Switched Chats? How to Keep Your Agent Working on the Same Project
You spend an afternoon with an AI agent working out what a project needs. The next day you open a new chat, or try a different agent, and you are starting over: re-explaining goals, re-pasting notes, re-checking what was already ruled out.
The problem usually isn't the agent. It's that the project lives inside one conversation.
Put the project outside the conversation
A project that survives a chat switch needs a home of its own, with at least:
- Requirements, versioned, so you can see what changed.
- Materials you already have: specs, notes, constraints.
- Research that came back, with its sources and what still hasn't been verified.
- Decisions and the reasons behind them, so nobody re-argues a settled choice (more on why decision logs matter).
- A clear next step.
ThingWeave is a private project workspace shared by you and your agent. These pieces live in one project, so a new session can pick up from what is already saved.
How an agent reads and writes the project
ThingWeave exposes your project to agents over MCP (Model Context Protocol). You connect an MCP-capable agent to https://thingweave.com/mcp, sign in with Google and explicitly approve which project it may use. After that, the agent can read the project's current state, submit requirements and research results back to it, and record plan decisions with their reasons. Setup steps are on the agent setup page. Limits and privacy details are in the workspace and agent access docs.
Research is carried out by your own agent. ThingWeave keeps the results, sources and open questions together so the next session doesn't start from zero.
Four starting points, one workspace
- Just an idea: you have a goal and need a first step.
- Existing materials: plenty of documents, no thread to follow.
- A prototype: one version built; what should you test next?
- A paused project: you changed chats or agents and want to continue — here is a checklist for restarting a stalled project.
What ThingWeave does not do
ThingWeave is in pilot. Hardware is the first concrete scenario. Engineering sign-off, procurement, manufacturing and physical fulfilment are not offered. Research outputs are drafts with sources and open risks; they are not engineering certification.
We are in pilot, and you can sign up and try it. Start your first project →
FAQ
What is AI agent project memory?
Keeping a project's requirements, sources, research and decisions outside any single chat, so an agent in a new session can read where things stand and carry on.
Does ThingWeave work with my agent?
It works with agents that can connect to a remote MCP server. You connect to https://thingweave.com/mcp and approve the authorization yourself.
Is my project private?
Projects belong to your own account, which you sign in to with Google. An agent can only use a project after you approve the connection, and you can revoke that access.
Does ThingWeave run research by itself?
No. Your agent does the research, and ThingWeave stores the results with their sources and status.
How much does it cost?
ThingWeave is in pilot, and payments are not currently enabled.