Z.AI

What it is

Z.AI (Zhipu AI) is the platform behind the GLM model family. Its chat endpoint is OpenAI-compatible, while the rest of the capabilities live on separate paths of the same API: image and video generation, web search, speech recognition and a tokenizer. The integration connects with a personal API key and lets a bot scenario hold a conversation, read images, search the web, generate pictures and video, transcribe audio and count tokens.

Requirements

An account at z.ai.
A positive balance — every operation is billed.
An API key from the API Keys section.

How to connect

Sign up at z.ai and top up your balance.
Open API Keys and create a key.
Copy the key.
In the constructor open Integrations → find the Z.AI card → Connect.
Paste the key into the connection field and save.
Click Test connection — the integration makes a lightweight call to the tokenizer and reports the result.

Available operations

Operation	What it does
Send message	Chat with the model: system prompt, conversation history, JSON mode, reasoning mode, streaming output
Image analysis	Answers a question about an image (URL or a Telegram file_id) using a vision model
Image generation	Draws a picture from a description with `glm-image` and uploads it to Telegram for a stable `file_id`
Reaction to a message	The model picks one emoji reaction from a given list
Web search	Returns the structured Z.AI search result: titles, links, text excerpts and dates
Video generation	Creates a clip from a description or a source image with `cogvideox-3`
Audio transcription	Turns a voice message or audio file into text with `glm-asr`
Token count	Counts how many tokens a text takes in the selected model

Limits

An image sent to a vision model must be under 5 MB and no larger than 6000×6000 pixels.
Audio for transcription must be wav or mp3, up to 25 MB and no longer than 30 seconds.
A video prompt is capped at 512 characters; duration is 5 or 10 seconds, frame rate 30 or 60 fps.
Sizes for glm-image: each side between 1024 and 2048 pixels and divisible by 32. cogview-4 is stricter — each side divisible by 16 and no more than 2 MP in total.
The tokenizer only accepts glm-4.6, glm-4.6v and glm-4.5.
Links to generated images and videos expire after 30 days.

Notes

Video generation is asynchronous: the integration submits the task and waits for the result, polling Z.AI for up to 5 minutes. The block holds the scenario for that whole time — don't put it on a branch where the user is waiting for a quick reply.

Uploading the result to Telegram is controlled by the "Don't upload to Telegram" checkbox in the block — as in every other integration, it is on by default. While it stays on, only {{result.url}} is available — a Z.AI link that stops working after 30 days, so save the file yourself if you need it long-term. Clear the checkbox to have the bot upload the result to Telegram and return a permanent {{result.file_id}}.

Z.AI publishes no models-listing endpoint, so the dropdown in the block is filled from the model codes documented by the platform. If Z.AI ships a new model, its code can be typed into the "Model" field by hand.

"Web search" is not a conversation with a model but a direct query to the search engine: it returns an array of sources ({{result.0.title}}, {{result.0.link}} and so on) rather than a finished answer. To get readable text, feed the findings into "Send message" as the next step.