heimdall
View on GitHubA lightweight CPU only memory approach with ranked retrieval. Simple, yet effective.
CPU-only local memory layer for AI coding agents: tree-sitter plus local embeddings index code across repos, and a hybrid ranked kb_search returns trust-verified STRONG/WEAK/STALE hits. Integrates with Claude Code, Codex, Cursor, Windsurf and pi.
Use Cases
cross-repo code and notes memory for AI coding agentssemantic search over local source code and docsfast agent session orientation via one kb_search calltrust-verified retrieval with STRONG/WEAK/REBUILT/STALE verdictslocal private indexing of code without LLM calls or cloud uploadswiring persistent memory into Claude Code, Codex, Cursor, Windsurf, piCLI search/insert as a standalone knowledge baseingesting personal context such as notes and email into the same graph
Built With
- Language
- JavaScript
- Frameworks
- Node.js · TypeScript · SQLite · tree-sitter · llama.cpp · typebox · @earendil-works/pi-coding-agent · launchd
Tags
memory · cross-repo · embeddings · semantic-search · hybrid-retrieval · knowledge-base · local-first · cpu-only · tree-sitter · sqlite · cli · ai-agents · claude-code · developer-tools · code-intelligence · mcp