Castor: A Self-Hosted AI Agent for Business Workflows

Castor is a self-hosted AI agent developed by deepfounder-ai, built to integrate into various business workflows such as customer operations, internal automation, knowledge retrieval, and scheduled reporting. The agent is designed to deploy on a laptop, workstation, or private server, ensuring that data remains on the user's infrastructure unless explicitly sent to a third party.

Key features of Castor include compatibility with any OpenAI-compatible LLM provider, such as Azure OpenAI, AWS Bedrock, OpenAI, Groq, OpenRouter, DeepSeek, and Together. It also supports local models via tools like LM Studio or Ollama for on-premise data processing. The system prioritizes the surrounding infrastructure to handle heavy lifting, employing tool search, semantic memory, and a scheduler to keep prompts lean and manage tasks efficiently. Castor offers various interfaces, including a web UI, terminal, and Telegram bot, and provides direct USB/serial access for hardware integration with devices like scales, scanners, and GPS. It also includes a sandboxed canvas panel for rendering HTML artifacts.

The latest release, v0.23.4, addresses critical security fixes related to secret scrubbing. This update ensures that sensitive information is properly redacted across chat history, entity/wiki summaries, and audit trails, closing bypasses identified in a codebase architecture review. The release also activates a previously defined trajectory pruning function to manage audit trail retention.