Solo founder: switch from Claude to ChatGPT without re-explaining your project

Illustration scenario. It is inspired by how Pryzm is built by its founder, Aurélien, but it contains no figures and no anecdotes: it's a typical workflow, not a testimonial.

A solo founder carries everything: product, code, infrastructure, website, support. AI assistants fill part of the missing team. But each one works with its own memory, and the project's continuity depends on one person copying context from tool to tool.

This scenario follows a typical week of building a SaaS product with three tools: Claude to think and write, ChatGPT for a second opinion, and Claude Code in the terminal to write and fix code.

Before: the project lives in your head

On Monday you discuss with Claude how to isolate each customer's data in the database. You compare two options and pick one, for specific reasons. On Tuesday, in Claude Code, you implement it. The terminal knows nothing about yesterday's discussion: you summarise the decision from memory and forget one of the reasons.

On Thursday you ask ChatGPT to review your backup strategy. It suggests a sensible approach… that contradicts Monday's choice, because it has never heard of it. You spend ten minutes restoring context, then wonder whether you ever wrote down what was left to do.

None of this is serious on its own. Repeated every day, it has three effects: time lost re-explaining, weaker answers because they are based on partial context, and decisions that quietly come undone.

After: one memory, connected everywhere

With Pryzm, all three tools connect to the same memory server over MCP: one URL, one OAuth sign-in per tool. When a decision is made in a conversation, the AI saves it. When another AI needs it, it finds it.

• Technical decisions follow the project. Monday's choice with Claude, and its reasons, are there when Claude Code opens the repository on Tuesday.
• Fixed bugs stay documented. The cause of an incident and the fix applied are saved; if the symptom comes back, the AI picks up the trail instead of starting over.
• Todos don't get lost between tools. What's left to do is noted when it comes up, whichever assistant is open.
• Preferences apply everywhere. How you write commit messages, the tone of the website, what you refuse to add to the product: said once, respected by all.

Search combines keywords and meaning, so you don't need the exact term you used three weeks ago. The dashboard, organised by project, shows you what was saved — you can correct or delete it.

What you actually ask your AIs

A few typical requests, as you'd type them:

• “What did we choose to isolate customer data, and why?”
• “Which bugs did we fix in the login module this month?”
• “What's left to do before launch?”
• “Review this backup plan against our infrastructure decisions.”

The last one says it all: the review becomes useful because the AI starts from your real constraints, not from an imaginary project.

What shouldn't be saved

Not everything deserves to be kept. A message that starts with ## is never saved: handy for thinking out loud, a rough draft, or something you don't want resurfacing later.

On the data side, servers are in Germany and each account is encrypted with its own key (AES-256-GCM). One limit worth knowing: during a session, the server decrypts your memories in memory so it can search them. It is not end-to-end encryption, and we'd rather say so.

When it isn't useful

If you only use one AI, its built-in memory is probably enough. Shared memory pays off from the second tool onwards — even more when one of them is a coding agent that never sees your chats.

How it works, without jargon

MCP (Model Context Protocol) is an open protocol that lets an AI call external tools. Pryzm provides a few of them: search the memory, save a memory, get a project overview. You add a single URL to each assistant, sign in once with OAuth, and each tool gets its own token, which you can revoke at any time.

The “main memory” instructions do the rest: the AI checks Pryzm at the start of the conversation and before answering anything that depends on your context, then saves on its own what is worth keeping. You don't have to think about “saving”: you work, and the memory fills up.

Good habits

• Name your projects clearly. One name per product or workstream is enough; the AI files memories in the right project when you say which one.
• Give the reason with the decision. “We're keeping PostgreSQL” is worth less than “we're keeping PostgreSQL because the rest of the infrastructure depends on it”.
• Review the dashboard now and then. An abandoned decision should be corrected, or it will keep steering answers.
• Ask for a status update at the start of a session. A question like “where is the project at?” puts everyone, you included, on the same page.
• Keep secrets out. Passwords and API keys don't belong in a memory, shared or not: keep them in a password manager.

Getting started

1. Create an account: free up to 100 memories, then Solo at €9.99/month.
2. Add the connector URL to each tool (Claude, ChatGPT, Claude Code, Cursor, Gemini CLI…).
3. Paste the “main memory” instructions from the connection page: they tell the AI to check Pryzm at the start of every conversation and to save what matters.

The step-by-step guide is on the connection page.