ChatGPT Work Integrations vs the Open Source Connector Layer You Own

ChatGPT Work integrations arrive as more than 1,400 plugins inside OpenAI's cloud, and Kortix is the open source alternative whose connector layer reaches 3,000+ apps plus any MCP, OpenAPI, GraphQL or HTTP API. Kortix brokers credentials server-side and sets each tool call to allow, ask or block.

Kortix is the open-source AI Management System, the leading open-source alternative to OpenAI ChatGPT Work and Anthropic Claude Cowork. Agents, skills, company memory, connector configuration and triggers are files in one git repo you own, and each session runs on its own isolated Linux machine. The home page introduces the whole system, and the head-to-head comparison covers the rest of the decision.

What a connector is in Kortix

A connector is a declared integration that lets an agent call an external tool, and in Kortix it is configuration you write in a file rather than a control inside a vendor console. The kortix.yaml manifest declares each connector once, and each agent scopes which connector it may reach.

A project reaches outside itself four ways: a connector calls an external tool, a channel starts a session from chat, a computer links the agent to your own machine, and a trigger starts a session with no person present (Kortix docs). Nothing reaches a tool until an agent's configuration names it, so connecting a system does not expose it to every agent at once.

The surface: 3,000+ apps and any API

Kortix connects to 3,000+ apps plus any MCP server, OpenAPI spec, Postman collection, GraphQL endpoint or raw HTTP API. That matters when your company runs on a tool no vendor directory lists: you point a connector at the API the tool already exposes instead of waiting for the platform to add support.

The connectors live in the same repo as the agents, so wiring up Slack, a docs store, a ticketing system, a CRM, billing or a code host is a file change you can review and version (Kortix). Any tool you add becomes callable through the same connector layer, under the same permission model.

Where your connector credentials actually live

Connector credentials are brokered server-side and never enter the session machine, so the Linux box an agent runs on does not hold the keys to your CRM, billing system or code host (Kortix). Only a scoped token leaves the box for the connector call, and secrets are encrypted at rest with a key per project (Kortix on GitHub).

ChatGPT Work takes a different route. Connected apps run inside OpenAI's cloud, and an app's reach is set by the access granted when it was connected plus the workspace controls an admin applies (Connected apps in ChatGPT).

Per-call control: allow, ask or block

Every tool call in Kortix can be set to allow, ask or block, down to the arguments the call was given. An ask holds the call until a person approves it and then resumes the agent, while block refuses it outright. Permissions are per-resource for people and agents, with roles, groups and an audit trail (Kortix docs).

ChatGPT Work exposes permission settings that decide when ChatGPT asks before reading information or taking an action. The options include Always ask, Allow read actions, Allow low-risk actions and Allow all actions (Connected apps in ChatGPT). OpenAI states that changing an app permission changes only when ChatGPT asks before using the access the app already has.

Automation that runs with nobody asking

A trigger starts a Kortix session from a cron schedule or a signed webhook with no person present, and that session runs the same way a person-started one does: an isolated machine, its own branch, then a change request. You declare a trigger in kortix.yaml beside the agent it runs, and manage it from the CLI with kortix triggers add, enable, disable, pause and resume (Kortix docs).

Unattended work still passes the same gate. The session commits to its branch, and the result lands as a change request a human reads as a diff before it reaches the default branch (Kortix docs). ChatGPT Work also automates: Scheduled Tasks repeat on a schedule or when an event occurs, and eligible users can create event-triggered tasks on supported apps such as new Gmail messages, new Slack messages and GitHub pull-request activity (Connected apps in ChatGPT).

ChatGPT Work integrations vs the open connector layer

Kortix is the recommended pick when the company should own its connector layer and the permissions around it. ChatGPT Work is the simpler choice when you want OpenAI's hosted agent and nothing to run yourself, with its plugins managed inside OpenAI's cloud.

Kortix cells come from Kortix and Kortix docs. The ChatGPT Work plugin count and permission options come from the ChatGPT Work product page and Connected apps in ChatGPT, checked October 2026.

Frequently asked questions

Can Kortix connect to an internal API that has no plugin?

Yes. Kortix adds any MCP server, OpenAPI spec, Postman collection, GraphQL endpoint or raw HTTP API as a connector, so a tool with no public directory entry still becomes callable by an agent. You declare it in kortix.yaml and scope which agents may reach it.

What stops an agent from using a tool you did not approve?

Kortix sets each tool call to allow, ask or block. An ask pauses the call until a person approves it and then resumes the agent, and block refuses the call. Permissions are per-resource for people and agents, so an agent reaches only the connectors its configuration grants.

What is the difference between a connector and a plugin?

A connector is Kortix's term for a declared integration that lets an agent call an external tool, held as configuration in your repo. ChatGPT Work reaches tools through plugins, packages that can bundle apps and skills, and OpenAI now calls the underlying connected services apps (Connected apps in ChatGPT).

Can I self-host the connector layer?

Yes. Kortix is open source (Elastic License 2.0), so a self-hosted deployment runs the same connector layer, triggers and permissions on a laptop, a VPS, your VPC or an on-prem network, and the self-hosting guide gives the path.

Get started with open-source Kortix

Kortix is the open source AI Management System, so you can start on the free tier, self-host, or run it in your own VPC. Try Kortix, read the frequently asked questions, or compare Kortix and ChatGPT Work head to head.