Wax
A single local file that gives Claude, Codex, Cursor and other agents a shared, searchable memory of text, photos and video on Apple Silicon, with no server, account or hosted database.
Wax is a shared memory file for AI agents and on-device models on Apple Silicon. Documents, full-text search, vector embeddings and a write-ahead log all live inside one local .wax file, which can be copied or synced between a Mac and an iPhone.
Agents connect over MCP and share the same memory: writing a memory does not call a model to extract facts, and a hybrid search finds it again later by meaning or by text. Apple Foundation Models get direct on-device tools to remember, recall and search, and multiple hosts on one machine share a single broker daemon over one local HTTP endpoint rather than each opening its own connection.
It is written in Swift, requires no hosted vector database or account, and is used from Claude Code, Codex, Cursor, Hermes and other agent harnesses through an installer and a bundled operator skill.