ASMI - An AI chatbot with a face — your coding agent ships it

Hi Product Hunt 👋 I'm Stian, a solo developer in Norway. I built ASMI because every "AI avatar" or chatbot widget I evaluated either locked you into a vendor runtime — your widget dies when their backend does, your data flows through their servers, your costs scale per-visitor on their pricing — or made you hand-roll the whole face / state-machine / expression layer from scratch. What it is, plainly: ASMI builds chatbots with a face. A visitor types a question, your avatar answers from a knowledge base you control, and its expression shifts as the conversation does. You design that avatar in a visual editor; your coding agent wires it into your own site. Build avatars once, run them anywhere. ASMI splits the problem cleanly: 🔹 Design-time — a visual editor at broen.tech/apps/asmi. You shape the chatbot: face and expressions (multi-wardrobe sets with event scheduling), brand voice (seven sliders), the conversation state machine (XState-compatible, structured guards), a company knowledge base, and an awareness module (nine holiday calendars including Norwegian, business hours, time-of-day greetings). No code. 🔹 Implementation — in your own repo. You paste a one-paragraph briefing into your AI coding tool (Claude Code, Cursor, Lovable, v0, Replit, Cline, base44). It calls ASMI's MCP server, fetches the avatar plus a ~1000-line embedding guide tuned to your framework, and writes the integration. After that, your site owns it. What's new in v1.1 (June 2026): - Six demo chatbots across six industries — pediatric dental (live on a Lovable site today), urban mobility, museum education, dev-tools support, hotel concierge, sustainable fashion. Each built end-to-end by a coding agent following the MCP flow. - Consistency by construction: every face and animation frame is QA-checked against the locked identity and auto-regenerated when it drifts. - A knowledge-base crawler that turns your crawled company pages into draft answers you accept one by one. What this buys you: - No runtime lock-in. ASMI going offline doesn't break your live site — the chatbot is a snapshot in your repo, not a live dependency. - Your LLM keys, your costs. ASMI doesn't proxy chat traffic. - Your data stays on your stack. - Two public MIT npm packages. The runtime has zero dependencies. - A verify_deployment MCP tool — your coding agent re-runs the canonical happy-path I recorded in the editor against its own implementation and gets a pass/fail checklist to iterate on. Test-driven integration instead of vibe-checking. Honest about what it doesn't do: - No "drop in one script tag" path for serious integrations. The architecture needs a coding agent (or a developer following the embedding guide). - I don't host the LLM. You bring OpenAI / Anthropic / Gemini. Genuinely free tier: build, simulate, connect knowledge — no card. Deploying to a live site and sharing to the gallery are the moments worth charging for, so they sit on the paid tiers. See it: - 2-minute walkthrough: broen.tech/apps/asmi/developers - Live chatbots in the wild: broen.tech/apps/asmi/showroom I'm specifically after: bugs, pricing thoughts, missing framework integrations, and MCP-client compatibility reports. I'll be in the comments all day. 🙏