OS

The open-source AI agent platform.

Build, deploy, and manage intelligent conversational agents — from prototype to production — in minutes.

Demo by Miguel Amigot, CTO at ibl.ai

Features · Screenshots · Quick Start · Deployment · Troubleshooting · Testing · Contributing

---

Features

• AI Agents — Create custom agents with configurable LLMs, system prompts, tools, and safety filters
• RAG Training — Upload documents, connect Google Drive, OneDrive, Dropbox, or crawl websites to ground agents in your data
• Voice Calls — Real-time WebRTC voice chat powered by LiveKit
• Deep Research — Extended multi-step reasoning for complex queries
• Canvas / Artifacts — Generate, edit, and version rich documents alongside chat
• Screen Sharing — Share your screen directly inside a chat session
• Web Search — Ground responses with live web results
• MCP Servers — Extend agent capabilities with Model Context Protocol tool servers
• Analytics — Usage dashboards, topic analysis, transcript viewer, and financial reporting
• Projects — Collaborative workspaces to group agents with shared context and goals
• Cross-Platform — Ships as web, desktop (macOS, Windows, Linux), and mobile (iOS, Android)
• Multi-tenancy — Full tenant isolation with per-org configuration, branding, and user management
• SSO — Single Sign-On with configurable identity providers (OAuth, OIDC, SAML)
• RBAC — Granular role-based access control with policies and group-based permissions
• Stripe Billing — Subscription management, free trials, and usage-based pricing
• Embed Mode — Embed agents in any website via iframe with custom styling
• Custom Domains — Host agents on your own domain
• API Keys — Programmatic access for integrations and automation
• Whitelabeling — Custom branding, logos, and disclaimers

---

Available On

Platform	Status
Web	Production at os.ibl.ai — works on any modern browser
macOS	Native desktop app — lightweight, fast, system-integrated
iOS	Native mobile app — available on iPhone and iPad
Android	Native mobile app — available on phones and tablets
Windows	Native desktop app
Linux	Native desktop app

One codebase, six platforms. OS runs natively everywhere your users are — lightweight, fast, with near-native performance.

---

Screenshots

Configure agents with LLMs, prompts, safety filters, and explore conversation starters

Customize agent identity, description, and appearance

Connect to external services via Model Context Protocol — GitHub, Notion, Slack, and more

Manage agent memory with knowledge gaps, learning goals, and user preferences

Control agent visibility, access permissions, and LTI integration

---

Quick Start

git clone https://github.com/iblai/os.git
cd os
pnpm install

Using Claude Code? Run /setup — it will walk you through connecting your ibl.ai tenant and configuring .env.local automatically.

Manual setup: Copy .env.example to .env.local, then set NEXT_PUBLIC_MAIN_TENANT_KEY to your org key from login.iblai.app/me.

cp .env.example .env.local   # then edit NEXT_PUBLIC_MAIN_TENANT_KEY
pnpm dev

Open http://localhost:3000. See the full Development Guide for environment variables, scripts, and architecture details.

---

Deployment

OS is the frontend for the ibl.ai platform. Choose your path based on your backend setup:

Option A: Existing ibl.ai Tenant

If you already have an ibl.ai tenant (organization key):

1. Configure your tenant

   cp .env.example .env.local

   Update these values with your tenant details:

   NEXT_PUBLIC_TENANT=your-tenant

2. Deploy with Docker (recommended)

   docker build -t os .
   docker run -p 5000:5000 --env-file .env.local os

   Or deploy standalone:

   pnpm build
   PORT=3000 node server-wrapper.js

   The build emits a self-contained server under .next/standalone/ (Next.js standalone output). next.config.ts pins outputFileTracingRoot to the project directory so the output always lands at .next/standalone/server.js with its static assets alongside it. See Troubleshooting if the app loads to a blank screen.

Option B: Enterprise Deployment

If you need full backend infrastructure:

1. Get an enterprise license

   Reach out at ibl.ai/contact to get a license of the enterprise platform (full backend codebase).

2. Deploy with our infra CLI

   If you already have access to our Docker images, deploy them easily via iblai/iblai-infra-cli.

Note: OS requires the ibl.ai backend platform for authentication, AI agent APIs, and data services. The backend is not included in this repository — visit ibl.ai to get started.

Desktop & Mobile

See docs/development.md for native app build instructions.

Full deployment docs: Docker & Standalone

Troubleshooting

The app loads to a blank page or stays stuck on the loading spinner (no redirect to login).

Open your browser's DevTools → Network tab and reload. If every request under /_next/static/... and /env.js returns 404/503, the server isn't finding its static assets. Two common causes:

• A duplicate or stale server is bound to the port. An older node/next process from a previous run can keep listening on :3000 and shadow the new one (a process bound to a specific address such as 127.0.0.1 wins over a wildcard bind). Find and stop strays before starting fresh:

  lsof -nP -iTCP:3000 -sTCP:LISTEN   # list listeners on the port
  kill <PID>                         # stop the stale one

• The standalone output was nested under an unexpected path. Next.js infers the file-tracing root from the nearest lockfile. A stray lockfile in a parent directory (e.g. ~/package-lock.json) makes it treat your home directory as the workspace root and emit the server at .next/standalone/<path-to-project>/server.js instead of .next/standalone/server.js — post-build.sh then copies static assets next to the wrong path and server-wrapper.js can't find the server. This repo pins outputFileTracingRoot in next.config.ts to prevent it; if you still hit nesting, remove the stray parent lockfile and rebuild.

---

Testing

This project is covered by Playwright end-to-end tests in e2e/. Run the E2E suite for any change so nothing regresses:

make e2e-ui

make e2e-ui launches Playwright in interactive UI mode — watch the journeys run, step through them, and re-run individual tests. The first time, install the browsers once:

make e2e-install

Other useful targets:

Command	What it does
make e2e	Run the full suite headless (all browsers)
make e2e-headed	Run with a visible browser
make e2e-chrome	Run on Chrome only
make e2e-journey J=01	Run a single journey spec
make e2e-report	Open the last HTML report

See e2e/COVERAGE.md for current coverage. Coverage must not regress — add or update a journey whenever you change user-facing behavior.

---

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines. If you'll be working with AI-assisted tooling, read AGENTS.md first — it documents the formatting, lint, and push protocol rules that the husky hooks enforce.

---

Resources

• Documentation
• Development Guide — setup, scripts, architecture, configuration
• iblai-app-cli — CLI for scaffolding ibl.ai apps
• @iblai/mcp — MCP server for AI-assisted development
• Vibe — developer toolkit for building with ibl.ai
• Vibe Starter — pre-wired Next.js + ibl.ai SSO template

---

License

MIT License. See LICENSE for details.