Freelancers: keep each client's preferences across missions and AIs

Illustration scenario. The freelancer and the clients described here are fictional. There is no testimonial and no statistic: just a typical workflow.

Freelance work is mostly context. Each client has its own vocabulary, brand guidelines, contacts, sensitive topics and past decisions. The deliverable matters, but what brings a client back is not having to repeat everything.

Take an independent copywriter and web developer. She works with several clients in parallel, sometimes over years. Some missions last a week, others come back every quarter, and between two of them she forgets details that the client, naturally, remembers perfectly. She writes with Claude, prepares quotes and meeting notes with ChatGPT, and builds websites in Cursor.

Before: one notes file per client, never up to date

She keeps a file per client. At the start of a conversation she pastes the relevant part. When a decision is made on a call or in a meeting, she promises herself to write it down, and does so half the time.

Anyone who works for several clients knows the result: copy written in another client's voice, a colour rejected six months ago turning up again in a mockup, a question already settled being asked again and making you look like you don't follow. And each AI has its own memory, when it has one, mixing every client together.

After: one project per client in a shared memory

With Pryzm, each client becomes a project. Claude, ChatGPT and Cursor connect to the same memory over MCP. What is said in one is available in the others, filed in the right place.

• Each client's preferences. Tone, formality, banned words, text length, invoice format.
• Mission decisions. The hosting choice, the call on a page, the scope approved by email: with the date and the reason.
• Open items. What is waiting on the client, what is promised for the next delivery.
• What has already been tried. A rejected idea isn't pitched again three months later as if it were new.

When she resumes a mission after a break, the AI finds the right client's context, not whatever was discussed last. Search combines keywords and meaning: “the client selling hiking gear” is enough, even without the name. The dashboard, organised by project, lets her review or correct what was saved for a given client.

Example requests

• “Write this month's article for the estate agency, following their tone guidelines.”
• “What did we decide with the ceramics shop about the home page?”
• “Draft notes from today's meeting and save the decisions to the client's project.”
• “Before we launch this site, list the items still open with this client.”

The third request is the habit that changes everything: what comes out of a meeting goes into memory straight away, instead of waiting for a quiet moment that never comes. The fourth one is the safety net before a delivery: the AI goes through the client's project and brings back what was promised, what is still waiting on them, and what was explicitly ruled out. It doesn't replace your own checklist, but it catches what slipped through it.

Confidentiality: what to know

Working for clients means handling information that isn't yours. A few points:

• Servers are in Germany.
• Each account is encrypted with its own key (AES-256-GCM). One limit to know: during a session, the server decrypts memories in memory so it can search them. It is not end-to-end encryption.
• A message that starts with ## is never saved. Useful for something a client told you in confidence that must not resurface.
• You decide what is kept: everything can be reviewed and deleted from the dashboard.

If a contract forbids you from sharing a client's data with a third-party service, that applies to an AI memory too. In that case, keep that client out of the shared memory.

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

• One project per client, with a stable name. That's what stops one client's preferences from bleeding into another's.
• Save while it's fresh. Right after a call, one sentence to the AI is enough: “save that the client approved the mockup, except the pricing page”.
• Clean up at the end of a mission. Archive or delete what you won't need, especially if the client asks you to.
• Keep credentials out. Client passwords and access keys don't belong in a memory: keep them in a password manager.

Getting started

1. Create an account: free up to 100 memories, enough to try it with one or two clients. Then Solo at €9.99/month.
2. Connect your tools with the connector URL: Claude, ChatGPT, Cursor, Le Chat, Perplexity, Claude Code or Gemini CLI.
3. Paste the “main memory” instructions: the AI checks Pryzm at the start of every conversation and saves what matters.
4. Tell it which client you're working for: it files memories in the right project.

Everything is explained on the connection page.