Getting started

The normal way to use Kiwi is simple: you work with a Kiwi agent through the web interface — sign in, pick or build an agent, and have a conversation. The agent reasons about your request, uses its tools when needed, and streams the result back.

Open the interface here: kiwi-ui.lionspace.ai.

Before you begin

• Access is granted through your organization's SSO — contact the Kiwi team if you don't have it yet.
• Nothing to build first. A default Kiwi agent is always available with all 12 built-in tools enabled, so you can start chatting right away.

This guide covers what you can do, in two parts: first set up an agent, then work with it. You don't have to do these in order — with the default agent you can start chatting immediately and come back to the rest as you need it. Other ways to use Kiwi — the API and connectors — are covered in Architecture and Coming soon.

In a hurry?

Jump straight to Interact with your agent and chat with the default Kiwi agent now. Build your own when you need one.

Set up your agent

Define an agent, decide who can use it, and give it the tools and knowledge it needs.

Create an agent

An agent is a reusable configuration: a name, instructions (its prompt), a default model, and the tools it may use. To build one:

1. In the sidebar, open Agents — your agents are grouped as My agents, Shared with me, and Public. Select New agent.
2. Give it a clear name (for example Brand Tone Checker).
3. Write its instructions — describe what it does and how it should behave. This is the most important part; it defines the agent's role.
4. Pick a default model. If unsure, start with the default — you can change it later or override it per conversation.
5. Save. You don't need an administrator role to create an agent for yourself.

Share agents

An agent can be kept private, shared with specific people (as owner or member), or made public for everyone. Publishing a public agent requires an administrator role. Configure an agent once and a whole team can reuse it.

Configure tools

New agents automatically receive every current and future built-in tool, such as retriever_search for documents, web_search for facts, and generate_image for visuals. You can instead select an explicit tool set when you need a focused or tool-free agent. Connector tools are always added explicitly. See Tools for the full list.

Tip

Use an explicit tool set when tighter capability control is more important than automatically receiving new built-in tools.

Upload files to the library

In the Library, upload or import your files — PDF, Office documents, CSV, text, images, and common media. Kiwi automatically extracts the text and the images inside them and prepares them for search. Newly uploaded files are processed first, so watch the processing status and wait until a file is ready before relying on it.

Organize documents into collections

A collection is a named group of documents you reuse instead of attaching the same files over and over:

1. In the Library, create a new collection and give it a clear name (for example Brand guidelines).
2. Add the documents you want to it.
3. Attach the collection where you need it — to an agent (default context), a session (a whole conversation), or a single message (one-off context).

Group documents by purpose so an agent only sees what it needs.

Work with your agent

Hold a conversation, bring in context, adjust on the fly, and pick up where you left off.

Interact with your agent

From the Dashboard, select Start New Chat (or pick a recent agent) to open Chat. The default Kiwi agent is ready to use, so you can start typing immediately. Type your message and send it — the response streams back, showing the agent's reasoning, the tools it uses, and any media it generates. Each conversation is a session with memory, so follow-ups work naturally and sessions are titled automatically.

Make the agent search the web

When a question needs current or external information, let the agent reach the public web. In Chat, open Agent Settings and switch web search on. The agent then uses web_search to find relevant results and web_scraper to read specific pages. Switch it off to keep answers grounded only in your own knowledge and attached documents.

Attach documents to the chat

Beyond an agent's default knowledge, you can attach files or a collection to a single session or message — useful for a one-off briefing without changing the agent. The agent retrieves and quotes the relevant passages (retriever_search), so each answer shows where it came from.

Change agent configuration on the fly

You don't have to edit the saved agent to adjust a single conversation. In Chat, open Agent Settings to switch to another agent, override the model, set the reasoning level, and toggle web search — all for the current conversation only.

Edit a generated image

After the agent creates an image (generate_image), you can refine it: edit by instruction (edit_image), or open the visual editor to annotate specific regions and regenerate just those areas — without starting over.

Navigate between sessions in the sidebar

Your conversations are kept as sessions in the sidebar, each auto-titled, so you can switch between them and resume a long analysis days later with its history intact. Use New Session to start fresh.

What's next

• Understand the ideas → Concepts — what an agent is, the tools, and how data is stored.
• See it applied → Use cases — media generation, document analysis, spreadsheets, and more.
• What's planned → Coming soon.