What is OpenAI ChatKit?
ChatKit is OpenAI's toolkit for putting an agent-style chat window into your own product. The frontend is a customisable embed, available as React bindings, a JavaScript library or a CDN script, with streaming replies, file attachments, tool-call display, interactive widgets and visible reasoning steps. A Python SDK on the server handles storage, access control and tools, and can sit in front of any agent, including one built with OpenAI's Agents SDK.
There were two ways to run it. The hosted route, backed by an Agent Builder workflow, ends with Agent Builder's shutdown on 30 November 2026; OpenAI now recommends the self-hosted custom server route for all new apps and migrations, and says ChatKit itself remains available.
The libraries are Apache-2.0 on GitHub. chatkit-js has about 1,967 stars and its last release, version 1.9.0, came on 31 July 2026; the Python SDK's last release, 1.6.5, was on 19 May 2026.
How much does OpenAI ChatKit cost?
| Plan | Price | What you get |
|---|---|---|
| ChatKit libraries | Free, Apache-2.0 | |
| Upload storage | $0.10 per GB-day | After 1 GB free per account each month |
No fee for ChatKit itself; you pay model tokens at normal API rates. File and image upload storage costs $0.10 per GB-day after 1 GB free per account each month.
Is OpenAI ChatKit worth it?
Building the chat window is the part of an AI feature that looks trivial and is not. Streaming text, retries, file uploads, showing which tool ran and keeping threads in order add up to weeks of work.
ChatKit hands you that window. You drop a React component or a script tag into your page, and it handles the streaming, attachments and the little cards, called widgets, that let an agent show a form or a product list instead of a wall of text. OpenAI's own examples run from HR onboarding to shopping and troubleshooting assistants.
The plan changed this year. ChatKit launched with a hosted option where Agent Builder ran the agent and you only embedded the window. OpenAI announced on 3 June 2026 that Agent Builder shuts down on 30 November 2026, and the guide now tells everyone to use the custom server route, where your server runs the Python SDK and whichever agent you like.
The catch: that route makes you the backend. You authenticate users yourself, mint a short-lived client secret for each session, store threads and run the agent, and the sample servers check tokens against a hard-coded list of users that the docs warn you to replace. One example even passes a device ID as the user, which other examples say never to do. Activity is slowing too: the Python SDK has not had a release since 19 May 2026.
OpenAI's data rules are less friendly than the rest of its API. Threads on /v1/chatkit/threads are kept until you delete them and are not eligible for zero data retention, and uploads cost $0.10 per GB-day after the first 1 GB each month. API data is not used for training by default.
Our take: ChatKit is a good component and a weak platform. Teams that want OpenAI's look without writing a chat UI will save time; teams hedging across model vendors will find the open alternatives, such as the Vercel AI SDK, carry less baggage.
Where it earns its place
A two-person product team at a B2B SaaS start-up
Ship an in-app support assistant in a fortnight
Embed the React component, run the Python SDK on an existing FastAPI server and point it at an Agents SDK agent with file search over the help docs. ChatKit adds no fee; the bill is model tokens plus storage beyond 1 GB.
A backend engineer whose ChatKit app runs on an Agent Builder workflow
Keep the assistant running after 30 November 2026
Rebuild the workflow as code with the Agents SDK, move to the custom server integration and keep the same frontend. Budget for adding real user authentication, since the hosted route did part of that work.
A data protection officer at a European insurer
Check whether a ChatKit assistant can meet a no-retention policy
ChatKit threads are kept until deleted and are outside zero data retention, so either schedule deletions through the API or keep conversation storage on your own server with the self-hosted SDK.
Before you commit
- Setup
- Install @openai/chatkit-react or load chatkit.js, run the ChatKit Python SDK on your server, and add an endpoint that authenticates the user and returns a client secret; OpenAI provides a starter app and sample servers to copy from.
- Integrations
- React, plain JavaScript and a CDN script on the frontend; the Python SDK on the server, with examples in Python, Ruby, JavaScript, Go and Java; works with the Agents SDK, Responses API and hosted tools.
- Data and privacy
- API data is not used for training by default. /v1/chatkit/threads keeps data until you delete it and is not eligible for zero data retention.
- Cost at scale
- Model tokens at the chosen model's API rate, plus $0.10 per GB-day for upload storage; keeping 50 GB of attachments for a 30-day month comes to about $150 before the free 1 GB.
- Limits
- The hosted Agent Builder route ends on 30 November 2026; the `user` value must be unique per end user; the docs do not cover rate limits specific to ChatKit.
What’s good
- Ready-made chat window with streaming, attachments, widgets and tool-call display
- React, plain JavaScript and CDN options on the frontend
- Self-hosted server can front any agent, not only OpenAI's
- Apache-2.0 libraries with no fee beyond tokens and storage
- Widget builder, playground and starter apps
What’s not
- Hosted Agent Builder route shuts down on 30 November 2026
- Self-hosting leaves auth, storage and the agent loop to you
- ChatKit threads are not eligible for zero data retention
- Python SDK has not had a release since May 2026
- Inconsistent auth guidance across the official examples
Best for
Product teams already building on OpenAI who want a polished chat interface without writing one from scratch.
OpenAI ChatKit alternatives
OpenAI ChatKit: common questions
- Is OpenAI ChatKit worth it?
- ChatKit saves a team weeks of building a polished chat window, but the easy hosted version dies at the end of November and the self-hosted one leaves the backend to you. We rate it 3.6 out of 5.
- How much does OpenAI ChatKit cost?
- No fee for ChatKit itself; you pay model tokens at normal API rates. File and image upload storage costs $0.10 per GB-day after 1 GB free per account each month.
- What are the downsides of OpenAI ChatKit?
- Hosted Agent Builder route shuts down on 30 November 2026; self-hosting leaves auth, storage and the agent loop to you; chatKit threads are not eligible for zero data retention; python SDK has not had a release since May 2026; inconsistent auth guidance across the official examples.
- Who is OpenAI ChatKit best for?
- Product teams already building on OpenAI who want a polished chat interface without writing one from scratch.
- How much does OpenAI ChatKit cost for a team or in production?
- Model tokens at the chosen model's API rate, plus $0.10 per GB-day for upload storage; keeping 50 GB of attachments for a 30-day month comes to about $150 before the free 1 GB.
- Is OpenAI ChatKit safe for company data?
- API data is not used for training by default. /v1/chatkit/threads keeps data until you delete it and is not eligible for zero data retention.
- What are the best OpenAI ChatKit alternatives?
- The closest alternatives we have reviewed are OpenAI API, Voiceflow and Intercom Fin.
Answers checked Nov 6, 2026.