RU

Whirl: open-source AI chat with memory, live documents, and MCP

Published: 2026-10-03 · Author: AI Release · @ai_release1
Whirl: open-source AI chat with memory, live documents, and MCP

⚡ The gist in 5 seconds - Whirl is a full-stack open-source AI chat, published on GitHub under the MIT license. - The app is available as a service at whirl.chat and can also be self-hosted. - It requires Convex, Clerk, and OpenRouter keys; some features are enabled with separate keys. ### 🔍 What we found Whirl is built on Next.js 16, React 19, Tailwind CSS v4, and Motion. The backend uses Convex (database, functions, streaming, crons), Clerk for authentication, and models are connected via OpenRouter using the AI SDK. The tooling is Bun workspaces and TypeScript. In a single conversation, you can chat with all the top models, routing goes through OpenRouter, and the level of "reasoning" is configurable. The app can create live artifacts: documents, charts, and interactive pages that remain editable in the sidebar and can be shared via link. A key feature is integrations with your own tools via MCP (including OAuth) and "skills" that extend the chat's capabilities. Long-term memory stores preferences and projects across chats (SuperMemory), and there's live web search and page reading. Security comes from encrypted chats: encryption happens on-device with a password the server never sees, and responses are only given by models with zero data retention. Also available: incognito mode, message queueing, voice input, image generation, attachments, folders, sharing, light/dark theme, accent colors, and an installable mobile web app. ### 💡 Why it matters Whirl lets you work with all the top models in a single conversation without switching between services, while keeping context through long-term memory and live documents. The open-source code lets you run your own instance, study the architecture, and adapt the app to your needs. For geeks, it's a convenient entry point into the world of multi-model chats while retaining control over your data. ### 🧩 Context Besides the main web app (apps/v2, the one in production), the repository contains an admin console for managing models and integrations (console), a native mobile app built with Expo (mobile), a separate waitlist page (waitlist), and a legacy version that is unsupported and kept for reference only. The backend is split into a separate package, packages/backend, which also hosts the documentation and branding. To run the project you'll need Bun, a Convex account, a Clerk app, and an OpenRouter API key — all steps are described in detail in the self-hosting and configuration guides.

🔗 Read on github.com

🤖 AI summary
AI-чатopen-sourceсамохостингNext.js
📖
Read the guide on this topic
Read →
← PreviousExperiment: how to make Windows 11 tell the SSD "Insert a disk into drive A:"Next →react-native-blur: frosted glass and backdrop blur for React Native and Expo

Source: github.com · post in Telegram