR "scan & create".
ChatGPT and Claude subscriptions, the allowed way
OpenAI and Anthropic don't offer third-party sign-in for consumer subscriptions, and reusing their tokens elsewhere breaks
their terms. OpenCue runs your installed, logged-incodex, claude, grok, muse, vibe or agy binary instead, so your subscription stays in
the official client and OpenCue never sees the tokens. Those agents still get OpenCue's tools — memory, web, skills,
computer, files, email, catalog apps — through a local MCP server (opencue mcp-serve --agent <id>).
Develop
Requirements: Rust (stable), Node 22+; on Linux libwebkit2gtk-4.1-dev libayatana-appindicator3-dev librsvg2-dev; on
macOS the Xcode command-line tools.
cd apps/desktop
npm install
npm run tauri dev # desktop app with hot reload
npm run dev # UI only, in a browser, with an in-memory mock backend
cargo test --workspace # core tests, incl. an end-to-end agent loop against a mock LLM
cargo run -p opencue-cli -- --help
Release builds sign update bundles, so a local npm run tauri build needs TAURI_SIGNING_PRIVATE_KEY, or add
-c '{"bundle":{"createUpdaterArtifacts":false}}'.
crates/opencue-core Rust core: agents, providers, runtime, memory, files, MCP client and server, web, skills,
automations, email, catalog apps, computers (Docker/Podman/SPACE/WSL/host), voice and local
speech models, PersonaPlex, terminals, the AI council, coding groups (git/GitHub), storage
crates/opencue-cli opencue-cli companion
apps/desktop Tauri 2 shell (src-tauri) + React/Tailwind UI (src)
site/ website (GitHub Pages): home, /docs, /privacy, /changelog, install scripts, llms.txt
docs/ design doc, SPACE setup guide
.github/workflows CI (Linux/Windows/macOS), tagged releases (macOS/Linux/Windows), website deploy
macOS
Linux
Windows
~/Library/Application Support/opencue
~/.local/share/opencue
%APPDATA%\opencue\data
Inside: models/ (downloaded speech models, the sherpa-onnx runtime and PersonaPlex's Python environment), space/
(the SPACE checkout, when OpenCue installed it), opencue.db (SQLite: agents, groups, messages, providers, MCP servers, automations, email accounts, settings,
web cache), agents/<id>/ (INSTRUCTIONS.md, MEMORY.md, workspace/, space.json), files/ (shared files),
memory/GLOBAL.md and skills/<name>/SKILL.md. Set OPENCUE_HOME to use another folder. API keys, email passwords and
tokens live in secrets.json (mode 0600) in the data folder; catalog app tokens are in oauth/<app>.json (also 0600).
The system keychain is not used.
Bump version in Cargo.toml, apps/desktop/package.json and apps/desktop/src-tauri/tauri.conf.json, then push a
tag vX.Y.Z. GitHub Actions builds the macOS (universal), Linux and Windows installers, signs the update bundles with
the TAURI_SIGNING_PRIVATE_KEY secret, publishes latest.json for the in-app updater and attaches the CLI zips.
Changes to site/ deploy the website.
0.2: drawn avatars, automations, email, computers with terminal and file browser, skills, Composio, voice,
website.
0.3: files in chats and groups, OpenCue tools for CLI agents (MCP), Docker desktop view, auto-update, macOS
builds, one-line install.
0.3.1: first-run walkthrough, separate Providers and Models pages (show/hide models, search), model picker
when creating agents, VMPal machines, / commands (/create-automation, /create-skill, skills), self-building
memory, About me, floating avatar picker and app-drawn dropdowns, menus and dialogs.
0.3.2: base system prompt, agents that write their own instructions and soul (SOUL.md) and greet you by
asking about their job, Grok CLI, system notifications, tray menu, VMPal as a computer option and fixes for the Claude
and Codex CLIs, resizable panels.
0.3.3: per-agent working folder, searchable Composio app list (and the right key header), links as pills, real
model lists for the Claude/Codex/Grok CLIs, and a sturdier agent loop (stable prompt prefix, long results to files,
compaction, plan recitation, research helpers).
0.3.4: Composio consumer keys through Composio Connect, local VMs with a live desktop, memory/CPU limits and a
folder that keeps their state, Perplexity SPACE with its GitHub link, and provider logos on the website.
0.3.5: agents that talk to each other in 1:1 chats (send_to_agent, @ and a "Chat with …" pill), the
yellow ask/permission card (ask_user), shared or per-agent Docker VMs with an idle stop and a setup checklist, a
Composio refresh button and a fix for Claude Code agents hitting permission errors on Composio.
0.3.6: Podman as an alternative to Docker, a Catalog of OAuth apps instead of Composio (red/yellow/green cards),
a terminal that opens by itself, a right-click menu on agents and groups (rename, pin, clear, reset, delete), voice
calls in a phone-sized window, and secrets in a private file instead of the keychain.
Use cases
A paralegal agent with its own inbox that emails courts, tracks replies and reminds you of hearings.
A reminders agent that checks your boards every Monday at 9:00 and tells you what is due.
A dev agent with a Docker computer and desktop that runs scripts, opens apps and sends you the results as files.
A group chat where a planner hands tasks to a researcher and a writer.
License
OpenCue is free to use, modify and share, but not to sell. It is licensed under the
Apache License 2.0 with the Commons Clause: you may not sell the software, or a product or service (including
hosting or paid support) whose value derives substantially from it. Because of that condition it is source-available
rather than OSI open source.
Security & privacy
Report vulnerabilities privately — see SECURITY.md. What leaves your computer is described in the
privacy policy.
0.3.7: permission cards for Claude Code agents, a stop button for VMs, Podman without Docker, Composio in the
Catalog (with the apps it connects), a live green card for agent-to-agent chats and logos from your assets on the site.
0.3.8: a working Podman desktop (Cua's XFCE image), stop-all and play buttons in Settings → Computers, desktop
view on by default, the chosen VM respected over VMPal, folded reasoning, a gray call button without an ElevenLabs
key and an Agents / Skills menu under the chat's +.
0.4.0: a built-in browser agents can drive (Browser tab next to the VM, terminal below), WhatsApp through a QR with
green cards in the chat, templates for agents and teams with an x.ai import, a message search (⌘K) and the 5-hour and
weekly limits of your Codex and Claude Code subscriptions.
0.4.1: Google Gemini, Meta Muse (API and Muse Code CLI), complete templates (soul, skills, memories), agent chat
cards in each agent's color, one browser page per agent and a reliable test notification.
0.4.2: Artifacts (HTML, SVG, Markdown, Mermaid, code, and real Excel, Word, PowerPoint and PDF files, with
versions, kept in each agent's Files), a dashboard home screen with Esc to return and /dashboard to redesign it,
copy / edit / regenerate on messages, Mistral API, Groq, Mistral Vibe and Antigravity providers with their logos, and
browser tabs.
0.5.0: local speech models (Parakeet, Canary, Whisper, SenseVoice, Kokoro) and NVIDIA PersonaPlex for
real-time voice (also from a phone), /plan with a live status card, system and user skills in the chat's +, /canvas
and /artifact, artifact cards with progress, resizable and floating panels, terminals for Claude Code, Codex, Grok
and more, the AI council, coding groups with pull requests and a protocol for agents to talk, SPACE that installs
itself, ordered providers and a fix for the browser covering its toolbar.
0.5.1: agents in a group work in parallel (and you see it in the plan card and the sidebar), a message queue
while an agent works, question cards with three options, permissions answerable inside groups, a coding-team window
(name and folder, git init for you), live files per agent, coding-agent CLIs as agents, drag-to-reorder and a
collapsible sidebar, notifications in the notch, the light Moshi voice engine for Apple silicon, and the browser
placed inside the app's own page.
0.5.2: /plan as a self-ticking checklist, coding teams working in one shared project folder, permissions of
parallel agents inside the group, a tabbed Computers page that always shows the SPACE setup.
0.5.3: four permission modes (nothing runs, ask, ask only the critical, allow all) that work the same in chats and
groups and map to each CLI's sandbox, a Reply button, diffs/commits/PR files for coding teams, brain + Lumina + WorkStudio
connected from Settings → Family, update buttons for the CLIs and the skills your CLIs already have.
0.5.4: one orchestrator per /plan that hands the steps out, a new plan card (checklist with who does each
step, live states, nothing shown twice), Settings → Storage, Providers + Models in one page, WorkStudio inside Apps,
Reply above the message box and a more polished design.
Still open: notarized macOS and code-signed Windows installers (need an Apple Developer ID and a Windows certificate).
