Kimi

What it is

Kimi is Moonshot AI's line of large-context (up to 1M tokens) multimodal AI models, with an OpenAI-compatible API. The integration connects with a personal API key and lets a bot flow chat with the model, analyze images, use the model's built-in web search, and extract text from uploaded files — all directly from the flow.

Requirements

An account at platform.kimi.ai.
An API key from Console → API keys.

How to connect

Sign in at platform.kimi.ai and open Console → API keys.
Create a key and copy it.
In the constructor, open Integrations → find the Kimi card → Connect.
Paste the key into the connection field and save.
Verify the connection with Test connection — it shows the current account balance.

Available operations

Operation	What it does
Send message	Chat with the model: system prompt, conversation history, JSON mode, streaming output
Analyze image	Answers a question about an image (URL or a Telegram file_id)
React to message	The model picks one emoji reaction from a given list
List models	Returns the available models and their parameters
Web search	Chat that automatically uses the model's built-in web search when needed to answer
Extract text from file	Uploads a file (PDF, DOCX, TXT, etc.) and returns its extracted text

Limits

Every operation calls Kimi's chat endpoint (/v1/chat/completions) — tokens are billed per your account's plan.
Kimi's API has no image generation, embeddings, or moderation endpoints — there are no corresponding operations here.
"Web search" may exchange a few requests with the model within a single operation call (until the model gives a final answer) — this adds a bit of latency and token usage compared to a plain message.

Notes

The "Extract text from file" operation doesn't answer questions by itself — it only pulls the file's text into a variable. To ask a question about a document, pass {{result.text}} into the "Send message" operation's system prompt as the next step of the flow — this is the approach Kimi's own docs recommend.

In "Web search", the model itself decides whether search is needed to answer — if the question doesn't need fresh information from the internet, no search runs and {{result.used_search}} stays empty.