Open Source  /  Self-Hosted  /  Private

Your AI.
Your Hardware.

Big tech wants your chats. Chakor keeps them on your machine.

A free, open source AI workspace you run yourself. Chat with any local or cloud model, search the web, talk to your documents, and compare models side by side. Zero tracking. Zero cloud dependency.

Get Started Free Explore Features
Local Models Web Search Document Chat Model Compare
100%
Self-Hosted
0¢
Cost to Run
5+
AI Providers
MIT
Open Source

Why Chakor

Built different, on purpose.

Three things you will not find stitched together anywhere else.

01

Knows what your machine can actually run

Most apps let you download a model and find out it was too big when it crashes. Chakor reads your RAM and GPU first and tags every model FITS, TIGHT, or TOO BIG. A 4 GB laptop and a 24 GB workstation each get an honest answer.

FITS TIGHT TOO BIG
02

Download models without touching a terminal

Search Hugging Face from inside the app. See every quant with its size and a fit tag, click once. It downloads in the background, straight into your models folder. Close the tab and it keeps going.

▓▓▓▓▓▓▓░ llama-3.2-3b-q4.gguf   2.1 GB/s
03

Switch engines without the crashes

Flip between llama.cpp, Ollama, and LM Studio in one tap. Chakor unloads the previous model first, so two engines never fight for the same VRAM and crash on a modest GPU.

Ollama → llama.cpp

Everything it does

One workspace. Every AI feature.

From local hardware to cloud APIs, Chakor has it all in one place.

models

Local and Cloud Models

Ollama, llama.cpp, LM Studio, or your own keys for OpenAI, Anthropic, Google, and OpenRouter. One interface for everything.

search

Web Search Built In

Per-message web search with SearXNG, then Brave, then DuckDuckGo fallback. Real answers grounded in live sources.

rag

Talk to Your Documents

Drop PDFs, Markdown, or text files and ask questions. Chakor retrieves relevant passages and answers from them.

compare

Blind A/B Model Compare

Run the same prompt across models side by side, vote blind, and build a personal leaderboard of which model wins.

memory

Memory Across Conversations

Tell Chakor facts you want remembered. They persist across every conversation, so you never repeat yourself.

notes

Notes and Checklists

A built-in Keep-style notepad with checklists, colors, pinning, and archive. Your workspace, your notes.

privacy

Zero Telemetry

No analytics. No tracking pixels. Conversations stay in a local SQLite database on your disk. Forever.

multi-user

Multi-User Admin Panel

Run for your whole team. Invite codes, registration control, and an admin panel to manage accounts.

pwa

Installable on Any Device

Works as a PWA on phones and desktops. Even runs the server on Android via Termux.

Comparison

Chakor vs the usual options.

Both Open WebUI and LM Studio are good. Here is where Chakor is different.

Feature Chakor Open WebUI LM Studio
Open source✓✓✗
Self-host on a server✓✓Desktop only
Run on Android (Termux)✓Partial✗
Local models (Ollama / llama.cpp)✓✓✓
One UI for llama.cpp + Ollama + LM Studio✓Partial✗
Hardware fit tag per model✓✗Partial
Download GGUF from Hugging Face in-app✓✗✓
Background downloads (close tab and wait)✓✗✓
Frees old model before loading new one✓✗n/a
Bring your own cloud keys✓✓Limited
Web search built in✓✓✗
Chat with your documents✓✓Beta
Multi-user + admin panel✓✓✗
Blind A/B compare + leaderboard✓Partial✗
Notes + memory across conversations✓Partial✗
Rebrand from one config file✓✗✗
One small codebase you can read✓HeavyClosed

Quick Start

Up and running in one command.

All three paths land on http://localhost:3001. The first account you register becomes the admin.

bash · Linux, macOS, Windows 
# Clone and start (Docker Engine or Docker Desktop required)
git clone https://github.com/TYFSADIK/chakor.git && cd chakor
docker compose up -d

# Optional: bundle a local model with Ollama
docker compose --profile ollama up -d
docker compose exec ollama ollama pull llama3.2
bash · Linux, macOS 
# Installs Git, Node, Ollama and starts Chakor
git clone https://github.com/TYFSADIK/chakor.git && cd chakor
bash install.sh

# Want everything including CUDA for NVIDIA GPUs?
bash install.sh --all

# Fresh server, fully unattended, runs as a service
sudo bash install.sh --yes --service
bash · Node 20+ 
git clone https://github.com/TYFSADIK/chakor.git && cd chakor
npm install --legacy-peer-deps
cp .env.example .env.local      # then set AUTH_SECRET
npm run build
npm start

# Development with hot reload
npm run dev
Termux · Android 
# 1. Install Termux from F-Droid (not Play Store)
# 2. Open Termux and run:
pkg install git
git clone https://github.com/TYFSADIK/chakor.git && cd chakor
bash install.sh

# Your phone is now the server.
# Open http://localhost:3001 in any browser.

FAQ

Common questions about Chakor.

Chakor is a free, open source AI workspace you run on your own hardware. It supports local AI models through Ollama, llama.cpp, and LM Studio, as well as cloud models via your own API keys (OpenAI, Anthropic, Google, OpenRouter). Everything stays on your machine with zero tracking or cloud dependency.
Yes. Chakor is completely free and open source under the MIT license. You run it on your own server or machine at no cost. Cloud model API costs (OpenAI, Anthropic, etc.) depend on your usage and the provider's pricing. Chakor never charges for those.
No. CPU works fine, just slower for local inference. Or use cloud API keys and run nothing locally at all. Chakor reads your actual hardware and tags each local model FITS, TIGHT, or TOO BIG before you load it, so you never accidentally load something that crashes.
Chakor is open source (unlike LM Studio), runs as a server you can access from any device including phones (unlike LM Studio's desktop-only mode), and is lighter and simpler than Open WebUI. Chakor uniquely tags every model FITS, TIGHT, or TOO BIG based on your actual RAM and GPU, lets you download GGUF models from Hugging Face inside the app, and switches engines without OOM crashes.
No. Chakor has zero telemetry. All conversations are stored in a local SQLite database on your disk. Data only leaves your machine when you explicitly use web search or choose a cloud model provider, and only to that provider. There is no analytics, no tracking, no data collection of any kind.

Your AI should answer to you,
not to a data broker.

Run Chakor on your laptop. Run it on a server. Run it on an old phone in a drawer. Keep every word.

Get Started Free Star on GitHub

MIT Licensed · No Account Required · No Cloud · No Tracking