Every AI you already pay for, working as one team. They reach every app any of them is connected to, brief you each morning on what needs you, and ask before they act. From your browser, your phone, or inside Claude Code.
Website · Install · Features · Full guide · Contributing
A group chat: Claude builds it, Antigravity reviews it, Codex updates the tests. (Demo recorded with scripted replies.)
You probably already pay for one or two AIs, and each lives in its own app or terminal with its own connected apps. RAPR AI puts them all in one place, and the whole is bigger than any one of them:
- More reach than any single AI. Claude Code has your claude.ai connectors (Gmail, Calendar, Slack, Drive…) and its MCP servers; Codex and Gemini have theirs; RAPR adds its own (any MCP server, Zapier, Composio). RAPR uses each app through the AI that already has it, with that AI's sign-in, so nothing is set up twice. Different accounts can live in different AIs: a work Gmail through Claude, a personal one through Codex.
- It comes to you. Each morning your brief lists what needs you, from all of those apps at once plus your approvals, unfinished sessions and scheduled tasks, with one tap to hand each item to the AI that has the app. Schedules and check-ins run as often as you like, on your computer or your own server.
- It asks before it acts. When an AI wants to send an email, post, merge, pay or delete in a connected app, it waits for your OK on your phone or in the app. That covers everything Claude Code can reach (its MCP servers, your claude.ai connectors and RAPR's apps) and tools any AI calls through RAPR, under one set of your rules and a tamper-evident audit log.
- Use them together, not one at a time. Group chats, bake-offs and second opinions: Claude builds it, Codex reviews it, the best answer wins.
- Everywhere you are. Browser, Telegram, WhatsApp, Discord, Slack, phone notifications, and inside Claude Code, Codex, Cursor or Windsurf as an MCP server.
- Bring the AIs you already have. RAPR runs the official command-line tools with your own sign-ins, so your existing plans and limits just work. Nothing to sign up for. Free and open source (MIT); everything runs on your computer or your own server.
An assistant from one AI company sees that company's connectors, runs that company's model, and lives in that company's app. RAPR sits above all of them:
| One AI's own assistant | RAPR AI | |
|---|---|---|
| Apps it can see | That AI's connectors | Every connected app of every AI you use, plus any MCP server, Zapier or Composio |
| Accounts | Usually one per app | Several: each AI can hold a different account of the same app |
| Who does the work | Its own model | Claude Code, Codex, Antigravity, Copilot, Cursor, Ollama and more; pick one, let Auto choose, or make them compete |
| Schedules and briefs | What that product offers | As many as you want, on your computer or server, running on your AI plans |
| Where you use it | Its own apps | Browser, Telegram, WhatsApp, Discord, Slack, and inside Claude Code, Codex, Cursor and Windsurf (MCP) |
| Before it acts | Its own rules | Your rules and one audit log over everything Claude Code reaches and every tool call through RAPR |
| Price | A subscription | Free (MIT), on the plans you already have |
| ☀️ Your brief | Each morning (and check-ins if you like) one list of what needs you: approvals, unfinished sessions, scheduled results, and what each AI you pick finds in its own connected apps. Arrives on Telegram, WhatsApp, Discord, Slack, your phone and under Needs you; one tap hands an item to the AI that has the app. |
| 🛡️ Asks before acting | An AI that wants to send, post, comment, merge, pay or delete in a connected app waits for your OK first, including Claude Code's own MCP servers and claude.ai connectors. Your rules can always allow, always ask or never allow; every decision is logged. |
| 💬 Group chats | Several AIs in one conversation. @mention who should answer, or @all. Ready-made crews give each AI a job: code review, research, launch, debate. |
| 🧪 Bake-off | Give the same task to several AIs. Each works in its own copy of your project; compare the diffs, test results and time, then merge the winner. Share the result as a card, or publish it to the public RAPR Arena leaderboard. |
| 🔌 RAPR inside your editor | RAPR is an MCP server: from Claude Code, Codex, Cursor or Windsurf, ask another AI to review your changes, answer a question or join a bake-off, or ask "what needs me today?" for your brief. Runs on your computer; your files stay untouched unless you allow edits. Set it up |
| 🐧 Monthly recap | Your month as a shareable card: your AI crew and its MVP, streaks, bake-offs, and what running AIs locally saved you. Worked out on your computer. |
| 🧭 Smart routing | Choose "Auto" and RAPR picks the AI for each task: free local models for easy work, the strongest coding agent for hard work, skipping any AI that's over budget. |
| 📱 Phone control | Telegram bot, WhatsApp (scan a QR code, no Business API), Discord and Slack. Start sessions, send tasks, approve actions, get your brief and hand items off with /do, get files back. |
| 🖥️ Parallel sessions | Many sessions at once, each with its own AI, folder and history. Resume any past conversation. |
| 🧠 Shared memory & playbooks | What one AI learns is available to the others. RAPR keeps how-to notes from past tasks. |
| ☁️ Your own server | One command installs RAPR on any Linux server (Oracle's free tier works) with a free public address, so it runs while your laptop is off. |
Coding agents (they use your own sign-in): Claude Code · Codex · Antigravity · Copilot CLI · Cursor · OpenCode · Droid · Qwen Code · Kiro (Amazon Q) · Aider · Goose · Kilo Code · OpenClaw · Open Interpreter
Free and local: Claude Code and Codex on a local model through Ollama, plus Ollama on its own
Models through an API key: Groq · OpenRouter · GitHub Models · LM Studio and any OpenAI-compatible server
Anything else: add any agent that speaks the Agent Client Protocol in Settings → Agents over ACP, or any command-line AI in Settings → AI Integrations.
Any app with an MCP server: click + Add MCP server in RAPR Packages, paste its URL (or command, or the JSON snippet from its docs) and sign in. Every AI in RAPR can then use its tools.
Install the tools you want, and RAPR finds them. Settings → AI Integrations lists every supported AI with its install command.
One command on each platform. It installs the latest version, and the same command updates it later.
Windows 10/11 (in PowerShell):
irm https://raw.githubusercontent.com/rawalrahul/raprai/main/scripts/install-windows.ps1 | iexAdds RAPR AI to your Start menu and desktop, plus a rapr command. Python 3.12 is installed for you if it's missing.
macOS 12+ (in Terminal):
curl -fsSL https://raw.githubusercontent.com/rawalrahul/raprai/main/scripts/install-mac.sh | bashAdds a RAPR AI app to your Applications, plus a rapr command.
Linux server (Ubuntu/Debian, x86 or ARM, over SSH):
curl -fsSL https://raw.githubusercontent.com/rawalrahul/raprai/main/scripts/install-server.sh | bashRAPR runs in Docker and prints a free public address to open. See docs/cloud-setup.md.
On first launch, a short setup page asks for a PIN and, optionally, a Telegram bot. Then pick an AI and start a session.
| Windows / macOS | Linux server | |
|---|---|---|
| Update | Settings → Updates → Update now, or rapr update (or run the install command again) |
rapr update |
| Uninstall | rapr uninstall: type yes to keep your chats and settings, all to delete them |
rapr remove |
| Start / stop | rapr start / rapr stop (or the RAPR AI app) |
rapr restart |
| Status and logs | rapr status, rapr logs |
rapr status, rapr logs, rapr url |
How RAPR knows about updates: every hour it asks GitHub whether the branch it was installed from (main by default) has new commits. That's an ordinary web request carrying RAPR's version number; nothing else is sent. When there's something new, a dot appears on Settings, and Settings → Updates lists what changed. Offline, the check just skips. Turn it off in the same place (or set UPDATE_CHECK=0 in .env); Check now still works.
Your chats and settings live in %LOCALAPPDATA%\RAPR AI (Windows) or ~/Library/Application Support/RAPR AI (Mac), so updating keeps them. Used the old Windows .exe installer? Run the Windows command: it uses the same folder, so everything carries over. Then remove the old app in Windows Settings → Apps.
From source: see the full guide.
RAPR is an MCP server, so the AI you're already coding with can ask the others for help. After installing RAPR, add it once:
claude mcp add --scope user rapr -- rapr mcp # Claude Code
codex mcp add rapr -- rapr mcp # CodexOn Windows, end both with -- cmd /c rapr mcp. For Codex, also add tool_timeout_sec = 600 under [mcp_servers.rapr] in ~/.codex/config.toml: Codex gives up on a tool after 60 seconds, and a review usually takes a few minutes. Cursor, Windsurf and other clients: add a server with the command rapr and the argument mcp.
Then ask in plain words: "get a second opinion from Codex on my changes", "ask Claude how it would structure this", "run a bake-off between Claude and Codex on this task, with pytest -q", "what needs me today?".
| Tool | What it does |
|---|---|
second_opinion |
Another AI reviews your uncommitted changes (or the last commit, or changes since a branch you name) and reports bugs, most serious first. |
ask_agent |
Puts any question or task to another AI and returns its reply. |
start_bakeoff, bakeoff_status, merge_bakeoff, discard_bakeoff |
A bake-off from your editor: 2 to 4 AIs, each in its own copy of the project; merge the one you like. |
list_agents |
The AIs installed and signed in on this computer. |
latest_brief |
What needs you, from your latest brief in the RAPR app. |
The other AI works in a temporary copy of your project, so your files only change if you allow edits or merge a bake-off. Nothing is hosted and the RAPR app doesn't need to be open. More in the full guide.
Browser ─┐ ┌─ Claude Code ─ Gmail, Calendar, Slack (claude.ai), GitHub
Telegram ─┤ ├─ Codex ─────── Linear, a second Gmail
WhatsApp ─┼─▶ RAPR AI (one Python process) ──┼─ Antigravity
Discord ─┤ sessions · brief · group chats ├─ Ollama (offline)
Slack ─┤ approvals · rules · audit └─ RAPR's own MCP servers, Zapier, Composio
Your editor (Claude Code, Cursor, Windsurf: MCP)
RAPR starts each AI's own command-line tool in the folder you choose and streams its replies to every channel. Each AI keeps its own connected apps (Claude Code's claude.ai connectors and MCP servers, Codex's, Gemini's); RAPR reaches them by asking that AI, and adds a hook so Claude Code asks you before it acts in one. Your chats, settings and keys stay in RAPR's data folder on your machine; keys are encrypted.
- Full guide: Telegram setup, every command, sessions, history, files, memory, packages and backups
- Run it on your own server
- Release notes
Ideas, bug reports and pull requests are welcome. See CONTRIBUTING.md, and look for issues labelled good first issue. If RAPR AI is useful to you, a ⭐ helps other people find it.
MIT: use it, fork it, build on it.
