# Hypernym

- **Event:** [WeaveHacks 3: Self-Improving Agents Hackathon with Weights & Biases](https://cerebralvalley.ai/e/weave-hacks-3-self-improving-agents-hackathon-with-weights-and-biases-7014fe80)
- **When:** Jan 31 at 9:00 AM – Feb 1 at 5:00 PM (PST)
- **Where:** 400 California St, San Francisco, CA
- **Team:** [Nick Lulofs](https://cerebralvalley.ai/u/hype)
- **GitHub:** https://github.com/HypernymAI/Cache
- **Demo video:** https://youtu.be/hcTZ7i29ciA
- **Gallery:** https://cerebralvalley.ai/e/weave-hacks-3-self-improving-agents-hackathon-with-weights-and-biases-7014fe80/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/weave-hacks-3-self-improving-agents-hackathon-with-weights-and-biases-7014fe80/hackathon/gallery/15

Cache: Self-Improving Agent Memory for Claude Code

  AI coding agents lose all context between sessions—successful patterns vanish, mistakes get repeated, and teams never build on their wins. Cache solves this by creating organizational memory for AI agents. How it works:
  1. Monitor - Connects to Claudestorm to track active Claude Code sessions in real-time                                                   
  2. Detect - A control panel watches for success signals: git commits, passing tests, successful deploys, user-marked "magic" checkpoints, and issues/blockers    
  3. Enrich - Each event gets tagged with tech stack, goal type, token usage, and confidence scores 
4. Store - Patterns push to Weave
  5. Retrieve - Future agents query these patterns via MCP to learn from past successes 
  The feedback loop: Agent succeeds → Cache detects → Push to Weave → Future agents query → Better outcomes
It transforms individual session wins into collective intelligence.

---

Markdown version of https://cerebralvalley.ai/e/weave-hacks-3-self-improving-agents-hackathon-with-weights-and-biases-7014fe80/hackathon/gallery/15. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt