HomeServicesOpenAI Integrations

OpenAI Consulting and ChatGPT Integration Services

There are three ways to connect ChatGPT to a business system, and they differ in who can use them, what they can reach and how they are billed: a connected app inside your ChatGPT Business (formerly ChatGPT Team) or ChatGPT Enterprise workspace, a custom MCP server or plugin that an admin publishes to the workspace, and an integration built on the OpenAI API that runs in your own systems. OpenAI consulting at NetSys is choosing the route for each job, building it with least privilege and approvals, and proving it against acceptance criteria before anyone depends on it.

See Custom AI Applications

By The NetSys Group · Published October 17, 2026 · Editorial policy

The short answer

ChatGPT integration services connect a company's ChatGPT workspace, or the OpenAI API, to the systems where its work lives: SharePoint, Outlook, Teams, Google Drive, Slack, a CRM, an accounting system or a database. NetSys delivers them in three forms, chosen per job: connected apps and Company Knowledge inside the workspace (read access within each user's permissions), custom MCP servers and plugins published by an admin for read and approved write actions, and API integrations built on the Responses API or Agents SDK for work outside ChatGPT. Every route starts with the data it may read, the actions it may take and who approves them.

Closest related page: Custom AI application development. That page covers full applications built on Claude and Microsoft tooling with their own screens and hosting; this page covers connecting ChatGPT and the OpenAI API to systems you already run, from a workspace app to an API integration.

Who OpenAI consulting and integration services are for

Companies on ChatGPT Business or Enterprise whose staff keep pasting content out of SharePoint, Outlook and the CRM into a chat window. Teams that want ChatGPT to answer from company documents without copying them anywhere. Businesses with a workflow that should run on the OpenAI API inside their own systems (an intake form, a document pipeline, a reporting job) and need someone to build it with the boundaries set first.

The vocabulary changed in 2026 and it matters for the plan you buy. What OpenAI used to call connectors are now apps; plugins package skills and apps or a custom MCP server into a workflow; synced connectors were retired between August and October 2026, so access is real-time through apps. OpenAI-built apps are search-only; writing into a system needs a custom MCP server, a workspace agent or an API integration. A ChatGPT workspace and an API Platform organization are separate systems with separate billing, and a Business subscription includes no API usage.

Route, boundary, proof of concept, acceptance, support

Each integration starts with the job and the system, then the route: a connected app when read access inside existing permissions is enough, a custom MCP server or plugin when the workspace needs approved write actions, an API integration when the work runs outside ChatGPT. The boundary is written before the build: which identity it acts as, which data it may read, which actions it may take, who approves writes, where the logs go and what it may cost per month. A proof of concept runs against real data with a small group and acceptance criteria agreed in advance. Only then does it reach the wider team, with monitoring, cost caps and a monthly review under managed AI services.

What OpenAI and ChatGPT integration services include

Route selection and OpenAI consulting

Workspace app, MCP server or API build, decided per job.

Each job mapped to a route with its reach, audience, approval model and monthly cost
A plan check: what Business and Enterprise allow for apps, custom MCP servers, workspace agents and role-based control
API architecture when the work leaves ChatGPT: the Responses API, the Agents SDK, or ChatKit for an embedded chat, with Agent Builder avoided because OpenAI has scheduled its shutdown for November 30, 2026
Build-or-buy advice, including when Microsoft 365 Copilot or Claude already covers the job

Connected apps: SharePoint, Outlook, Teams, Google Drive, Slack

Read access inside the permissions people already have.

Apps enabled per workspace from the approved list, with the approval policy set to ask before any write
Microsoft Graph permissions granted by your Entra administrator for SharePoint, Outlook and Teams, with the consent recorded
Company Knowledge turned on where staff should get answers from connected sources within their own permissions
Enterprise content restrictions applied where available: SharePoint site collections, sensitivity-label exclusions, Google shared-drive allow lists

Custom MCP servers and plugins

The route for systems OpenAI has no app for, and for write actions.

A custom MCP server for your CRM, accounting system, ticketing tool or database, with read tools first and write tools behind approval
Published by an owner or admin to the workspace, with the tool list refreshed and republished after each change
Authentication through OAuth with refresh tokens or scoped keys held by the server, never in a prompt
A plugin that packages the server with skills for a defined workflow, shared to the team that needs it

API integrations with business systems

Work that runs in your systems, billed per token.

Intake, document processing, classification and reporting jobs built on the Responses API
Agents built with the Agents SDK where the steps vary, with handoffs, guardrails and tracing
Project-level API keys, spend limits, and zero data retention or modified retention requested from OpenAI where the data requires it
Hosting in your Azure or cloud environment, with logs you own

Security boundaries

Set in writing before the first connection.

Identity: an app runs as the signed-in user; an MCP server or API integration runs under its own account with least privilege
Data: sources and fields listed, sensitive sites and labels excluded, test data separated from production
Actions: read by default, writes named one by one and gated by an approval, kill switch documented
Prompt injection tests on documents, emails and web content before launch, and the logs reviewed with you

Evaluation, usage costs and support

Proof of concept, acceptance criteria, then the bill watched.

A proof of concept on real data with a small group and acceptance criteria agreed first
Credit usage in the workspace and token usage on the API estimated before the build and capped where the platform allows
Acceptance sign-off, deployment to the wider team and an employee guide
A monthly review of logs, failures, cost and vendor changes under managed AI services

Why NetSys

Why businesses bring NetSys in for OpenAI integrations

Tell us which system you want ChatGPT to reach, what the job is, whether it needs to write anything, and which plan you are on. A NetSys engineer will name the route, the boundary and the proof of concept, and quote the work after a scoping call.

Who does the work: meet the NetSys team on our About page.

We run the Microsoft 365 tenants these integrations connect to, so Entra consent, SharePoint permissions and sensitivity labels are handled by the same engineers
Vendor-neutral across OpenAI, Microsoft Copilot and Claude: the route is chosen by the job, and NetSys resells nothing
The same boundary-first method as our published agent and custom application work: identity, data scope, approvals and logs before the build
Delivered AI work: a Copilot rollout with custom agents for 18 people, Claude apps and reporting roundups for a lean team, and a Copilot and Claude rollout for 25 employees with connector access controls
Acceptance criteria and a proof of concept before anyone depends on the integration
Month-to-month support afterward, from one office in Brooklyn, in business since 1998

From our published case studies

Custom Apps, Dashboards and Financial Reporting on Anthropic Claude

A lean team needed more output without more hires. NetSys set data-handling rules first, then built custom apps, dashboards and reporting roundups on Claude.

The same boundary-first approach on another platform: access and data-handling rules agreed before any source was connected, then apps and reporting roundups built on that footing.

Read the case study

Three ways to integrate ChatGPT with business systems

Read from OpenAI's help center and developer documentation on October 17, 2026. The route can change as OpenAI adds apps, so each job is re-checked at scoping:

	Connected app in ChatGPT	Custom MCP server or plugin	API integration
What it is	An app (SharePoint, Outlook, Teams, Google Drive, Slack, GitHub, Box, Notion and others) enabled in the workspace	A server you publish to the workspace that exposes read and write tools for your systems; a plugin packages it with skills	Software built on the Responses API or Agents SDK that runs in your own systems
Who can use it	Workspace members, within the permissions they already hold in the source	Members the admin shares it with; on Business only owners and admins create and publish	Whoever your application serves, inside or outside the company
What it can do	Search and read; OpenAI-built apps are search-only as of October 2026	Read and, where you build it, write, with approvals set per action	Anything you build: intake, classification, drafting, reporting, agents
Identity	The signed-in user, after they authenticate with the source	The server's own credentials, scoped to the tools you expose	A service identity with least privilege in each system
Billing	Workspace seats and credits	Workspace seats and credits, plus hosting the server	API tokens, billed per model, plus hosting
Admin control	Enable or disable per workspace; Enterprise adds per-role assignment and content restrictions	A frozen tool snapshot until an admin refreshes and republishes	Project keys, spend limits, retention options and your own logs

Where common integration searches land (as of October 2026)

The apps named here appear on OpenAI's official pages as of October 2026; availability depends on plan, region, workspace and role. Where no app exists, the route is a custom MCP server or an API integration:

System	Route today	Note
SharePoint and OneDrive files	The SharePoint app, or Company Knowledge for answers across connected sources	Needs Microsoft Graph consent from your Entra administrator; OneDrive is not named as its own app on the OpenAI pages we opened
Outlook email and calendar	The Outlook Email and Outlook Calendar apps	Enterprise can exclude emails by sensitivity label
Microsoft Teams	The Teams app for search, and @ChatGPT inside Teams for Business and Enterprise	@ChatGPT needs workspace credits, and a Microsoft Global Administrator or Privileged Role Administrator to set up
Slack	The Slack app, and @ChatGPT in Slack	Event-triggered tasks in ChatGPT Work can respond to new channel messages where an admin allows it
Google Drive, Gmail and Google Calendar	The Google Drive app (with Docs, Sheets and Slides actions), and the Gmail and Google Calendar apps	Drive actions are on by default in Business and off until enabled in Enterprise
GitHub, GitLab, Box, Notion, Asana, Trello, Airtable, Figma	Apps are listed for each on OpenAI's pages	Check write support per app; OpenAI-built apps are search-only, so writes need a custom MCP server or a workspace agent
CRM (HubSpot, Salesforce), accounting, e-commerce, ticketing	A custom MCP server or an API integration	No HubSpot, Salesforce or Shopify app appeared on the OpenAI pages we opened; we build the server or integration against the vendor's API
Excel, Word and PowerPoint	The ChatGPT for Word, Excel and PowerPoint extensions	A Microsoft 365 admin must allow the add-in
Zapier, Make and other automation tools	An API integration, or a workflow step that calls the API	Covered under AI workflow automation; a workspace app is not the route

Step-by-step guides: ChatGPT and SharePoint: setup, permissions and plan limits · ChatGPT Outlook integration: permissions and safe email workflows · ChatGPT CRM integration: Salesforce, HubSpot and custom APIs · AI workflow automation with Power Automate, Zapier and Make.

Security boundaries we set before connecting

Every route gets the same seven boundaries, written into the design and checked at acceptance:

Boundary	How we set it
Identity	Connected apps act as the signed-in user after they authenticate with the source. MCP servers and API integrations get their own account or app registration with least privilege, never an employee's login.
Data scope	Sources, sites, folders and fields listed per job. On Enterprise, SharePoint site collections, sensitivity labels and shared drives restricted in the workspace settings; elsewhere, restricted at the source.
Actions	Read by default. Each write named, built as its own tool and set to ask for approval. A kill switch (disable the app, unpublish the plugin, revoke the key) documented and tested.
Untrusted content	Documents, emails and web pages tested for hidden instructions before launch; write tools never triggered from unverified content.
Retention and training	Workspace and API data is not used for training by default; API abuse-monitoring logs are kept up to 30 days; zero data retention or modified retention is requested from OpenAI where the data requires it and stated per route in the design.
Cost	Credit limits per seat type in the workspace; project spend limits on the API; usage reviewed monthly.
Review	Permissions documented so your security owner can see what each integration reaches and revoke it without calling us.

How an integration project runs

The order is the same for a workspace app, an MCP server and an API build:

Scoping: the job, the system, read or write, the audience and the plan you are on.
Route and boundary: app, MCP server or API build; identity, data scope, actions, approvals, retention and cost written on one page.
Access: Entra or Google consent, source permissions, API keys or OAuth set up by your administrator with us.
Proof of concept: built against real data for a small group, with acceptance criteria agreed before it starts.
Testing: the test set, prompt injection checks and a review of the logs.
Acceptance and deployment: sign-off, rollout to the wider team, an employee guide and a register entry.
Support: a monthly review of logs, failures, cost and vendor changes under managed AI services.

Deliverables, owners and boundaries

Each deliverable has an owner on both sides and a boundary written into the scope:

Deliverable	NetSys	Your company	Boundary
Route and boundary document	Writes it	Approves identity, scope, actions and approvers	One job per integration; writes only where named
App and consent configuration	Proposes and configures it	Your Entra or Google administrator grants consent	Only the apps the job needs; approvals set to ask
Custom MCP server or API code	Builds, tests and documents it	Owns the repository, credentials and hosting account	Hosted in your environment; you keep the code
Acceptance test results	Runs and records them	Signs off	No accuracy or uptime guarantee from NetSys; OpenAI's terms govern its service
Cost estimate and caps	Estimates usage and sets the caps	Approves the limits and pays the vendor	Vendor usage billed to you by OpenAI
Employee guide and register entry	Writes them	Publishes them	Kept in your workspace

What moves the quote for an OpenAI integration

NetSys publishes no price list. Configuration and builds are quoted after a scoping call, monthly support runs month to month, and these are the inputs that move the number:

Factor	Why it moves the price
Route	Enabling an app is configuration; an MCP server or API build is development
Systems and write actions	Each system and each write adds tools, tests and approvals
Identity and permissions state	Clean permissions in the source are quick; oversharing must be fixed before an app sees it
Plan	Enterprise role assignment and content restrictions add configuration; Business has fewer controls to set
Data sensitivity	Regulated data adds retention decisions, exclusions and sometimes a different route
Volume	API token usage follows how much the integration reads and writes, estimated before the build

OpenAI's own prices are on its pricing pages and change; the API cost question below quotes them as of October 2026.

Staff pasting SharePoint and CRM content into ChatGPT?

Tell us the systems and the plan you are on. We will say which route gives them the data inside their permissions, and what it would take to let ChatGPT write back safely.

Common Questions

OpenAI Integrations FAQs

What does an OpenAI consultant do?

An OpenAI consultant helps a company decide how to use OpenAI's products (ChatGPT Business or Enterprise, connected apps, workspace agents and the API) and then builds and secures the pieces. At NetSys that means choosing the route for each job, setting identity, data scope, approvals and cost caps before anything connects, building the app configuration, MCP server or API integration, proving it against acceptance criteria and supporting it monthly. We are independent: NetSys holds no OpenAI partner, certification or reseller status, and you buy seats and API usage from OpenAI directly.

What is a ChatGPT integration?

A ChatGPT integration is any connection between ChatGPT and another system. As of October 2026 there are three kinds: connected apps inside a Business or Enterprise workspace (what OpenAI used to call connectors), which search and read within each user's permissions; custom MCP servers and plugins published by an admin, which can also write where you build that in; and integrations on the OpenAI API, which run in your own systems and are billed per token. Synced connectors were retired between August and October 2026, so access is real-time. Source: OpenAI help center, connected apps and custom MCP servers in ChatGPT, read October 17, 2026.

Can ChatGPT connect to SharePoint and Outlook?

Yes, on Business and Enterprise through the SharePoint, Outlook Email and Outlook Calendar apps, and through Microsoft Teams. Some actions need Microsoft Graph permissions granted organization-wide by a Microsoft Entra administrator; approval in ChatGPT alone does not enable them. ChatGPT respects the permissions each user already has in Microsoft 365, and Enterprise can restrict SharePoint site collections and exclude content by sensitivity label. OpenAI-built apps are search-only, so writing back to SharePoint or sending mail needs a custom MCP server, a workspace agent or an API integration. Our SharePoint and Outlook posts cover the setup step by step. Source: OpenAI help center, administering connected apps, read October 17, 2026.

How do we connect ChatGPT to our CRM, such as HubSpot or Salesforce?

Through a custom MCP server or an API integration. No HubSpot or Salesforce app appeared on the OpenAI pages we opened on October 17, 2026, so we build an MCP server against the CRM's API with read tools (look up an account, list open deals) and, where you want them, write tools (log a call, create a task) that ask for approval. On Business only owners and admins can create and publish it, and ChatGPT then uses a frozen snapshot of its tools until an admin refreshes it after a change. For CRM work that should run without a person in ChatGPT, an API integration or an agent is the better fit. Our CRM integration guide compares the options.

What does the OpenAI API cost?

It is billed per million tokens, by model, with no seat fee. As of October 2026, OpenAI's pricing page lists gpt-6-astra at $10.00 input, $1.00 cached input and $50.00 output per million tokens, gpt-6.1-sol at $2.00, $0.10 and $10.00, and gpt-6-luna at $0.10, $0.01 and $0.50; the Batch API is 50% of standard for most models, cached input is about 90% below standard, and data residency or FedRAMP endpoints add about 10%. A ChatGPT Business subscription includes no API usage. We estimate the monthly token cost of each integration before building it and set project spend limits. Source: OpenAI developer documentation, API pricing, read October 17, 2026.

Is data sent through the OpenAI API used for training, and how long is it kept?

Not for training: OpenAI's developer documentation says data sent to the API is not used to train or improve its models unless you explicitly opt in. Retention: abuse-monitoring logs are kept for up to 30 days unless the law requires longer, and the Responses API stores application state for 30 days by default when storage is on. Zero data retention and modified abuse monitoring are available subject to OpenAI's prior approval through sales. We state the retention terms for each route in the design, and we never promise zero retention on a route that does not have it. Source: OpenAI developer documentation, your data, read October 17, 2026.

What are ChatGPT Enterprise connectors, and how do they differ from Business?

The same apps, with more administrative control. In Business, apps are enabled by default and an admin decides which stay on and sets an approval policy per app. New Enterprise workspaces start with a selected set enabled and new apps disabled until an admin turns them on; Enterprise can assign apps per custom role, restrict content by SharePoint site collection, sensitivity label or shared drive, restrict which account domains may connect, and see GPT conversations in the Compliance Platform. OpenAI's pricing comparison also lists a connector registry for Enterprise only. Source: OpenAI help center, administering connected apps, and the business pricing comparison, read October 17, 2026.

How is this different from custom AI application development?

Scope and ownership. This page connects ChatGPT, or the OpenAI API, to systems you already run, from enabling an app to an API integration that performs one job. Custom AI application development builds a complete application (its own screens, retrieval, hosting and reporting) on Anthropic Claude and Microsoft tooling for work that needs more than a connection. Many clients need both: an app in the workspace for staff questions, and a built application for the monthly reporting roundup. Our custom AI applications page covers that build.

Sources and technical references

Keep Reading

Guides on this topic

Explore More

Connect ChatGPT to the systems where the work lives, with the boundary set first.

Tell us the system, the job and your plan. We will name the route, write the boundary and scope a proof of concept with acceptance criteria.

See Custom AI Applications 845-203-3914