Features

An agent works through a conversation: it draws on a chosen model, calls the tools it has been granted, and keeps the session's memory. The schema below shows how these fit together; each capability is detailed on its own page.

flowchart TB
    User(["You / your app"])
    docs[("Your documents<br/>and collections")]
    conn["Connectors (MCP)"]

    subgraph Agent["Kiwi agent"]
        direction LR
        core["Agent<br/>prompt + model + tools"]
        mem["Session and memory"]
        core --- mem
    end

    subgraph Models["Models"]
        direction LR
        m1["OpenAI"]
        m2["Google Gemini"]
    end

    subgraph Data["Documents and data tools"]
        direction LR
        t1["retriever_search"]
        t2["summarizer"]
        t3["spreadsheet_schema"]
        t4["spreadsheet_query_sql"]
    end

    subgraph WebCode["Web and code tools"]
        direction LR
        t5["web_search"]
        t6["web_scraper"]
        t7["code_interpreter"]
    end

    subgraph MediaStudio["Media Studio tools"]
        direction LR
        t8["generate_image"]
        t9["edit_image"]
        t10["generate_video"]
        t11["generate_audio"]
        t12["caption_media"]
    end

    User -->|message| core
    core -->|streamed response| User
    core --> Models
    core --> Data
    core --> WebCode
    core --> MediaStudio
    docs --> Data
    conn -. adds tools .-> core

Each page explains a capability in plain terms. Most include an example ask, plus a short "What's coming" note where there's something on the way.

Tools the agent can use

Every built-in tool an agent can use today, by its name in Kiwi:

Tool	What it does	Learn more
generate_image	Create an image from a prompt	Media Studio
edit_image	Edit an existing image from an instruction	Media Studio
generate_video	Create a short video from a prompt	Media Studio
generate_audio	Create music or sound effects	Media Studio
caption_media	Describe an image or video in words	Media Studio
retriever_search	Find and quote relevant passages from your documents	Documents
summarizer	Condense long documents	Documents
spreadsheet_schema	Inspect a sheet's columns and shape	Spreadsheets
spreadsheet_query_sql	Answer questions over a sheet with read-only SQL	Spreadsheets
web_search	Look things up on the public web	Web
web_scraper	Fetch and read specific web pages	Web
code_interpreter	Write and run Python on your data	Execute code

Media Studio also powers a visual image editor (annotate regions and regenerate), offered through the interface rather than as a tool the agent calls.

Read on in order, or jump straight to the capability you care about.