Whirl opens self-hosted AI chat

Whirl’s strongest idea is treating chat as an extensible workspace rather than a disposable conversation. The feature breadth is compelling, though deploying it still requires several external services and API keys.

• –Editable artifacts turn model output into working documents, dashboards, and interactive pages.
• –MCP support gives developers a path to connect proprietary tools and workflows directly inside conversations.
• –Long-term memory and model switching make Whirl more useful for ongoing projects than single-session chat clients.
• –MIT licensing and self-hosting appeal to teams that want control over data, integrations, and provider choice.
• –The architecture’s reliance on Convex, Clerk, OpenRouter, and optional hosted services means “self-hosted” does not necessarily mean fully local.