AstroBob: Self-Teaching Memory for AI Agents

AI coding agents are powerful, but they suffer from amnesia. Every new session, they forget your conventions, your workflows, your hard-won bug fixes — forcing developers to re-explain context endlessly. AstroBob solves this with a self-teaching memory system built for AI coding agents. It introduces three types of memory inspired by human cognition: semantic (stable facts about your project), episodic (events that happened), and procedural (reusable playbooks). Agents store experiences as they work, recall relevant context on demand, and reflect on past episodes to write their own skills automatically. Built on an open MCP protocol and backed by AstraDB with NVIDIA vector embeddings, AstroBob works with any MCP-compatible agent — IBM Bob, Claude Code, Cursor, and beyond. It's not tied to a single vendor or use case. Developers use it for coding conventions. Testers use it for edge cases. Business analysts use it for decision logs. The result: an agent that compounds knowledge with every session, evolving from generic assistant into domain expert. One memory layer. Every agent. Every role.