Free, open-source Cowork desktop app that unifies Claude Code, Codex, Gemini CLI, and 20+ AI agents into a single platform with multi-agent teams, 24/7 cron automation, and zero-config built-in agent.
AionUi is a cross-platform Electron desktop application that redefines how developers and knowledge workers interact with AI agents. Rather than juggling separate terminal windows for Claude Code, Codex, Gemini CLI, and other CLI-based agents, AionUi surfaces them all through a single unified interface — auto-detecting whatever you have installed and letting you run parallel sessions with independent context per agent.
At its core, AionUi ships a built-in agent engine that requires zero external CLI installation. Drop in any API key — Anthropic, OpenAI, Google, AWS Bedrock, Ollama, or any of 30+ supported platforms — and you get a full autonomous cowork agent capable of reading and writing files, browsing the web, generating images, and executing multi-step tasks. The same agent capabilities are available regardless of which model you choose.
The platform’s Team Mode takes multi-agent collaboration further: a Leader agent receives your instructions, breaks them into subtasks via ACP (Agent Communication Protocol), and delegates to Teammate agents that execute in parallel and share results through an async mailbox. Teams share a common workspace folder while maintaining independent permission dialogs per agent.
Beyond interactive sessions, AionUi supports 24/7 unattended automation through cron scheduling, remote access via a built-in WebUI server (accessible from mobile), and messaging integrations with Telegram, Lark/Feishu, DingTalk, and WeChat. Twenty-one built-in professional assistants — covering document creation (PPT, Word, Excel via OfficeCLI), academic writing, financial modeling, and UI/UX design — are ready to use immediately without any additional configuration.
Architecture
AionUi follows a classic Electron layered architecture partitioned across a monorepo with four packages: desktop (the core app), web-host (the remote WebUI server), web-cli (browser-based CLI interface), and shared-scripts. The desktop/src/process main process handles data persistence via a SQLite database (with WAL mode, foreign keys, and a well-indexed schema covering conversations, messages, teams, mailbox, and cron jobs), auto-update logic, and bridge IPC channels. The renderer process in desktop/src/renderer is a React 19 SPA organized into pages per domain (conversation, team, cron, settings) with platform-specific sub-modules for ACP, Aionrs, Gemini, and legacy agent adapters. Agent Communication Protocol (ACP) underpins multi-agent orchestration, with AcpChat.tsx, useAcpMessage.ts, and teamSendRuntime.ts forming the collaboration layer between Leader and Teammate agents. This clean separation means the main process owns persistence and process management while the renderer owns all UI state and agent communication, with IPC bridges connecting the two via typed contracts in common/adapter/ipcBridge.ts.
Tech Stack
AionUi is built with TypeScript throughout (strict mode, Node ≥22) and packaged as an Electron 37 desktop app using electron-vite 5 for the build pipeline. The renderer is React 19 with React Router 7, UnoCSS for utility styling, and Arco Design for the component library. Code editing surfaces use CodeMirror 6 with multi-language support. The main process uses better-sqlite3 (synchronous API, WAL mode) for local data, croner for cron job scheduling, express 5 for the built-in WebUI server with JWT/bcrypt auth, and grammy for the Telegram bot integration. Agent integrations span @anthropic-ai/sdk, openai, @google/genai, @aws-sdk/client-bedrock, and the custom @office-ai/aioncli-core and @agentclientprotocol/sdk packages. Linting uses oxlint and formatting oxfmt (Rust-based, fast), with Husky pre-commit hooks enforcing lint-staged on every commit. Testing employs Vitest for unit/integration tests and Playwright for E2E scenarios.
Code Quality
The codebase has extensive test coverage across unit, integration, and E2E layers — the tests/ directory contains dedicated suites for conversation, team, cron, settings, assistants, skills, and database drivers, with Playwright E2E tests specifically for team mode lifecycle, whitelist, and communication scenarios. TypeScript strict mode is enforced across the entire monorepo, and the Rust-based oxlint/oxfmt toolchain ensures consistent code style on every commit via Husky lint-staged hooks. Error handling is explicit and typed — the database schema uses ISqliteDriver interface abstraction allowing driver swapping between better-sqlite3 and Bun’s native SQLite, and the IPC bridge uses typed contracts rather than stringly-typed channels. The assertBridgeSuccess.ts utility enforces contract validation at bridge boundaries. Overall code quality signals (typed, tested, linted, CI-enforced) are comprehensive for a project of this velocity (nearly 700 commits/month).
What Makes It Unique The core innovation is ACP (Agent Communication Protocol) — a purpose-built multi-agent coordination layer that allows a Leader agent to decompose tasks, communicate subtasks to Teammate agents via an async mailbox backed by SQLite, and aggregate results, all within a shared workspace folder. This is distinct from simply running multiple chat sessions: agents have shared file context, team-isolated permission dialogs, and dynamic scaling (add/remove Teammates mid-run). The built-in OfficeCLI integration for generating Morph-animated PPTX files, production-ready .docx documents, and formula-rich .xlsx spreadsheets directly from agent chat is genuinely uncommon in the cowork agent space. The unified MCP management layer — configure once, sync to all connected external CLI agents — addresses a real fragmentation pain point that other multi-agent UIs leave to users to solve manually.
Licensing Model Apache-2.0 licensed — all features available in self-hosted and local deployments with no restrictions, license keys, or paid tiers required.
Self-Hosting Restrictions None. The entire codebase is open-source under Apache-2.0 and every feature — including Team Mode, cron scheduling, WebUI remote access, messaging integrations, and all 21 professional assistants — is available without any license key or account registration.
License Key Required No. AionUi is fully free and open-source. You supply your own API keys for the AI providers you choose to use (Anthropic, OpenAI, Google, etc.); there is no AionUi subscription or license fee.
Automation · Productivity · AI Assistants
Build, deploy, and run autonomous AI agents that automate complex multi-step workflows using a visual block-based graph editor.
No Code Platforms · AI Development · Developer Tools
Visual LLM workflow platform with RAG pipelines, agent capabilities, and model management for building production AI applications.
Productivity · Project Management · Collaboration
The open-source AI workspace that puts your data, your rules — with local LLMs, CRDT collaboration, and full self-hosting built in.