Getting started
From downloading the app to a real conversation — this chapter takes five minutes.
AgentKit Studio runs in three shapes. Pick the one that matches you:
Download the latest signed installer (.exe / .dmg) from the download page
and double-click. No Node.js, pnpm, or build toolchain required.
Install @kingsoftcloud/agentkit-studio into an existing DeepSeek Harness environment to
attach Studio capabilities to your own Host. For teams that already run DSH.
For development, patch debugging, or verifying behaviour against a real Host. Requires Node.js 24 and pnpm 10. See Contributing.
What is in this chapter
5-minute quickstart
Install → configure a model → create an agent → send the first message.
Install
Per-platform notes, requirements, data directories, and uninstall.
First conversation
What you are looking at: tool calls, file delivery, run records, and recovery.
Before you start
| You need | Notes |
|---|---|
| macOS 13+ (Apple Silicon) or Windows 11 (x64) | macOS ships a signed and notarized DMG; Windows uses an NSIS installer |
| An API key for a model provider | Any OpenAI-compatible endpoint, or your organization's model gateway |
| Optionally, cloud credentials | Only required to deploy agents to the cloud |
Cloud features need a control plane
Deployment, managed runtimes, knowledge bases, and memory depend on the AgentEngine control plane and credentials. A working local install does not imply cloud resources are reachable — see Managed runtime.
Next
Start with the quickstart. If you would rather understand the product model first, read Core concepts.