ChatGPT Agent Kit: AgentKit, ChatKit, and the Agents SDK

"ChatGPT agent kit" is not one product. Depending on where someone found the term, they may mean the OpenAI Agents SDK for building multi-agent systems in code, Coinbase's AgentKit for building blockchain-capable AI agents, or ChatGPT Work — OpenAI's consumer agent experience inside the ChatGPT app. Each is a different tool with a different audience.

The confusion matters because picking the wrong one wastes time. This article explains what each tool does, who it is for, and when to use it. Whatever tool you choose, keeping shared project decisions and a handoff brief outside those tools is where something like fenbs fits in: the work context that survives when you switch tools or hand off to a teammate.

ChatGPT Work — the consumer agent in the ChatGPT app

ChatGPT Work (formerly called ChatGPT Agents, renamed in July 2026) is the agent feature inside the ChatGPT web and mobile app. You give it a goal in plain language and it completes a multi-step task on your behalf: browsing the web, running code, filling in forms, managing files.

ChatGPT Work is not a developer toolkit. You do not write code to use it. It is a product feature, not an API you call. The right audience is anyone who wants an AI assistant to complete browser-based tasks without writing any code.

• Available inside the ChatGPT app — no separate installation.
• Works through natural language instructions, not code.
• Connects to the web and to file storage within the ChatGPT environment.
• Not the same as the OpenAI API or any developer SDK.

For pricing and plan availability, check OpenAI's pricing page↗ (opens in a new tab) directly — availability changes and quoting plan names or prices here would go stale quickly.

OpenAI Agents SDK — the developer framework

The OpenAI Agents SDK is a Python library for building agentic systems in code. It provides abstractions for agents, tools, handoffs between agents, and guardrails. You import it, define agents with instructions and tools, then orchestrate how they coordinate.

This is the tool for teams building AI-powered products, not for people who want to use an AI assistant. It calls the OpenAI API on your behalf. Using it requires an OpenAI API key, Python knowledge, and familiarity with how tool-calling and multi-agent patterns work.

• Install with pip install openai-agents and import in Python.
• Define agents with a name, model, instructions, and a list of tools.
• Orchestrate handoffs: one agent passes work to another based on the task.
• Add guardrails to validate or filter agent inputs and outputs.
• Open-source; the repository is at github.com/openai/openai-agents-python↗ (opens in a new tab).

Coinbase AgentKit — for blockchain-capable agents

Coinbase AgentKit is an open-source toolkit for building AI agents that interact with blockchain networks and the Coinbase Developer Platform. It gives agents the ability to check balances, send transactions, deploy contracts, and call DeFi protocols.

AgentKit works with several AI frameworks including the OpenAI Agents SDK. It is not a ChatGPT product — it is a third-party toolkit. The audience is developers building applications where an AI agent needs to act on-chain. If your use case has nothing to do with blockchain or cryptocurrency, this is not the tool you are looking for.

You can find Coinbase AgentKit at github.com/coinbase/agentkit↗ (opens in a new tab). The pricing model for the underlying Coinbase Developer Platform is on Coinbase's own developer pages.

How the three compare

Keeping shared context when you use these tools

All three tools share a limitation: the context they hold is private to a session or an agent instance. When you hand off work to a colleague, switch from ChatGPT Work to a custom agent, or migrate from one framework to another, the conversation history stays behind. The project decisions, the rules the team agreed on, and where things stand are not automatically available in the new tool.

fenbs does not replace these tools, but it does fill that gap. It keeps the shared work record — decisions, tasks, and progress — outside any one tool, so each agent or session starts from the same known state. When a connected AI agent reads the fenbs context at the start of a session, it reads the same rules and current tasks that a teammate's agent would read.

fenbs does not connect to ChatGPT Work directly or integrate with the Coinbase Developer Platform. It works with the AI apps listed on Integrations: Claude, Claude Code, ChatGPT (developer mode), Cursor, Codex CLI, Gemini CLI, GitHub Copilot, and Devin Desktop.

When the tool changes, the work should not have to restart. fenbs keeps Where we left off, active tasks, and Decisions and rules in one place. Any connected AI app reads the same record, so a switch between tools or teammates is a session start, not an explanation session.

How to do this with fenbs

1. Create a free account and add your project. Write the key decisions and rules that every agent working on this project should follow.
2. Connect a supported AI app to fenbs over MCP from Integrations. For Claude or Claude Code, sign in with a browser and approve the connection.
3. For ChatGPT in developer mode, use the MCP setup described in the integrations guide. Which ChatGPT plans support developer mode is on OpenAI's help pages.
4. At the start of each agent session, ask the agent to read the project context from fenbs before starting work.
5. Before the session ends, ask the agent to save progress and any new decisions so the next session — on any tool — starts from the same point.

The agent harness topic hub has more guides on connecting agents to shared context, including chatgpt agent mode for how the ChatGPT app feature works and microsoft copilot agents for the Microsoft ecosystem.