If yours is a team that uses ChatGPT, you can connect it to your CRM.
The surface the connection will rest on is live today: a REST API authorized with OAuth2, describe that returns the object schema, composite that carries 25 sub-requests in a single request, and bulk endpoints that handle up to 200 records per call. We build it as an integration project whose scope is defined in the discovery call.
The team works in ChatGPT, the record sits in another window
If your team no longer uses ChatGPT as an experiment — the subscription has been bought, the team has settled in, most of the work runs in that window — then the CRM is sitting somewhere else during the same day. The mechanism is this: with no defined link between the two windows, a human does the transfer, and the moment it flows across, the origin of the record is lost. Three outcomes we have seen in our own projects and in our buying conversations:
- Data is moved by copy-paste. The summary is produced in ChatGPT, then pasted into the CRM by hand; where the record came from and who produced it are lost.
- In the procurement meeting someone asks, “Is your CRM AI-powered?” Saying “yes” is easy; once they ask with which model, under what authorization and where the data goes, the answers get blurry.
- IT asks something else: which data goes to which provider, who turned it on, can it be revoked? The answer to that question can be left to the contract stage — that is, to the worst possible moment.
This page answers all three openly — and on one of them it answers against our own interest. First what there is, then what has been measured, then what has not been measured. A page that does not write the third one teaches it to the buyer on installation day.
First the sentence against our own interest: this product's assistant runs on Claude
There is no out-of-the-box module for ChatGPT, and we are not going to say there is. The connection is built as an integration project — with REST API, OAuth2, webhooks and import/export — whose scope is defined in the discovery call.
Now the thing that really needs saying. Rapi, the assistant inside the product, runs on Anthropic Claude — not on GPT. That is why you will not read the sentence "GPT-powered CRM" on this page; writing it would help us in search results, but it would not be true. With the same care: our 18-tool MCP package, downloaded from the Setup screen, is defined today as a Claude connector. We cannot say it has been tested with any other AI client; and because we have not tested it, we do not say it. Setting out these two sentences up front ends the "but the website said so" conversation before it can start after purchase.
So what is actually there? The way to connect ChatGPT to the CRM runs through the product's open surface, and that surface is live today: the product ships with 183 business objects and each one is published in the REST API with its own data and schema endpoint — you can read the field structure from outside with describe. Alongside it there is authorization with OAuth2, composite carrying 25 sub-requests in a single request, and bulk endpoints that handle up to 200 records per call. That is exactly what an AI client needs in order to be able to look at the CRM.
Which connection route is used on the ChatGPT side — depending on the client's capabilities at the time and on your subscription type — is chosen together during discovery. We can say that this surface really is usable from outside on the strength of a measured fact: in one live customer installation, a third-party ERP integrator reads and writes records over this same REST API. We do not give a name because we do not have permission to publish it; what we are saying is a measured fact, and this is a single installation, not a list of cases. We have described the product's own assistant and the MCP package on the Claude integration page, the same framework inside an ERP on the Netsis integration page, and our general approach on the integrations page.
Thursday 14:25 — a rep is preparing for a meeting. Once the connection is in place, this is what changes:
-
1
Data is not copied, it is called
Once the connection is in place, the client reads customer, contact and opportunity records from the API. The copy-paste step in between disappears; which record an answer rests on becomes clear.
-
2
Authorization comes from OAuth2, not from a person
Access is granted with OAuth2; no user password is shared. You define the scope, and you can take it back the way you granted it — from a single screen.
-
3
The schema is not tracked by hand
The object schema is read from the describe endpoint. When you add a new field to the CRM, the connection code does not have to chase after it; what it reads changes along with it.
-
4
Write access is a separate and deliberate decision
Reading and writing are not the same thing. Which objects get write access is written out one by one in the scope; we do not grant access that is open to everything by default.
-
5
Every change made leaves a trace
Writes made over the API land in the field-level change history: when which field moved to which value. An assistant coming from outside does not skip the audit trail.
-
6
The relationship with the provider stays in your name
You use your own ChatGPT subscription; the contract on that side is between you and your provider. We do not sell tokens, credits or models in between.
The surface the connection rests on — the part that is live today
183 business objects, each with its own endpoint
The product ships with 183 business objects and each one is published in the REST API with its own data and schema endpoint — from lead to quote, from invoice to task. You can read the field structure from outside with describe.
Authorization with OAuth2
Access is granted with OAuth2. No integration is built by sharing a user's password; authorization is taken back the way it was granted.
Schema reading with describe
The object schema is read from the API. Custom fields the tenant adds without writing code also appear in that schema; the connection stays current.
composite and bulk endpoints
Several operations are sent in a single request — composite carries 25 sub-requests; bulk endpoints handle up to 200 records per call.
Change history covers writes as well
Writes made over the API land in the field-level change history. The fields the AI touched can be seen one by one afterwards.
A separate database per tenant
Your CRM data is held in a separate database that belongs to you; access arriving over the API also stays inside that boundary.
What was measured, what was not?
| Topic | Where it stands today | What that means |
|---|---|---|
| The assistant inside the product | Rapi runs on Anthropic Claude; it reads and writes CRM records | We make no claim of a “GPT-powered CRM”; we write down which model we run on, by name |
| MCP package — 18 tools | It is defined today as a Claude connector and is downloaded from the Setup screen | We have not tested the package with ChatGPT; and because we have not tested it, we do not say it |
| REST API + OAuth2 | Live; each of the 183 business objects that ship out of the box is published with its own data and schema endpoint, and describe and composite are supported | This is the surface a ChatGPT connection will rest on, and it works today |
| The connection route on the ChatGPT side | It depends on the client's capabilities at the time; we have no measured record of it | The route is chosen together during discovery; we do not ship an out-of-the-box GPT or plugin |
| Where the data goes | You use your own ChatGPT subscription | The contract on that side is between you and your provider; we do not sell tokens or credits in between |
What we have verified
How is the connection brought live?
-
1
Discovery
We work out how you use ChatGPT today, which subscription type you are on, and which records you want to look at in the CRM. We recommend that your counterpart on the IT and compliance side is in this call.
-
2
Scope and authorization table
Which object gets read access and which gets write access is written out one by one. Which provider the data will go to, and what that means in terms of transfer abroad, sits in the same document.
-
3
Configuration and a limited start
OAuth2 access is defined and the connection is opened with a narrow scope. We recommend limiting the initial scope to reading only; write access is widened after the team's flow has been seen, and at your decision.
-
4
Training, auditing and go-live
The team is shown how to follow the assistant's work through the change history. We do not commit to a timeline before discovery; after the discovery call we give you a date, not a range.
Our limits: what we do not promise on this page
AI is an easy headline to overstate, and an expensive one. We write down what we do not promise, by name.
- We do not say “GPT-powered CRM”. Rapi, the assistant inside the product, runs on Anthropic Claude. That sentence would help us in search results; we do not write it because it is not true.
- We do not say that our MCP package works with ChatGPT. The package is defined today as a Claude connector and we have not tested it with another client. If you want to connect another assistant, the route is REST API and OAuth2; that is a separate piece of work whose scope is defined in discovery.
- We do not ship an out-of-the-box GPT or plugin. There is no listing from us in the store.
- We do not sell an autonomous agent. The number of steps that can be taken in a single request in the product's own assistant is limited — at most eight. We do not offer a system that runs on its own through the night.
- We do not promise that your stock will stay in sync, that the order → invoice → collection chain will flow by itself, or production/MRP automation. Connecting an assistant does not change those three. We have written the reason for all three, and what can be done instead, in one place: the three things we do not promise.
AI and consent: the product's own assistant is off by default; unless you turn it on, your CRM data is not sent to any AI provider. When you connect your own ChatGPT client, that decision is yours, and we define its scope in writing at setup: which object, which permission, which user. Seen from Turkey this is a transfer abroad and falls under Article 9 of Law No. 6698; we establish its basis together during the contract process. Our sub-processor list is on our security page.
Data location: your data is held in the European Union — the database in Helsinki (Finland), files in Frankfurt, backups in Falkenstein (Germany). We have no component in Turkey and we have no region-selection feature. The detail is on the security page and in the KVKK privacy notice.
Certification: we do not hold an ISO 27001 or SOC 2 certificate, and we will not claim one until we do. We fill in your enterprise security assessment forms and describe the architecture in writing.
Plan prices are published; scope comes out of discovery
The scope of a ChatGPT connection — and any fee for it — comes out of the discovery call. We publish plan prices, and separately priced items such as Rapi AI credits, on a single page.
Three plans, a scope table and separately priced items, on one page.
Frequently asked questions about the ChatGPT integration
Can we connect ChatGPT to Rapitek CRM?
Is Rapitek CRM a “GPT-powered” CRM?
Do you provide a ready-made Custom GPT or plugin?
Does your MCP package work with ChatGPT as well?
Which model does the product's own assistant run on?
Will we be using our own ChatGPT subscription?
If the connection is set up, can ChatGPT write records to the CRM?
How do we audit the changes the assistant makes?
Does our CRM data go to OpenAI?
Is the AI assistant on by default?
Why is the product's assistant not ChatGPT?
How long does setup take, is it included in the plan price?
Where is our data held?
Can we see it before buying?
Let us spend 30 minutes on which record we open with which permission
In the demo, let us show the API surface, OAuth2 authorization and how write operations land in the change history, on your own scenario; and let us narrow the scope together.
The authorization table is drawn up in discovery: which object, which permission, which user.
