Kijito builds a 'prosthetic hippocampus' for LLMs, cutting memory tokens ~20x
jasonridesabike · reddit · 2026-10-24
An engineer spent a year building Kijito.ai, a persistent associative memory engine for LLMs modeled on how the human hippocampus works. Instead of exhaustively rereading notebooks of context, agents recall a handful of meaning-associated memories — including when they were wrong and what they learned. He reports it at times beats competitors on leading memory benchmarks while using 20x fewer tokens; in Claude Code it adds only 1.5k tokens at session start, letting him run more agents on a single Max 20 subscription. Under the hood: an extended embedded graph/vector database, a quantized embedding model, and an MCP-compatible agent layer. Memories are encrypted at rest; updates ship in a midnight window, restarts hold requests instead of dropping them, and nightly encrypted backups are restore-tested. Full benchmark results coming soon.
More from coding & agent
- Screenshot shows multi-agent AI team with a Chief of Staff that self-corrects — RachelVT42 · 2026-10-24
- This all-in-one ESP32 pocket dev board is programmable by Codex and fits on a keychain — thesephist · 2026-10-24
- Desktop ComfyUI was unusable on phones, so he built a free open-source mobile UI — Deneme123deneme · 2026-10-24
- Open-source coding agent Kvist tackles crash recovery between file writes and checkpoints — ChatGPT_Codex · 2026-10-24
- Security veteran: 0day patch windows are now hours, security debt interest just spiked — xeophon · 2026-10-24
- Mubit open-sources Banyan Sapling, a verification model with calibrated probabilities (Apache 2.0) — testingcatalog · 2026-10-24