How to keep one client's context out of another's work, what contractors should and should not reach, and what a per-seat stack costs when headcount moves with the contracts
Aug 18, 2026 · 13 min read
An agency should organise its AI work by client rather than by person or by model, which means one container per account with its own files, instructions, chats, member list and retention rules. Client separation matters more here than the size of any shared memory, because an undifferentiated store can carry one account's material into another's work. The eight workspaces compared on the same criteria below are Playgram, WorkLLM, nexos.ai, Langdock, TeamAI, Aymo, Magai and TypingMind.
Headcount is the second thing that makes agencies different. A team may need ten people on an account during a campaign and six once the freelancers leave, while a per-seat plan keeps charging until somebody removes the seat. As an example, four single-vendor team plans came to about $101 per person a month at July 2026 list prices1, 2, 3, 4. Each fully provisioned contractor adds the same amount again.
This guide sets out what client separation should look like in practice, how the eight products document it, and what to test with a real contractor before confidential material moves. It ends with a two-week pilot run on your own client files and your normal approval process.
You own the client relationship and the material that must not cross between accounts.
Freelancers join for a campaign and need one client's context and nothing else.
You add and remove access every month and answer for what the AI bill did.
A solo consultant or a pair on one model, with no reusable client context.
An agency changes shape every quarter, and the four layers below each behave differently when the people on an account come and go.
Separate subscriptions make the bill follow licensed headcount rather than productive work, so an agency that needs ten people during a campaign keeps paying for ten after six freelancers leave. As an example, four single-vendor team plans came to about $101 per person a month at July 2026 list prices, taking ChatGPT Business at $25, Claude Team at $25, Gemini Business at $21 and Grok Business at $301, 2, 3, 4. Read it as an illustration rather than a rate, because a cheaper mix is easy to assemble. The commitment shape then decides how quickly you can undo it, since a fixed-term plan usually allows additions during the term while reductions wait for renewal, and Claude Team prorates changes while enforcing a five-member minimum32, 2.
Every account repeats the same steps in a different tab. Somebody copies the client brief into each tool, uploads the same deck and spreadsheet again, moves one model's output into another for review, and puts the result into the project-management system. Then a contractor joins and the whole sequence runs once more for them, which is why the subscription is only part of what this costs.
A chat can hold a great deal of client knowledge while belonging to one person and one conversation, so the account manager, the writer, the analyst and the freelancer each rebuild the background separately. Chat history is a poor shared knowledge system for this, because nobody can tell which conversation holds the approved tone, the current offer, the legal restriction or the client's decision. A general automatic memory creates the opposite risk, which is that material from one account influences work for another unless the scopes are explicit.
Personal subscriptions leave an agency unable to say who can see a client's files, whether a departing contractor kept a copy, which model received confidential documents or who is consuming the budget. Nobody can stop usage before an overage, and ownership of chats and prompts is unclear the day somebody leaves. What client work needs is both collaboration and compartmentalisation, and a consumer account offers neither.
Five checks for a team whose work is divided by client rather than by department. The second one is where most products in this category are weakest.
The team should be able to use different models for research, writing, analysis and review without buying every provider for every user. Changing model partway should keep the conversation and its attachments, because an account team that loses the brief at each stage rebuilds it instead.
Check each product for image generation, video generation, web research with citations, document work for briefs and decks, spreadsheet work for performance data, code review for technical accounts, and chats that leave nothing behind. No product here documents all seven at the same depth.
Each account needs its own project or workspace with its files, instructions, chats and member list, so a teammate continues without re-uploading the brief. A new contractor should receive the approved files, templates and previous decisions through that container rather than from a colleague.
A contractor should reach only the clients they serve, and removing them should not delete the agency's project history. An admin needs usage by person, model, client and period, plus budgets and hard limits that act before the bill rather than a report that arrives after it.
Headcount moves with the work, so price the shortlist at peak seats, at average active users and at real use. Then check the commitment shape, since a fixed-term plan often lets you add licences mid-term while reductions wait for renewal, and a light contractor should not cost what a lead does.
The multi-model workspaces an agency is most likely to weigh up, on the same criteria and to one standard. Where a vendor does not document something, the cell says so.
This table compares multi-model team workspaces with each other. The single-vendor plans a workspace replaces are priced further down, under 'Priced per seat', and are not rows here. Pricing is the lowest-priced paid plan that covers five users, at the monthly rate, so a product whose entry plan holds fewer than five people is shown on the plan that holds them. Each cell cites the page that documents that cell rather than one pricing page per row. Figures checked August 2026, and cells marked 'Manual test required' could not be confirmed from public documentation.
The same products on the criteria that decide daily use: what each one does besides chat, what it connects to, what an admin can see and cap, and where your data goes.
These criteria decide daily use more than the model list does, and vendors document them very unevenly. 'Not publicly documented' means the official sources checked did not state it, and it does not mean the feature is absent, so read those cells as questions to put to the vendor. Checked August 2026.
The published per-seat price of each major single-vendor team plan, billed monthly. Every contractor who needs all four arrives at this figure and leaves it behind for a while.
Buying all four for one person came to about $101 a month at July 2026 list prices. Read that as one example stack rather than a going rate, because a team can assemble a cheaper mix and the four plans do not buy the same amount of use. Commitment shape matters as much as the rate for an agency, since a fixed-term plan usually allows additions during the term while reductions wait for renewal. Figures checked July 2026, and the estimator further down runs the comparison on your own numbers.
Two of these follow the contracts and two follow the calendar, which is why an agency bill rarely matches the work in the same month.
Six ways to organise AI across accounts, ordered by how well each one keeps one client's material away from another's.
Each account gets its own project or workspace holding the files, instructions, chats and member list. nexos.ai Projects keep uploads, searches, conversations and instructions for their members, and Langdock shares projects with named users or groups under owner, editor and user roles11, 33.
Best for: Agencies running several accounts with changing teams.
Strengths
Trade-offs
One store the whole agency draws on, which is efficient for method and dangerous for client material. WorkLLM documents organisation memory without explaining per-client boundaries inside an agency tenant, and Aymo claims team memory without documenting its controls7, 21.
Best for: Agency method and templates rather than client material.
Strengths
Trade-offs
Standardise on a single vendor and separate clients using whatever containers it offers. Administration is simple and model choice is whatever that vendor ships, which shows up when a client asks for work another model does better.
Best for: Agencies whose work fits one model family.
Strengths
Trade-offs
Buy the plans each discipline prefers and accept the per-seat total. As an example, all four plans priced above came to about $101 per person a month at July 2026 list prices, and a fully provisioned freelancer adds the same1, 2, 3, 4.
Best for: Agencies that genuinely need each vendor's own tools.
Strengths
Trade-offs
Everyone expenses their own account and the client context lives wherever each person put it. It costs the least to start and the most to unwind when a client asks where their material is held.
Best for: A solo consultant or a pair with one client each.
Strengths
Trade-offs
Engineers build the client isolation, retrieval, access control, analytics and offboarding themselves. That becomes defensible when a contract dictates data location or a client wants the system on infrastructure they control.
Best for: Technical agencies with client-hosting requirements.
Strengths
Trade-offs
A live brief for one client, with the approved material in that client's container and a contractor who joins the account rather than the agency.
Everything stays inside one client container, so a contractor invited to this account sees the brief and the approved work and nothing from any other client. The account lead corrects the research before production starts and sends thin work back, and the approved language, the superseded drafts and the decisions stay behind when that contractor's access is revoked.
A system that remembers more is not automatically better here, because a fact being useful does not mean every account team should be able to reach it.
A context window is the material a model considers during one conversation, and it fills or falls out of use. Chat history is a record of previous chats that nothing retrieves for you. Memory is stored outside the window and pulled into later chats, which is where the client boundary belongs.
Some keep history alone. Some hold project knowledge filled in by hand, which Magai and Langdock document. Some learn automatically and keep it personal, as Langdock and TeamAI do. Some make it available across the organisation, which WorkLLM claims and Aymo markets without its controls33, 34.
Ask whether memory is tied to a person, a project, a client, a group or the whole company, and whether retrieval respects the permissions of the folder a file came from. Aymo suggests one organisation per client, and Langdock recommends a project per client34, 33.
Check that an owner can see what was saved, correct it and delete it, that deleting a source chat removes anything derived from it, and that a sensitive session can stay out of history entirely. A contractor should also be able to contribute to one account without reaching any other.
A vendor-neutral plan that ends with a real contractor being added and removed, which is the test a demo never covers.
List every account, who works on it, which files and folders belong to it and what the contract says about where the material may be held. Add the current AI subscriptions with their owners, commitments and renewal dates, and mark the seats that outlived the campaigns they were bought for.
Set up a project for one account with its permitted brand guide, audience notes, approved work, product facts, reporting template and current brief. Write the instructions with it, covering tone, prohibited claims, required citations, file naming and who approves what.
Research the brief on a web-connected model, have the account lead correct the assumptions, then continue in the same thread on a writing model and check that it inherited the research, the files and those corrections. Produce the assets the account actually needs and store them back in the same container.
Invite a real freelancer to that one client and check what they can reach, then confirm they cannot see other accounts or the agency library. Revoke the membership at the end, and check that the project history stays and that their chats or assets can be transferred to somebody who remains.
Delete a source chat and check whether anything derived from it also goes, then confirm who can inspect and correct a stored entry. Read usage by person, model and client, check that spending can be capped in advance, and read the training and residency terms against what the client contract requires.
An agency should choose on client isolation, context handover, contractor access and how the cost behaves when headcount moves, rather than on the number of models advertised. Project-based systems are usually safer for client work than one undifferentiated company memory. Per-workspace or usage-sensitive pricing also fits fluctuating teams better than a four-product per-seat stack, provided the credits, overages and external charges are understood.
Four limits apply. Plans with similar names do not carry similar capabilities. Stored context helps only when its scope, permissions, correction and deletion are clear. Several vendors describe memory or team context without documenting what an administrator can inspect, delete or restrict. And prices and plan caps move faster than a contract cycle.
So the decision comes from a controlled test with your own client files, a real contractor and your normal approval process. Add somebody to one account, check what they can reach, then remove them and see what stays behind. That sequence answers more than any feature table.
Upgrade as needed, and only pay for what you actually use
Save ~17% with the annual plan
Pro
Perfect for small and medium teams
Unlimited users & infinite memory
Multi-LLM chats
Granular access control to models
EU data residency
Ultra
Best for large, growing teams
Unlimited users & infinite memory
Multi-LLM chats
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
Enterprise
Get in touch
For organizations with advanced needs
Unlimited users & SSO
Priority Support
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
30-days money back guarantee
Playgram will automatically choose the most cost-efficient model suitable for the task. It will be chosen by users in approximately 80% of requests. Your models for the remaining 20%:
If you bought each separately: