Skip to content
ChatGPT logo

AI

ChatGPT integration

OpenAI models sorting email, extracting structured data from documents and summarising operations, connected to the company's systems.

Overview

ChatGPT

ChatGPT is known for chat, but in a business the bigger time saver is using the same models with no chat at all: read an email, decide whether it is a quote request, a complaint or a supplier invoice, and route the data to the right place.

Tejobit connects the OpenAI API (the technology behind ChatGPT, accessed programmatically) to the company's systems. With Structured Outputs, the model always returns JSON matching a defined schema: request type, tax ID, amounts, dates, urgency. The connection validates that result and only then creates the CRM contact, the Notion task or the draft document in invoicing.

Under OpenAI's policy, data sent through the API is not used to train models unless the customer opts in, and is retained for a limited period for abuse monitoring. Personal data is handled under GDPR: only what is needed is sent, under a data processing agreement with OpenAI.

Use cases

What it is for

  • Triage of a shared inbox: quote requests to the CRM, complaints to support, supplier invoices to accounts.
  • Data extraction from supplier invoice PDFs for checking and booking.
  • Categorising bank transactions by expense type for the accountant to review.
  • A plain-language daily summary of sales and payments for the manager.
  • Draft replies to common requests for the team to review before sending.
IN

What the connection listens to

  • New emails in a chosen inbox or label.
  • Documents uploaded to a folder or received as attachments.
  • New records in other systems (transactions, requests, tickets) and scheduled reports.
OUT

What the connection does

  • Classify messages and documents by type, urgency or department.
  • Extract structured fields (tax ID, amounts, VAT, dates, line items) into schema-validated JSON.
  • Write summaries and draft replies in English, Portuguese or other languages.
  • Create or update records in a CRM, spreadsheet or task tool with the result.
  • Route low-confidence cases to a person for review.

How it works

From the request to a working connection

  1. 01

    Request

    The form names the integration, the tools involved and what should happen in each case.

  2. 02

    Review and proposal

    Tejobit confirms what the API allows and sends a proposal with the flow, the price and the timeline.

  3. 03

    Setup and testing

    The connection is set up with the business's own accounts and tested with real cases before it goes live.

  4. 04

    Maintenance and monitoring

    Tejobit hosts the connection, follows up on errors and adapts it when the API changes.

The connection runs on Tejobit's infrastructure. The monthly fee covers hosting, monitoring and API changes.

What to prepare

  • An OpenAI platform account with billing enabled (usage is paid to OpenAI, separate from any ChatGPT subscription).
  • Real samples of the documents or messages involved, to tune the instructions and schema.
  • A decision on which calls the model may make alone and which need approval.
  • A GDPR review of what may be sent to the API, especially health data or data about minors.

For security, do not send passwords, API keys or tokens through the form. Credentials are requested through a secure channel when the setup starts.

Limits

  • Models can be wrong or make up values; anything with tax or financial impact goes through validation or human review.
  • Cost is per token and depends on the model, so long documents or high volumes need estimating first.
  • OpenAI enforces request and token per-minute limits tied to the account's usage tier.
  • Zero data retention and European data residency exist, but depend on OpenAI's approval and setup.

Frequently asked questions

Before you start

Do I need ChatGPT Plus or Team?

No. The connection uses the OpenAI API, which has its own account and billing. A ChatGPT subscription does not include API access and is not needed.

Does OpenAI train on the data sent?

Under OpenAI's current policy, API data is not used for training unless the organisation opts in. It is kept for a limited time for abuse monitoring. Check the wording in force before sending sensitive data.

How do you know the extracted data is right?

In layers: the schema forces a fixed shape, the connection checks tax IDs, totals and dates, and doubtful cases go to a review queue instead of flowing through.

Why ChatGPT rather than Claude or Gemini?

OpenAI models are a solid pick for structured extraction and high-volume classification. For very long documents or careful writing, Claude often does better; for photos and Google Workspace context, Gemini. The provider can be swapped without rebuilding the rest.

Set up the ChatGPT integration?

Describe what should pass between the tools. The reply confirms feasibility, price and timeline.