AI for consultants with multiple clients: keep them apart

You finish a proposal for one client at ten, answer another client's questions at eleven, and after lunch you rewrite a third client's homepage. Each of them pays a different day rate. Each has their own payment terms, their own tone of voice, and a list of things you must never say. You keep all of it in your head, and mostly that works.

Your AI tools keep none of it. Every new chat in ChatGPT, Claude or Copilot starts from a blank page. Either you paste the context in again, or you get a draft written for nobody in particular and fix it yourself.

This guide is for independent consultants, freelancers and fractional leads who work for several clients at once. It covers what goes wrong when your AI tools can't tell your clients apart, why chat memory doesn't solve it, and a setup that does. Most of it works with the tools you already have.

What goes wrong when your AI doesn't know the client

The mistakes are small, and that is what makes them expensive. They look right until the client spots them.

• Last year's rate in this year's quote. The AI took the figure from an old proposal you pasted in back in March.
• Net-30 for a client who is on net-45. Your standard terms, not the ones that client negotiated.
• A banned word in the client's own copy. Their brand guide says never "solutions". The draft says it twice.
• One client's details in another client's document. A product name, a price or a contact from one account turns up in the next account's draft.

Then there is the time you spend preventing them: pasting the rate card again, explaining the client's tone again, reading every draft line by line. Workday's research put a number on that kind of rework: nearly 40% of the time AI saves is lost to correcting errors, rewriting content and checking output from one-size-fits-all AI tools (Workday, January 2026). That survey was of employees at large companies, not solo consultants, but anyone who has fixed a generic AI draft will know the pattern.

Why chat memory doesn't fix it

ChatGPT, Claude and the rest can now remember things between chats, and projects let you attach files. Both help. Neither was built for one person working for many clients.

Memory keeps what you said, not what you agreed. It holds facts from your chats. Your rate card, your contracts and each client's brand guide live in Google Drive or Notion, and the current version of each is the one that counts.

Memory that spans every chat can carry one client into another. A detail from Monday's work for one client can surface on Tuesday in work for a different one. Some tools let you limit memory to a project, which helps, if you set it up for every client.

Attached files go stale. A project keeps the copy you attached. Raise your rates in January and every project still quotes December's prices until you replace each file by hand.

Every tool keeps its own copy. If you write in Claude, research in ChatGPT and build in Cursor, you set up the same context three times and keep all three in step yourself.

Four rules for AI that works across clients

Whatever tools you use, the fix comes down to four rules.

1. Keep one source of truth for each client

You already have it: the contract, the agreed rates, the brand guide, the notes from the kick-off call. Keep them as documents in one place, a folder per client, rather than retyping the important parts into prompts. A prompt is a copy, and copies drift.

2. Separate your defaults from each client's exceptions

Write your house rules down once: your standard day rate, your payment terms, how you lay out a proposal. Then record only where each client differs. When the two disagree, the client's rule wins. That's how you already think about it, and it's the order your AI tools should follow too.

3. Let only the current version count

When a rate or a term changes, change the document, and make sure the old version stops being used. In client work, the most common AI mistake isn't an invented fact. It's a real fact that has gone out of date.

4. Give the AI only what the task needs

More context is not better context. Researchers testing language models found that accuracy can drop by more than 20% when the information that matters sits in the middle of a long context (Liu et al., 2024). Pasting every document for every client into every chat makes the right answer harder to find. Send the rules for the client you're working on, and the ones that bear on the task.

A setup you can start today

You can follow most of those rules with tools you already pay for.

1. Write a house sheet. One page: your rates, your standard terms, your voice, and the things you never do. Keep it in Google Drive or Notion.
2. Write a sheet for each client. Where their rates or terms differ from yours, their tone, who signs off, the words they ban, and decisions already made. One page each.
3. Give each client a project in your AI tool. Attach your house sheet and that client's sheet, and nothing from any other client.
4. Date every sheet. When something changes, update the sheet and replace the copy in every project that uses it.

This works. Its weak point is step 4: you are the one keeping every copy current, in every tool, for every client. Miss one, and the AI quotes the old rate with complete confidence.

How Vythos does it for you

Vythos is built for this job: one person, or a small team, working for several clients with AI tools.

• Connect your work. Link Google Drive or Notion, or upload files. It all becomes memory: searchable, and private to you.
• Name your rules. Point Vythos at your rate card, your terms and each client's brand guide. You decide which documents are rules, and which client each one is for.
• A client's rules beat your defaults. Where a client pays a different rate or works to different terms, their rule comes first for their work.
• Edit the document, and the rule follows. Change your rate card in Drive or Notion and Vythos keeps in step. Only the version in force is sent, and earlier versions are kept.
• Use it in the AI tools you already have. Connect Claude, ChatGPT, Cursor or any tool that accepts MCP connectors, once. It gets your rules and memory before client work. Or ask Vythos directly, and see which documents each answer came from.

The Free plan covers one client, with no time limit and no card. Solo is $29 a month for all your clients, each kept apart. Connecting your AI tools is free on every plan.

Questions consultants ask

Can I keep clients apart in ChatGPT or Claude without another tool?

Partly. A project for each client, holding only that client's documents, keeps them apart. You still update every copy yourself when something changes, and you repeat the setup in each AI tool you use.

Is my clients' information used to train AI models?

Not by Vythos. Vythos doesn't train or fine-tune any AI model, on your data or anyone else's. The AI it uses runs on Microsoft Azure in the EU, where requests aren't used to train models. Check the data settings of each AI tool you connect, too.

How long does setting it up take?

About fifteen minutes for your first client: connect your documents, name the ones that are rules, and connect your AI tool. The docs walk through each step.

I run a small agency, not a one-person business. Does this still apply?

Yes, and it matters more, because several people are working from the same rules. Agencies of 3 to 20 people can also apply for the $1,000 Agency AI Grant.