Wax
Shared Single-file memory layer for all your agents, sub mili-second RAG over text, photo and video on Apple Silicon.. No Server. No API. One File. Pure Swift
What is Wax?
Shared Single-file memory layer for all your agents, sub mili-second RAG over text, photo and video on Apple Silicon.. No Server. No API. One File. Pure Swift
Overview of Wax
What it is
Wax is a shared single-file memory layer for agents and on-device models. It provides a local .wax file that can be used by Claude, Codex, Hermes, Cursor, and Apple Foundation Models. It offers searchable memory on disk with no account, no hosted vector database, and no server. It is written in pure Swift and designed for Apple Silicon.
Key features
- Shared single-file memory layer for multiple agents (Claude, Codex, Cursor, Hermes, etc.)
- Sub-millisecond RAG over text, photo, and video on Apple Silicon.
- No server, no API, one file, pure Swift.
- Uses MCP (Model Context Protocol) for communication between agents and the memory file.
- Supports FTS5 text search, CoreML vectors, and WAL inside the file.
- iCloud or AirDrop the file to another Mac or iPhone.
rememberdoes not call an LLM; stores sentences for hybrid search.- Apple Foundation Models get on-device tools: waxRemember, waxRecall, waxSearch.
- Hermes can use native wax-memory provider.
Use cases
- Multiple agents (Claude, Codex, Cursor, Hermes) reading and writing the same memories.
- On-device memory for Apple Foundation Models.
- Survives chat sessions; memory persists across sessions.
- No hosted vector DB, no account needed.
Getting started
bash npx -y waxmcp@latest install After installation, stage the server and teach the model when to use Wax. See the repository README for detailed host-specific wire-up instructions.
License and pricing
Licensed under Apache-2.0. Open source.
Based on the project's public README and repository metadata.