Yes — Perplexity connects to SAP Business One over MCP through TEKAI, and has been supported since TEKAI launched. You keep your own Perplexity subscription, and questions are answered from live ERP data under your existing permissions, read-only by default.
Yes, and it was one of the first. Perplexity has been supported since TEKAI launched, and it is the assistant a lot of teams tried first.
The connection is made over MCP rather than through any native integration, because none exists. A bridge sits between the ERP and the model: it interprets the question, enforces the asking user’s authorisations and logs what was asked.
Perplexity uses that bridge directly with your own subscription — the same route as Claude and ChatGPT, and one step simpler than the assistants that require an app or connector to be registered first.
The native TEKAI bridge, with the customer’s own subscription. Same bridge, same controls as every other assistant: a permission-scoped endpoint, read-only by default, per-user provisioning, every question logged.
There is nothing to migrate when moving between assistants. The endpoint and the permission model belong to your SAP Business One installation, not to whichever model is asking — MCP is model-agnostic, and one TEKAI endpoint serves all six.
That is a more useful property than it first appears. It means the decision you are making today is which assistant to start with, not which assistant to marry.
Customer-shaped questions are where this connection tends to be pointed, because they are the ones that come with a second and third question attached.
Churn signals are the clearest example: which accounts have slowed against their own ordering baseline rather than against a company average. That is a comparison nobody builds a standing report for, and exactly the kind of thing you want to interrogate rather than receive.
From there the natural follow-ups are what changed, when, and whether complaints or service calls preceded the drop. Because every answer comes from the same database under the same permissions, the follow-up costs nothing — which is the actual difference between an assistant and a dashboard.
The same connection also handles margin by customer with below-floor sales flagged, quotation conversion by owner, and receivables ageing when someone wants the detail behind a total.
A good first question is one where you already know the answer. That is how you find out whether to trust the second one.
Permission-scoped endpoint, read-only by default, per-user provisioning, every question logged. Answers are grounded in your data and your data is never used to train the model.
Write actions are opt-in and confirm before posting. Running read-only for the first months is the normal pattern and a sensible one — the value is in the answers long before it is in the automation.
The architecture behind all of this is documented on the SAP Business One AI agent page.
The single biggest predictor of whether an AI project sticks is which question you point it at first.
Pick one where you already know the answer. That sounds like a waste of a demo and is the opposite: it is the only way to calibrate how much to trust the second answer, which you will not be able to check.
Make it one sentence. If the question needs a paragraph to express, it is really three questions, and the assistant will answer a blend of them. “Which customers are over their credit limit with open orders?” works. “How is our receivables situation looking?” does not.
Choose something with a consequence attached. A question nobody acts on produces a nice demo and no change in behaviour — and the project quietly dies at renewal because it never earned a place in anyone’s week.
And write down the number before you start. If you cannot say what receivables ageing looks like today, you will not be able to say whether it improved, and the value of the whole exercise becomes a matter of opinion.
One more test, which sounds unserious and is not: pick something a person currently asks a colleague for. Those requests are the real evidence of an unmet need, and they are already costing somebody an interruption. A question nobody has ever had to ask is usually a question nobody wants answered.
Perplexity, Claude and ChatGPT share the simplest route: the native bridge with your own subscription, nothing to register first. Gemini, Microsoft Copilot and Grok each need their own app or connector step, which adds a one-time task and changes nothing afterwards.
The choice is a question about your team, not about SAP Business One. If you want the model SAP itself now builds Joule on, see Claude on SAP Business One; if your organisation runs on Google, see Gemini on SAP Business One; and if you want the connector route, see Grok on SAP Business One.
All six are compared, with the cost of building the bridge yourself, in the guide to connect AI to SAP Business One.
Yes. Perplexity has been supported since TEKAI launched, using the same bridge as every other assistant: a permission-scoped endpoint, read-only by default, provisioned per user, with every question logged.
It suits follow-up questions particularly well — the ones that arrive after the first answer and would otherwise mean a second report. For posting documents back into SAP Business One, any of the six can do it once write actions are enabled.
No. Answers about your ERP are grounded in your data, which is not used to train the model. Access is read-only by default and scoped per user against your existing SAP Business One authorisations.
TEKAI is ₹5,000 per company per month in India, or $150 per company per month globally, and you bring your own AI subscription. Pricing is per company rather than per user, so adding a colleague does not change the figure.
Yes. There is nothing to migrate when moving between assistants, and nothing stopping two teams using different ones — the endpoint and the permission model belong to your SAP Business One installation, not to whichever model is asking.
Thirty minutes on your own customer history, live.