The Challenge
Every AI chat app makes you pick a model before you've even formed your thought. Claude for writing, GPT for code, Gemini for research. The cognitive overhead of choosing the right tool interrupts the creative flow you came to accelerate.
For solo builders who think by talking through problems, context-switching between AI tools fragments the conversation and loses the thread.
The Approach
What if the chat app figured out who should answer? You just talk. The system classifies your intent and routes to the right specialist — a strategist for business questions, a coder for implementation, a critic for pressure-testing ideas.
Voice input makes it even more natural. Hold to record, release to send. Think out loud and let the system handle the rest.
The Solution
Six AI personas with distinct system prompts and expertise areas. A pattern-based intent classifier routes every message to the right specialist without user intervention. @mention syntax for explicit invocation when you know who you want.
Conversational voice input via Groq's Whisper integration. Hold-to-record with visual feedback. Conversations persist across sessions in Pocketbase, so you never lose context.
The Outcome
The model selection disappears. You think, you talk, you get answers from the right perspective. Conversations feel like working with a small team instead of switching between tools.
Deployed on Hetzner, accessible from anywhere. Voice input means you can brainstorm while walking, not just while typing.
What I Learned
Routing is the product. The personas are table stakes — anyone can write system prompts. The value is removing the decision of who to ask. When the system handles routing well, users stop thinking about AI models entirely.
Voice changes how people use AI. Typing encourages short, precise prompts. Voice encourages thinking out loud, exploring half-formed ideas. The tool becomes a thinking partner, not a search engine.