ChatGPT caps the workspace. ParleHub caps the engagement.

A workspace cap tells you the total. Your partners asked about one client.

Credit limits on the workspace stop the invoice being a surprise, which is worth having. They do not tell you what the Ashworth deal cost, because the four people who worked it also worked six other engagements that month. The allocation gets rebuilt by hand from a report that was organised around people.

• check_circle Budgets set on the project, with per-user and org-wide caps stacked underneath
• check_circle A send that would take the project over budget is blocked before it runs, with a warning at 80%
• check_circle Anyone blocked can request an increase in one click, routed to an admin to approve or reject
• check_circle Analytics group by project, person, provider, or model, and export as CSV under your project code

One vendor's models, or the models you already pay for

ChatGPT Business and Enterprise are a way to buy OpenAI. Ordinary chat is covered by the seat, which is genuinely clean, and the heavier agentic work draws on a credit pool you buy separately. It also ties the workspace and the model together, and a firm that has already signed an enterprise agreement with a provider is paying for the same tokens twice. With us the two are separate purchases: bring that agreement, or use models we host and pay for what you use.

There is a third case neither vendor covers. A good deal of the cheap, capable capacity now comes from labs your firm will never contract with, because a forty-person practice is not a procurement target for anyone and self-hosting an open-weight model is an infrastructure project nobody asked for. We host those models, so routine drafting and summarising can run on something that costs a fraction of a frontier call while the analysis a partner signs stays on the model you trust for it. Spend is tracked per project either way, so whether the cheaper model was good enough stops being an argument and becomes a number on the engagement.

• check_circle Connect OpenAI, Anthropic, Google Gemini, and Azure OpenAI with your own keys, held in Azure Key Vault and never exposed
• check_circle Or skip the keys entirely and use models we host, charged by what you use. Run both at once if you want, on the same ledger
• check_circle That includes open-weight models from outside the big three, which your firm will never sign a contract with and should not have to
• check_circle Azure OpenAI for firms that want inference inside their own Azure subscription
• check_circle Calls run under your firm's own provider agreement, which is the agreement your client questionnaire asks about
• check_circle Switch model per conversation, deliberately, with no automatic routing or silent downgrade

Lock-in is a price you pay in the second year

Credits make the first year simple. The second year is where it costs you, because a competing model will get cheaper, or better at the specific work your firm sells, and the only question that matters then is what it takes to act on that. When the projects, the instructions, the uploaded files, and the searchable history all live in one vendor's workspace, acting on it means rebuilding all of it somewhere else.

ParleHub keeps the provider at the edge. Everything your team builds sits above it and does not move when the model does.

• check_circle Point one conversation at a different vendor's model when a single question needs it, keeping the thread, the project files, and the agent tooling
• check_circle Move a project or the whole organization onto another provider as a settings change; projects, threads, files, budgets, project codes, roles, and the audit trail are untouched
• check_circle Your keys or ours, so moving spend to a cheaper model is a commercial decision rather than a migration project. The same budgets, caps, project codes and ledger apply either way
• check_circle One ledger measures every provider, so the comparison that justifies the move comes from your own usage rather than a vendor's pricing page