$ kin up —— self-hosted · any-model · sandboxed · AGPLv3

Your data. Your models.
Your servers.

Kin is a private, self-hosted AI-Agent workspace for small teams.

~/kin/agents/c
model: glm · any provider · live
session · refactor landing
you
Reskin the landing, then build.
kin · glm
On it — updating design tokens and verifying the build —
 bash · sandbox
# runs in an isolated sandbox
$ pnpm build
built in 6.2s · 0 errors
$
// works with any model · built on open standards
GLM · DeepSeek · GPT · Qwen · MCP · Skills · Kin Agent
[01] private

Private by deployment

One-command Docker. Single org, multiple trusted users. The image runs on the machine you choose — works out of the box.

screenshot · 自部署 · 团队登录
screenshot · 共享 Project · 团队协作
[02] co-work

Share a project with your team

Like ChatGPT's Shared Projects — but self-hosted. Sessions, knowledge and output are shared across the team, with invites and access you control.

[03] rag

Built-in RAG · your own vector DB

Upload documents into a knowledge base; the agent answers grounded in retrieval over your own vector database — something a desktop app simply can't give you.

screenshot · 知识库 · 向量检索
screenshot · OCR · 复杂 PDF / 表格识别
[04] plugins

Plugins · OCR for complex PDFs & tables

Extend Kin with plugins — like an OCR component that scans complex PDFs and extracts table structure.

[05] live preview

Run a live server in a sandbox · share a public URL

Spin up a React + Vite app inside an isolated sandbox and share a real public link with anyone — scripts, routing and storage all actually work.

screenshot · React+Vite 沙盒 · 公网分享
screenshot · 模型选择 / 健康看板
[06] any model

Any model, no lock-in

Bring any model via API — GLM, DeepSeek, GPT, Qwen — or point Kin at your own endpoint. No vendor lock-in; the menu only shows models that probe healthy.

[07] who it's for

Built for teams that can't trust public-cloud AI.

Data-sensitive teams under ~30 people — heavy AI users with no platform-ops team and no room for a data leak.

01

Law firms

Client confidentiality and privilege — work that public AI tools can break.

02

Funds

MNPI, deal and portfolio data that can never leak.

03

Family offices

Family privacy and assets, with zero exposure.

04

Research teams

Unpublished IP and grant data, kept in your own lab.

[08] deploy

Clone, and go — it really is that simple.

Optimized for "one-command Docker, works out of the box for a team", not elastic public-SaaS scale. Open-source (AGPLv3), self-hosted — your documents, sessions and audit stay on your server, and model calls go only to the endpoint you choose.

 deploy.sh
# three steps to launch
$ git clone github.com/deeptoai/kin
$ cp .env.example .env # add a model key
$ docker compose up -d
kin up · https://kin.local
$