Agam
github.com/CrypticCortex/agam →Memory for AI agents that activates, not retrieves
WHAT IT SOLVES
Most 'memory for coding agents' projects are just RAG with extra embeddings. That's a database, not a brain. You don't search your memories every morning — context activates them. Agam tries the activation model instead
WHY IT'S INTERESTING
Activation, not retrieval — that's a real mental model
The HN title draws a hard line: this isn't vector-search memory dressed up. It uses a knowledge graph where context activates related nodes — closer to associative recall than database lookup. That distinction alone shows the author has actually thought about what 'memory' means
Memory needs maintenance — and they built it in
The prompts directory reveals lifecycle and temporal-drift handling: obsolete, repair, digest, upgrade. Memories aren't write-once — they decay, get merged, need repair. Most projects store and forget. This one actually thinks about memory aging
TECH GUESS
Knowledge graph backed by SQLite FTS5 for tokenization/dedup, with prompt-driven memory lifecycle management; supports both Claude Code and Codex
Discussion (0)
- No comments yet — be the first.
