Nexus
Persistent memory for AI agents
Overview
A CLI tool that gives AI agents persistent memory across sessions. One binary, no daemon, no dependencies beyond Go. Solves context loss between sessions by tracking projects, sessions, decisions, preferences, and learnings with semantic search.
Architecture
Go binary with embedded SQLite and FTS5 for full-text search. The CLI stores projects, sessions, decisions, preferences, and learnings as structured records. Semantic search uses SQLite's built-in FTS5 with BM25 ranking. A companion MCP server wraps the CLI commands into tool calls, giving Claude Code drop-in memory.
The Story
Born from the frustration of telling Claude Code the same project context every session. The original prototype was a Python script that wrote to a JSON file. Rewriting it in Go was the turning point — the single-binary deployment model meant it could be dropped into any CI/CD pipeline without dependency hell. The MCP server wrapper came later, turning a good CLI into an essential tool for AI-assisted development.
Technology
Links
Features
- Cross-session memory for AI coding agents
- Project and session tracking with semantic search
- Decision and preference learning over time
- FTS5 full-text search with relevance ranking
- Single binary with zero runtime dependencies
Highlights
- Go CLI with no runtime dependencies
- MCP server wrapper for Claude Code integration