Your AI team, one message away.
Connect your messaging app to the AI tools on your machine. Send a message from your phone, and OpenBridge coordinates Claude, Codex, and Gemini to explore your workspace and execute tasks — using your existing subscriptions, at zero extra cost.
Features | Quick Start | Prerequisites | Examples | How It Works | Documentation
Claude, Codex, Gemini — working together on your tasks. OpenBridge discovers every AI tool installed on your machine and coordinates them automatically. One AI reads your codebase, another writes the fix, a third runs the tests. You send one message; the system figures out which AI handles which part.
Send a WhatsApp message, and a team of AI agents gets to work on your project. OpenBridge supports 5 channels — WhatsApp, Telegram, Discord, WebChat, and Console. Break complex tasks into subtasks, check progress, ask follow-up questions — all from your phone. Session continuity means the AI remembers every previous conversation.
The Master AI maintains a curated memory.md file (≤ 200 lines) inside your workspace. It is loaded on every session start so the AI picks up exactly where you left off — no re-explaining your project, no lost context. The AI updates the file itself: merging topics, discarding stale info, keeping it focused. Use /history to browse past conversations, search by keyword, or retrieve a full transcript.
Three access levels keep you in control: read-only (browse files), code-edit (modify files and run tests), and full-access (everything). The AI only touches the workspace folder you point it at — nothing else on your machine. A phone whitelist ensures only authorized users can send commands.
On startup, OpenBridge explores your workspace and builds a knowledge base inside your project. It detects git commits and file changes, then re-explores incrementally — so the AI's understanding is always current. Multi-turn conversations maintain context across messages: ask a question, get an answer, follow up — the AI remembers. Across sessions, persistent memory.md ensures key decisions, findings, and preferences survive restarts without bloating the context window.
Workers connect to external services (Gmail, Slack, Canva, databases) via the Model Context Protocol. Each worker gets an isolated MCP config with only the servers it needs — no cross-contamination of API keys. The Master AI autonomously decides which workers get which MCP servers.
Generate Word documents, PowerPoint presentations, Excel spreadsheets, and PDF reports from a chat message. Create diagrams (Mermaid/PlantUML), charts (D3.js/Chart.js), landing pages, slide decks, generative art, and brand assets. Generated files are delivered as attachments via WhatsApp/Telegram or download links in WebChat.
The Master AI plans before it acts — spawning read-only analysis workers before committing to code changes. Workers are grouped into coordinated swarms (research → implement → review → test) with automatic handoff. Test files are protected from unauthorized modification. Fix loops are capped at 3 iterations before escalating.
Domain-specific instruction sets that make workers experts at their task. Built-in packs for security audits, code review, test writing, data analysis, and documentation. Create custom skill packs in .openbridge/skill-packs/ or let the Master auto-create them from successful task patterns.
No API keys. No per-request fees. No new subscriptions. OpenBridge runs locally on your machine and uses whatever AI tools you already have installed — Claude Code, Codex, Aider, or anything else. Your subscription, your machine, your data.
No external accounts needed. Use the built-in Console connector to try OpenBridge immediately.
Prerequisites: Node.js >= 22, Claude Code installed
git clone https://github.com/medomar/OpenBridge.git
cd OpenBridge
npm installCreate config.json:
{
"workspacePath": "/absolute/path/to/your/project",
"channels": [{ "type": "console", "enabled": true }],
"auth": {
"whitelist": ["console-user"],
"prefix": "/ai"
}
}npm run devType a message in the terminal:
/ai what's in this project?
Prerequisites: Node.js >= 22, a WhatsApp account, Claude Code installed.
npx openbridge init
npm run devScan the QR code with WhatsApp. Then from your phone:
/ai what's in this project?
OpenBridge supports 5 channels: Console, WebChat, WhatsApp, Telegram, and Discord. See the Connector Setup Guide for all options.
Required
- Node.js >= 22 — nodejs.org
Optional — by connector / feature
- WhatsApp: Chrome/Chromium — auto-installed by Puppeteer, no extra steps needed
- Voice messages (recommended):
OPENAI_API_KEYenv var — uses the OpenAI Whisper API ($0.006/min, zero local setup; same key Codex uses) - Voice messages (offline):
pip install openai-whisperorbrew install whisper-cpp— free, runs locally - Telegram: Bot token from @BotFather
- Discord: Bot token + application ID from the Discord Developer Portal
You: /ai what's in this project?
AI: This is a Node.js REST API (Express + TypeScript) with:
- 12 routes in src/routes/
- PostgreSQL via Prisma ORM
- JWT authentication in src/middleware/auth.ts
- 47 tests (all passing)
- Docker setup with docker-compose.yml
You: /ai run the tests and fix any failures
AI: On it...
→ Reading test output... 45/47 passing
→ Fixing 2 failures in auth.test.ts
→ Re-running tests... 47/47 passing
Changes committed. All tests pass.
You: /ai how does the payment flow work?
AI: Payment flow (src/services/payment.ts):
1. POST /checkout → validates cart → creates Stripe session
2. Stripe webhook → POST /webhooks/stripe → updates order status
3. Order service notifies user via email (src/services/email.ts)
Key files: payment.ts, order.ts, stripe-webhook.ts
You: /ai what's our current coffee inventory?
AI: Based on inventory.xlsx:
- Arabica beans: 12kg (reorder threshold: 10kg)
- Robusta beans: 8kg
- Espresso blend: 15kg
You're below the reorder threshold on Robusta.
You: /ai which items sold best last week?
AI: Top 5 from sales_weekly.csv:
1. Iced Latte (142 sold)
2. Cappuccino (128 sold)
3. Americano (97 sold)
4. Cold Brew (89 sold)
5. Mocha (76 sold)
You: /history
AI: Recent conversations (last 10):
1. [2026-02-25] "Refactor auth to JWT" — 8 messages
2. [2026-02-24] "Fix failing tests in checkout" — 12 messages
3. [2026-02-20] "What's in this project?" — 4 messages
You: /history search authentication
AI: Found 3 sessions matching "authentication":
1. [2026-02-25] "Refactor auth to JWT"
2. [2026-02-18] "Add OAuth support" — 15 messages
3. [2026-02-10] "JWT token expiry bug" — 6 messages
You: /history 2026-02-25-auth-refactor
AI: [Full conversation transcript...]
You: /ai refactor auth to use JWT
AI: Complex task. Breaking it down...
→ AI 1 (fast, read-only): reading current auth code
→ AI 2 (powerful, code-edit): implementing JWT authentication
→ AI 3 (fast, code-edit): running tests and verifying
AI: Done. Refactored to JWT. 4 files modified, all tests pass.
One message from you. Three AI agents coordinated automatically. Each with the right capabilities for its subtask.
More examples: Use Cases — software teams, cafes, law firms, real estate, and more.
Your Phone / Browser Your Machine Your Workspace
───────────────────── ───────────────── ─────────────────
WhatsApp · Telegram OpenBridge .openbridge/
Discord · WebChat ──────> authenticates, ──────> workspace map
Console routes messages, learnings
<────── coordinates AI <────── session state
workers task history
- You send a message from any channel.
- OpenBridge authenticates it and routes it to the lead AI, which decides how to handle it.
- For complex tasks, the AI spawns focused workers — each with specific access permissions and capabilities — then synthesizes the results and responds.
Deep dive: Architecture | Project Overview | API Reference
| Guide | Description |
|---|---|
| Documentation Hub | All docs in one place |
| Project Overview | Vision, architecture, roadmap |
| Configuration Guide | All config options explained |
| Connector Setup | Setup guides for all 5 channels |
| Use Cases | Examples for every industry |
| Architecture | System design, message flow, layers |
| Deployment Guide | Docker, PM2, systemd setup |
| Troubleshooting | Common issues and solutions |
 Sayadi Med Omar Founder & Lead Developer |
 Haifa BEN LETAIFA Founder & Lead Developer |
We welcome contributions! Whether it's a new connector, AI tool integration, bug fix, or documentation improvement — see CONTRIBUTING.md for guidelines.
To report security vulnerabilities, please see SECURITY.md.
Apache License 2.0 — see LICENSE.