LiveAgent: LiveAgent – a local‑first desktop AI assistant that can read/write files, run shell commands and be accessed remotely via a web gateway.

What it is

LiveAgent is a cross‑platform desktop client (macOS, Windows, Linux) that lets a large‑language‑model‑backed agent act on your computer. It can:

• talk to Claude, OpenAI Codex, Gemini or any compatible API you point it at;
• read, edit, create and delete files, search with glob/regex, and run Bash commands;
• keep long‑running processes alive and expose them to the internet with a one‑click tunnel;
• store conversation history and knowledge in a local SQLite + Markdown store;
• run scheduled tasks (cron‑like) and remember context across sessions. All of this works offline on the desktop; a separate Go‑based Gateway can be deployed if you want to control the same agent from any browser.

Core components

Component	Tech	Role
Agent GUI	Tauri 2 + React 19 + Rust (Tokio)	Desktop UI, local tool bridge, SQLite memory, LLM SDKs
Gateway	Go 1.25	WebSocket/HTTP relay, Web UI, remote‑control API (Docker‑ready)
MCP protocol	stdio / HTTP bridge	Allows third‑party tools or “Skills” packages to plug into the agent
Skills ecosystem	“ClawHub” packages	On‑demand extensions (e.g., code formatters, web scrapers)

Getting started

1. Download the appropriate installer from the latest GitHub Release (DMG, EXE/MSI/portable zip, AppImage, DEB or RPM). The app is signed/notarized where applicable.
2. Run the desktop client, add your API keys (stored only on the local machine), and start a chat.
3. Optional – remote access: pull the Docker image ghcr.io/stack-cairn/liveagent-gateway:latest, run it with a token, and point the desktop Settings → Remote page to the gateway URL. Nginx reverse‑proxy config is provided for TLS.

How it works under the hood

• The UI talks to a Rust backend via Tauri’s IPC. The backend talks to LLM providers (Claude, Codex, Gemini) using the respective SDKs and routes requests based on the selected model.
• File‑system and Bash capabilities are exposed through a ManagedProcess layer that supervises commands, enforces a working directory and timeout, and can keep services alive.
• Conversation history is stored as segmented messages plus periodic summary checkpoints, enabling the agent to retain context even after long chats.
• A SQLite FTS index lets you search past knowledge with full‑text queries.
• The Gateway mirrors the same WebSocket protocol used by the desktop client, so a browser UI can issue the same commands. Short network interruptions are recovered by replaying a bounded sequence window.

Who might use it

• Developers who want an AI that can edit code, run builds, or manage local dev servers without leaving the terminal.
• Power users who need a personal assistant that can manipulate files, schedule scripts, and keep a searchable knowledge base.
• Teams that want a self‑hosted remote UI for a shared agent (e.g., via the Docker gateway).

License & contribution

The project is released under the MIT license. Contributions are welcomed; the repo includes CI checks for TypeScript, Rust and Go, a Makefile with common dev commands, and a contrib.rocks badge showing community contributors.