Sharing a magical tool to give AI Agents a "brain"!

Beads (gastownhall/beads)
This is an open-source project by Silicon Valley tech expert Steve Yegge.
📌 What pain points does it solve?
Everyone using AI to write code (like Claude Code, Cursor) has definitely encountered this crash:
Yesterday, I discussed a complex architecture with AI all day, and today when I open the terminal, it forgets everything!😵
AI's context window is limited, and it's hard to maintain long-term memory across sessions, making it impossible to autonomously advance long-term complex tasks.
📌 How does Beads work?
Simply put, it’s an external memory bank for an AI agent.
- Task as code: It stores tasks in JSONL format within the project directory.
- Git management: You can manage tasks like code—using Git branches, merges, conflict resolution to track progress.
- Dolt-driven: Uses Dolt (a version-controlled SQL database) at the core, supporting cell-level merges.
📌 Core features:
✅ Persistent sessions: AI will no longer have a "goldfish memory," and can pick up where it left off yesterday.
✅ Structured dependencies: Tasks have dependencies, so AI knows what to do first and what to do next.
✅ Multi-agent collaboration: Multiple AI agents can collaborate based on the same memory bank.
✅ Semantic auditing: All operations are traceable.
✅ Memory decay: Automatically compresses old tasks to save context window space.
If you often develop with AI agents and find their memory too poor, this tool is definitely worth exploring!
The project link is in the comments. 👇