Hold up

Inspiration

AI assistants are smart, but they do not know the context behind a person’s work. We were inspired by the frustration of repeatedly explaining projects, decisions, and past conversations every time we opened a new AI tool.

Hold Up creates a durable personal context layer so agents can understand your work without starting from zero.

What it does

Hold Up turns information from GitHub, Gmail, and imported chats into source-backed Artifacts such as projects, documents, decisions, tasks, and deadlines.

It then exposes that context through MCP, allowing AI agents to retrieve the right information with tools like get_context, search_sources, list_artifacts, and list_conversations.

How we built it

We built the dashboard with React, TypeScript, Vite, and a Node.js API. Context is stored in an encrypted local data store as source records, Artifacts, relationships, and activity history.

We also built an MCP server so AI tools such as Cursor and Claude can use the same saved context. Users can import chat transcripts or public share links, connect apps, review extracted context, and save only the information they approve.

Challenges we ran into

Our biggest challenge was keeping the dashboard, saved context, and MCP server synchronized.

We learned that it is not enough for an Artifact to appear in the UI—it must also be available to an AI agent through MCP. We also had to make deletion reliable so removing a chat Artifact removes its connected source data instead of allowing it to reappear later.

Accomplishments that we're proud of

We are proud that Hold Up makes AI context durable, reviewable, and source-backed.

Instead of treating memory as a hidden black box, users can see what was saved, where it came from, and what their agent can retrieve. We also created a complete flow from importing a chat or connecting an app to asking an MCP-enabled agent about that context.

What we learned

We learned that useful AI memory needs to be durable, source-backed, reviewable, and easy to retrieve.

We also learned that good AI context is not about storing everything. It is about preserving the information that remains useful after the original conversation or source has passed.

What's next for Hold Up

Next, we want to add more connected sources, stronger semantic search, automatic change detection, and better control over what an agent can access.

We also want to make Hold Up easier to use across more AI clients, so a person’s context can follow them between ChatGPT, Claude, Cursor, and future agent tools.

Built With

• codex
• context
• github
• mcp
• react
• typescript