ContextCollab

Private AI Chats with Selective Shared Team Context

ContextCollab is a hackathon product built around a tension every AI-assisted team feels: people want private, exploratory AI conversations, but teams still need a way to share only the useful parts. The app keeps each collaborator's private chat local by default, then lets them intentionally push selected context into a realtime shared branch.

The goal is not to expose everyone's entire prompt history. It is to create a privacy-aware collaboration layer where teammates can preserve their own thinking space while still pooling key discoveries, summaries, and reusable context.

During Microsoft GitHub Copilot Hackathon * NYC Tech Week, the team prototyped a browser-based workflow with two clients, private chat panes, a shared context branch, invite-link collaboration, and realtime fanout using Ably. OpenAI API calls power private responses, shared AI responses, summarization, and pull-context behavior.

The page below documents the shipped UX, the technical architecture, and the in-room hackathon moments behind the prototype.

Prototype Flow

The prototype flow keeps privacy as the default, then makes collaboration explicit. A user chats privately with Copilot, selects the messages worth sharing, summarizes and pushes them into a shared branch, and teammates can pull that shared context into their own private thread.

UX and Technical Architecture

Key Features

• Private chat as the default workspace for exploratory AI conversations.
• Selective sharing controls so users decide exactly which private context becomes team context.
• Realtime shared context branch scoped by room and powered by Ably sync.
• Summarize-and-push workflow that turns selected messages into compact shared cards.
• Pull-context action that lets collaborators bring shared context back into their own private chat.

Team Collaboration

• Collaborated as a team to define the core product idea: private AI chats with selective realtime shared context.
• Discussed and refined the feature set together, including private chat, shared context branch, invite links, summarize-and-push, and pull-context behavior.
• Worked through the system design together across browser clients, shared room state, OpenAI response modes, and Ably realtime sync.
• Vibe coded the prototype collaboratively during the hackathon, moving quickly between UX decisions, implementation, testing, and demo polish.
• Prepared the case-study framing after the hackathon to document the team's product thinking and build process.

Hackathon Moments

What I Would Build Next

The next iteration would focus on stronger realtime collaboration and permission controls: persistent rooms, authenticated invite links, granular shared-card visibility, better conflict handling between local and Ably-backed sync, and evaluation hooks so teams can mark whether shared AI summaries were useful, incomplete, or misleading.

Acknowledgements

Great building alongside the ContextCollab team: @Shreyas Kulkarni, @William Liu, @Emily Simmons, and @Joshua Solomon.