LocalLightChat

Portable. Fast. Yours.
The ultimate lightweight UI for AI chat, RAG and agents.

Use local models, cloud APIs or both. Keep huge chats manageable, search your history, work with files and documents, add RAG, or give an agent a shell and workspace — without turning the frontend into a heavyweight stack.

<1s cold start · 500k+ context · <60MB download · 100% portable

Website: https://www.locallightai.com/llc/

What it does

Chat that can get big

• Folder-based chat organization and management
• Full-text search across your chat history
• Compress & Clone for large conversations while preserving decision-critical context
• Documents and editable artifacts with version history
• Attachments including PDF, images, CSV, YAML, XML, logs and more
• URL sources and text notes
• Markdown, Mermaid, Math/TeX and syntax-highlighted code
• Templates, automatic chat naming, cloning, import and export

Give the model tools

• MCP server management with per-chat activation and support for remote MCP servers
• Native MCP agent stack with workspace-scoped filesystem tools
• Process and shell execution
• Persistent interactive terminal sessions
• Persistent code workspaces
• Optional Windows desktop tools for window management, screen capture and keyboard input
• Configurable agentic tool-call rounds and context compaction
• Large MCP results can be externalized as artifacts and selectively inspected later with artifact info, search and read tools instead of repeatedly filling the model context

RAG without the heavyweight stack

• Dedicated ragd service included in the project
• Reusable file library with RAG state
• Dense, hybrid and precise retrieval modes
• Hybrid retrieval combines vector search with BM25
• Configurable chunking, retrieval depth and optional reranking
• OpenAI-compatible embedding endpoint support
• RAG can be enabled globally or per chat

Search, images, voice and model control

• Built-in web search and fetch tools with configurable search providers
• Image generation through OpenAI-compatible APIs, OpenRouter, stable-diffusion.cpp or ComfyUI bindings
• Image editing through supported ComfyUI workflows
• Voice transcription and text-to-speech
• Local and cloud model connections through OpenAI-compatible APIs
• Multiple connection profiles with custom headers and model prefixes
• Advanced generation controls including temperature, sampling, DRY and Mirostat
• Configurable model pricing and usage information
• Role-based users: superadmin, admin and user

Download

Ready-to-use portable packages are available from GitHub Releases:

https://github.com/srware-net/LocalLightChat/releases

The repository contains packaging for:

• Windows
• Linux x64
• Linux ARM64
• macOS (coming soon)

Docker

Prebuilt image:

srwarenet/locallightchat:latest

Docker Hub:

https://hub.docker.com/r/srwarenet/locallightchat

Run locally:

docker run -d \
  --name locallightchat \
  -p 127.0.0.1:8181:80 \
  -p 127.0.0.1:8182:443 \
  -v locallightchat:/var/lib/locallightchat \
  --restart unless-stopped \
  srwarenet/locallightchat:latest

Open:

http://127.0.0.1:8181

HTTPS is also available on:

https://127.0.0.1:8182

The included certificate is self-signed, so a browser warning is expected.

Stop and remove the container:

docker stop locallightchat
docker rm locallightchat

Docker Compose / source build

The Docker setup runs nginx, PHP-FPM, MCP/Agent and optional RAG services in a single container and persists application data, uploads, temporary files and agent workspaces in Docker volumes.

Before building from the repository, copy the application files into docker/public/ and provide the matching ragd and mcp_server binaries in docker/native/<arch>/ as indicated by the placeholder files.

Then:

cd docker
cp .env.example .env
docker compose up -d --build

Default URLs:

http://127.0.0.1:8181/
https://127.0.0.1:8182/

By default Docker binds only to 127.0.0.1. Change the LLC_HTTP_BIND and LLC_HTTPS_BIND values in .env if you intentionally want LAN access.

RAG is disabled in the default Docker configuration until an embedding endpoint is configured in docker/config/ragd.conf. The MCP server starts with the container and uses a persistent workspace volume.

More Docker details: docker/README_DOCKER.md

Self-hosted

Use this if you want to deploy LocalLightChat on your own web stack.

Requires:

• nginx or a similar web server
• PHP 8.x
• SQLite

Upload the contents of:

www/

to your web server document root, for example:

htdocs/
public_html/
www/

Usually you should upload the contents of www/, not the folder itself.

Agent and RAG features require their corresponding services in addition to the web application.

Import from OpenWebUI

The included v4 converter can migrate an OpenWebUI SQLite database to the current LocalLightChat schema.

It supports chats, folders, messages, file metadata, image/audio attachments and optional alternate message branches as cloned chats. API keys, login credentials, sessions and provider secrets are deliberately not copied.

Basic usage:

python converter/convert_openwebui_to_locallightchat_v4.py openwebui.db

For media migration and advanced options, see:

converter/README_openwebui_to_locallightchat_v4.md

Project structure

LocalLightChat/
├─ www/          Web application
├─ mcp/          MCP gateway, configurator and native tool server
├─ ragd/         RAG service
├─ docker/       Docker image and Compose setup
├─ packages/     Portable platform packages
├─ converter/    OpenWebUI migration tool
├─ LISENSE
└─ README.md

License

Apache License 2.0. See LISENSE.