A team chat that keeps track
Ask Mesh a question or share an update. It pulls decisions and tasks into the panel on the right. The starting chat is a sample; your messages use live AI and are not saved after you leave.
Alright team, let's lock the core decisions for the MVP before EOD. Top priority: what does the AI participant actually DO in a conversation? Summarise, extract tasks, or something more proactive?
I think the minimum viable behaviour is: listen to everything, surface decisions and action items in a side panel, and only speak up when asked. No unsolicited replies. That's enough to prove the value without being annoying.
Agreed. From a design perspective, the AI should feel like a silent note-taker. Present but not intrusive. The side panel is the output, not the chat thread itself.
Got it. Recording that as a core design principle:
Decision: The AI participant operates in a "listen-first" mode. It surfaces structured context (decisions, tasks, links) in the side panel and only responds in the chat when explicitly addressed.
This is now tracked in Room notes on the right.
Perfect. Next: stack. Luca, what are you thinking?
Next.js + TypeScript for the frontend and API layer. Monorepo keeps it simple. For real-time we'll start with SSE, upgrade to WebSockets when we need bidirectional. OpenAI GPT-4o for the AI core, with the option to swap models later.
Budget estimate for infra in the first 3 months: around $4,000.
Works for me. I'll have the design system (tokens, component library, the base Figma file) ready by Friday. I'm scoping the messaging UI to feel like Linear meets iMessage.
Capturing a few things from that:
Decision: Tech stack is Next.js + TypeScript (monorepo), SSE for real-time, GPT-4o as the AI model.
Task: Priya to deliver the design system and Figma file by Friday.
Budget: $4,000 estimated for infra over the first 3 months.
Anything to add or correct?
One more: I'll set up the CI/CD pipeline and the staging environment by end of this week. Target is zero-downtime deploys from day one.
Great. On the product side, let's keep the MVP scope tight. Group chat with AI participant, context panel, and a shareable demo link. No auth, no persistence beyond the session for now. Ship fast, get feedback.
Here's the full PRD: /prd
Added that to the panel.
Task: Luca to set up CI/CD pipeline and staging environment by end of week.
Decision: MVP scope is group chat with AI participant + context panel + shareable demo link. No auth or persistence in v1.
Link: PRD spec
You're in good shape. Three decisions locked, two tasks assigned, infra budget set. Ready to build.
Enter to send · Shift+Enter for new line