ghostclaw
Ghostclaw is an open-source AI agent that runs locally on your computer and can be controlled via Telegram. It leverages Anthropic's Claude models to understand and execute tasks directly on your file system, command-line environment, and other local resources. The agent is designed around a dual-mode operation: a fast-path for simple queries that uses a lightweight API call to Claude Sonnet without spawning a full agent process, and a full agent mode for complex, multi-step tasks that require tools and state. This architecture keeps costs low for casual interactions while retaining the power to handle complex workflows. At its core, Ghostclaw is a Node.js application that listens for Telegram messages and dispatches them to either the fast path or a spawned agent process. The fast path uses the Anthropic SDK to generate quick answers and returns a HANDOFF signal if tools are needed, falling through to spawn a full agent. The full agent, built on the Claude Agent SDK, can interact with the user's environment, execute code, read and write files, and use custom skills. Skills are modular extensions that can be added to the agent, with templates for integrating services like Discord, Slack, and more. Usage and costs are tracked in a local SQLite database, and a daily budget cap enforced by the orchestrator prevents runaway spending. Ghostclaw provides a suite of Telegram slash commands for management, including /model to choose between Claude models (Sonnet, Opus, Haiku), /budget to set or check spending limits, and status reporting. It is designed to be a persistent, always-on assistant that lives on your computer, combining the convenience of a chat interface with the power of direct local access. The project is actively maintained, with regular updates focusing on performance, cost control, and simplicity.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Ghostclaw addresses the need for a personal, always-available AI assistant that can directly interact with your computer without relying on cloud services that have limited local access. Many users want to automate tasks, manage files, run scripts, and get quick answers through a chat interface, but existing solutions often require manual context switching or are entirely cloud-based. Ghostclaw solves this by running locally, giving it full access to the user's environment, while offering a familiar Telegram interface. It balances cost and capability by using a fast path for trivial requests and a full agent only when necessary, making it practical for daily use without excessive API fees.
Use cases
- Automating file management: renaming batches of files, organizing downloads, cleaning up directories.
- Running shell commands and scripts on demand via Telegram messages.
- Quick information retrieval: checking weather, stock prices, or summarizing notes without leaving Telegram.
- Scheduling and reminders: setting up one-time or recurring alerts, managing a personal calendar.
- Development assistance: asking for code explanations, generating boilerplate, or debugging from a remote location.
- Content generation: drafting emails, reports, or social media posts and saving them as local files.
Audience
- Developers
- Hobbyists
Pros
- Cost-efficient dual-mode operation that minimizes API spending
- Open source and self-hosted, ensuring data privacy and user control
- Flexible Telegram integration for remote access and control
- Built-in daily budget cap to prevent unexpected expenses
- Extensible skill system for adding custom functionality
Cons
- Limited to Anthropic’s Claude models; no support for other LLM providers
- Requires technical expertise to set up and maintain locally
- Telegram-only interface; lacks integration with other messaging platforms
Official website: https://github.com/b1rdmania/ghostclaw
Canonical LambdaBase page: https://www.lambdabase.com/tools/ghostclaw