Whirl: the open-source self-hosted chat app that fits every model in one window

What this project does

Whirl is an open-source AI chat application that crams nearly every mainstream large language model into a single chat window. Behind it is the full source code of whirl.chat, a real product already running in production, released entirely under the MIT license. In plain terms, it wants to be a self-hosted "everything chat hub": you no longer juggle between ChatGPT, Claude, and Gemini tabs, you can call every model from one interface, and let them share memory and your own tools. Its most interesting idea is "living artifacts" — analyses, charts, even a fully interactive web page generated in a chat stay editable in a side panel and can be shared by link, instead of vanishing into a long scroll of bubbles once the conversation ends. For anyone who lives in AI chat all day, that single shift from "disposable conversation" to "persistent workspace" is the whole reason to care.

Why it blew up recently

Whirl went open source in early October 2026 (the repo was created on October 2) and climbed past 450 stars in just three days. It hits the pain point of "AI chat app aesthetic fatigue": the official ChatGPT web app keeps piling on features while locking you deeper into a subscription, and open-source options like Open WebUI or LibreChat are either configuration-heavy or feel too geeky. Whirl sells "sweating the details" — smooth animations, dark mode, an installable mobile PWA, voice input — plus a privacy pitch ("locked chats encrypted locally with a password, answered only by zero-retention models"). The launch thread filled with comments like "finally an open-source chat app that respects my data and still looks modern," which is exactly the combination people felt was missing, and that is why it spread fast across Hacker News and dev communities.

Technical highlights

Whirl uses Next.js for the frontend and Convex for the backend and real-time database; the entire full-stack code is open, so you can read the implementation or deploy your own in one go. The features that actually stand out:

• All models in one frame: Routed through OpenRouter, GPT, Claude, Gemini, and open models can all be switched inside the same conversation, with adjustable "thinking level" — deep reasoning for coding, off for small talk to save tokens.
• Living artifacts: This is its biggest difference from others. Documents, charts, even complete interactive pages produced in chat stay in a right-side panel for further editing, and once edited you get a link to send to a colleague — turning AI chat into version-aware collaborative documents.
• Real memory: Long-term memory carries your preferences and ongoing projects across chats, instead of going blank every time you open a new window.
• Your own tools: Connect your own tools over the MCP protocol (with OAuth login built in), so for example you can let Whirl read a calendar or push to a Notion database, and install community skills to teach it new tricks.
• Privacy details: "Locked chats" are encrypted on your device with a password the server never sees, and only zero-retention models answer them. There is also incognito mode, message queueing, image generation, file attachments, folders, and a whole set of thoughtful touches.
• Images, voice, and mobile: Built-in image generation, voice input for hands-free chatting, and an installable PWA so the app lives on your phone home screen and feels native — details that usually take a weekend of glue code in other self-hosted setups.

Who it is for

If you are tired of jumping between multiple AI websites and want to gather your usual models and workflows into one self-hostable place, Whirl fits well. It is especially good for: people who want a clean, good-looking personal AI assistant they can install on their phone; people who need to share AI-generated analyses or prototypes with a team; and tech enthusiasts who care about keeping data in their own hands instead of being tied to one vendor's subscription. A concrete example: a PhD student can keep one research chat that remembers which papers were already summarized, so the next session picks up instead of restarting; a solo founder can keep product specs and competitor notes alive across dozens of conversations without copy-pasting into a separate doc.

Quick try

To run it locally (needs a Convex account):

git clone https://github.com/whirlchat/whirl && cd whirl && npm install && npx convex dev

Then npm run dev and open the local address. If that is too much trouble, just use the hosted version at whirl.chat.

Compared with competitors

Put it next to three common open-source chat apps:

• vs Open WebUI: Open WebUI is powerful but configuration-heavy and engineer-aesthetic; Whirl puts "good-looking and smooth" first, and living-artifact sharing is something Open WebUI lacks.
• vs LibreChat: LibreChat also aggregates all models, but Whirl's "living artifacts + real memory + locked encryption" combo leans more toward daily productivity than pure conversation.
• vs the official ChatGPT web app: The official app has many features but your data lives entirely with OpenAI and models are locked; Whirl is self-hosted, model-agnostic, and the privacy initiative stays with you.

In one line: if you want "the best-looking all-in-one open-source chat app," Whirl is now the strongest new candidate.