lightclaw: Run a Lightweight Local-First AI Assistant

Overview

LightClaw is a local-first AI assistant for people who want an agent with tools and persistent memory without a heavyweight application runtime. It packages the assistant as a Rust binary and connects to configured language-model providers, while keeping memory data in local SQLite storage.

It is a fit for users who prefer operating an assistant through Telegram or Discord, or running it as a background service. Its capabilities include shell and file tools, so it is best suited to users comfortable configuring and controlling an agent with access to those actions.

Key Features

• Single-binary distribution, with no Python runtime requirement.
• Agent tools for file operations, shell commands, web access, and scheduling.
• Telegram and Discord channel integrations.
• Local memory using SQLite, including conversation summaries and semantic retrieval.
• OpenClaw-style skills that can be discovered and activated by the agent.
• CLI commands for skill management and background service operations.
• Provider configuration with model fallbacks.

Use Cases

• A self-hosting user can run a personal assistant on a Linux machine or Mac and interact with it through Telegram or Discord.
• A Raspberry Pi or ARM server owner can deploy an assistant without setting up a Python environment.
• A developer can configure an agent to work with files, shell commands, web tools, or scheduled tasks.
• A privacy-conscious user can keep assistant memory in local SQLite storage instead of relying on an external vector database.

Getting Started

Install and start the setup flow:

curl -fsSL https://lightclaw.dev/install.sh | bash
lightclaw configure

See the README for configuration, platform support, and build instructions.

Alternatives

• goose: Goose is a native desktop, CLI, and API agent for installing, executing, editing, and testing, while LightClaw emphasizes a single-binary assistant with Telegram and Discord interfaces.

Considerations

• The assistant requires configuration for a model provider. Some integrations, such as web search and audio transcription, may also require provider credentials.
• Telegram and Discord use bot tokens, with optional user or channel allowlists to configure.
• Because the agent can use shell and file tools, review its permissions and configuration before granting access to sensitive data or systems.
• The README describes Windows support as less stable and less well-supported than the listed Linux and macOS platforms.
• The repository was created in February 2026, so its project history is relatively short. The supplied metadata does not specify a license.