← Platform overview
A — PERSISTENT MEMORY
Your agent, remembered.
Every project decision, every file structure, every stack choice — stored in local SQLite. Recalled in milliseconds.
Start remembering →
nexus-prime setup claude 7.8×
token compression
24×
faster memory access vs cloud
99.4%
reduction in re-reads
<10ms
memory recall latency
Claude Code, Cursor, and Codex all start fresh every session. They re-read your codebase, re-learn your conventions, re-understand your architecture. One developer tracked 99.4% of 100M tokens going to input re-reading alone.
Nexus Prime intercepts every agent session and writes structured memory records to a local SQLite database at ~/.nexus-prime/. The next session — same day or six months later — loads that memory in milliseconds. No cloud. No API calls. No token waste.
Project-scoped memory
Memory is namespaced per Git repo. Your monorepo and your side project never cross-contaminate.
Structured recall
Nexus stores file relationships, architectural decisions, code patterns, and convention rules — not raw chat logs.
Local SQLite, always
Your memory lives in ~/.nexus-prime/ on your machine. Zero data egress. Backup with cp.
Cross-agent reads
Memory written by Claude Code is readable by Cursor. One knowledge base, every agent.