One project, multiple AI tools — why give them a shared server

You discuss a website with ChatGPT and its structure with Claude, then keep carrying decisions from one chat to the other. You can set up a shared workspace that different AI tools can access.

A personal VPS—a rented virtual server—can host it. It can hold your project documents and run search and other tools. Those tools might prepare files, validate data or launch your service. The models themselves can still run with their providers. Your chat histories do not merge automatically.

Think of it as a shared workshop. At the centre are up-to-date documents. Each connection has its own permissions. Let's start with four steps.

First, create three documents: a brief, agreed decisions and the current status. Record the goal, constraints and next step. Together, they become your agreed reference for the project.

Second, set up secure access. One option is MCP, a standard for connecting AI applications to data and tools. Use HTTPS and authorization. Start with read-only access to a selected folder. Support depends on the application, account and settings. Set up the connection separately in each client.

Third, test it. In the first client, ask it to read the brief and identify the source. Ask the same question in the second client. Compare the document versions. Similar answers without actually reading the file don't prove anything.

When you have more material, add RAG. Before answering, search retrieves relevant passages from authorized sources. We are adding context to the request, not retraining the model. Search can get things wrong too. The system must check source permissions before retrieval.

Fourth, add one narrowly scoped action. For example, preparing an update to the project status. AI suggests: “Brief approved; next step: prototype.” You review the change and confirm it should be saved. This creates a new version and an action record. Ask the second client for the status and source again: it should now read the update.

Keep keys separate from documents, in a protected secret store. Don't paste passwords into chat. If you use a cloud model, the selected passages are sent to its provider. Set up backups and an activity log, cap the number of requests and monitor server and API spending.

Start with three documents and a reading test you can verify. Let the shared workspace grow as your tasks require.