mindsdb/anton

Executes multi-step tasks autonomously by generating and running code in isolated scratchpad environments, scraping live data, and iterating based on results—all without pre-configured skills. Features dual memory systems (semantic markdown rules and episodic conversation archives) enabling learning from past sessions, plus transparent reasoning via explainable scratchpad notebooks showing every executed cell and error. Integrates with Anthropic's Claude models for planning and coding, stores workspace context in local `.anton/` directories, and manages API credentials securely without exposing them to the LLM.