GitHub - AI agent with channels, SillyTavern cards, and sandboxing

Hi everyone, I just released the first version of DuckAgent, a local agent runner focused on making character-based agents easier and safer to run. It supports SillyTavern character cards, so you can reuse existing cards instead of rewriting personalities or prompts from scratch. It also includes a fairly complete sandbox, designed to limit what an agent can access or execute while still letting it use tools in a practical way. I’d love for people to try it out and share feedback, especially around the card workflow, sandbox behavior, and what would make it more useful for real experiments.