Stream the model output token by token in WordPress
SleekAI adds an AI chatbot for Streaming Tokens to your WordPress site. It answers from the content you choose, such as posts, pages, custom post types, post meta, taxonomies and user data, with your own OpenAI, Anthropic, Google, or OpenRouter key.
Token streaming, not request-response
A non-streaming chatbot waits for the model to finish, then drops the entire answer into the widget. On a Llama 3.1 70B reply that runs three paragraphs, the visitor stares at a spinner for four to eight seconds before anything shows up. That is the difference between a chatbot that feels alive and one that feels broken, and most WordPress AI plugins still ship the spinner version.
SleekAI streams. The PHP side opens a server-sent events connection to OpenAI, Anthropic, Fireworks, or any OpenAI-compatible endpoint that supports stream: true.
The streaming layer is invisible to the rest of the bot. Only the rendering changes, and visitors immediately feel the difference.
Features
SSE-based streaming
The visitor sees the reply build word by word instead of waiting for a full JSON response to land.
Sub-second first token
Perceived latency on a streaming bot is the time to first token, not total completion.
Answers from your data
Map post titles, content, post meta, taxonomies or a WP_Query into the instruction, and the bot answers from what your site actually publishes.
Where streaming changes the chat experience
Long documentation answers
Docs questions often need three or four paragraphs of explanation. Streaming makes those answers feel responsive instead of broken, because the visitor starts reading the first sentence while the model is still finishing the last.
Product recommendations
When a shopper asks for the right SKU, the bot can stream a tailored recommendation built from postmeta and product taxonomies. The first line appears in under a second, which keeps the visitor engaged through the rest of the explanation.
Tutoring and learning
Educational bots need step-by-step explanations. Streaming lets the learner read step one while the model is still composing step two, which feels closer to a real tutor than a request-response transcript ever does.
Stream tokens through SleekAI in four steps
Pick a streaming-capable provider
OpenAI, Anthropic, Fireworks, Together, Groq, and most self-hosted vLLM or TGI deployments support stream: true.
Set display conditions
Decide which pages, post types, roles or logged-in states show the chatbot.
Tune the proxy layer
If WordPress sits behind Nginx, Cloudflare, or a host WAF, allow the chat endpoint to bypass response buffering.
Map your data
Choose the posts, fields, taxonomies or users the bot should know and reference them in the instruction.
Why streaming is the baseline for modern chat
Visitors have been trained by ChatGPT, Claude, and Gemini to expect words to appear as the model is still thinking. When a WordPress chatbot reverts to a spinner and dumps the whole answer at once, the experience instantly feels years out of date even when the underlying model is the same.
The fix is not faster inference, it is streaming. Server-sent events let the server push tokens to the browser the moment the model emits them, which collapses perceived latency to whatever the first-token latency of the upstream provider happens to be.
On hosted providers that number is usually well under a second. SleekAI ships streaming as the default, not an upsell. The result is a chatbot that feels like the tools your visitors already use, on a stack you fully own, without writing a single line of SSE plumbing yourself. That alignment with modern UX expectations matters more for engagement than almost any other implementation detail in a chatbot.
Frequently asked questions
Does streaming work with Anthropic Claude?
Yes. From the WordPress side the configuration is identical to a streaming OpenAI bot.
What about WAFs that buffer responses?
Some WordPress hosts run a WAF or reverse proxy that buffers responses and breaks SSE. If your stack still buffers, disable buffering on the chat endpoint route.
Does streaming break with caching plugins?
Object caching is fine, but full-page cache plugins can intercept the chat endpoint and cache the SSE response, which obviously breaks streaming.
Which AI providers does it support?
OpenAI, Anthropic, Google and OpenRouter, plus custom OpenAI-compatible endpoints. You bring your own API key and the provider bills you directly.
Where does the chatbot get its answers?
From the WordPress data you map into its instruction: post titles and content, post meta, ACF and Meta Box fields, taxonomies, terms, users and user meta, or a WP_Query. Values resolve on every request.
Can it act on a visitor’s behalf?
No. A chatbot answers questions.
Can I control where the chatbot appears?
Yes. Display conditions cover post type, post, post status, post template, taxonomy, term, user role, logged-in state and WordPress conditional tags such as front page, archive, search and 404.
Are conversations stored?
Only if you turn on logging for that chatbot. SleekAI then stores each visitor message and each completed reply, which you read and delete on the chatbot’s Logs page, and a retention period can delete old logs.
Pricing
Explore our flexible licensing options tailored to your needs. Upgrade your license anytime to access more features, or opt for a lifetime license for ongoing value, including lifetime updates and lifetime support. Our hassle-free upgrade process ensures that our platform can grow with you, starting from whichever plan you choose.
…or get the Bundle Deal
and save €250
The Bundle (unlimited sites)
Pay once, own it forever
Elevate your WordPress site with our exclusive plugin bundle that includes all of our premium plugins in one package. Enjoy lifetime updates and lifetime support. Save significantly compared to buying plugins individually.
What’s included
SleekAI
SleekByte
SleekMotion
SleekPixel
SleekView
SleekRank
SleekTrack