christopherkarani/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 it solves

Wax provides a local-first, shared memory system for AI agents and on-device models. It eliminates the need for hosted vector databases or manual markdown scratchpads (like CLAUDE.md) by allowing multiple agents—such as Claude, Cursor, and Hermes—to read and write to a single, searchable .wax file on disk.

How it works

Wax uses a single binary file that combines Full-Text Search (FTS5), CoreML vectors, and a Write-Ahead Log (WAL). It operates as an MCP (Model Context Protocol) server, allowing various AI hosts to connect via HTTP to a shared memory store. For Apple ecosystem users, it integrates directly with Apple Foundation Models and provides a Swift framework for native iOS and macOS app integration, utilizing a built-in MiniLM embedder for hybrid search.

Who it’s for

  • AI Agent Users: People using tools like Claude Code, Cursor, Codex, and Hermes who want their agents to remember preferences and lessons across different sessions and projects.
  • iOS/macOS Developers: Developers building on-device AI applications that require durable, local memory without relying on cloud APIs.

Highlights

  • Local-First: No accounts or hosted databases; memory is stored in a single portable .wax file.
  • Cross-Agent Sharing: Multiple different AI agents can share the same memory store via a single MCP server.
  • Hybrid Search: Combines text and vector search for fast retrieval (p95 recall around 6.1 ms on M-series chips).
  • Apple Ecosystem Integration: Native support for Apple Foundation Models and a Swift package for app developers.

Related

  • Project
    mcncarl/agent-memory-vault

    A local-first, Git-backed long-term memory system that allows AI coding agents like Claude Code and Codex to share a shared, verifiable Markdown-based memory vault.

  • Project
    Fergana-Labs/stash

    A persistent knowledge base and shared memory layer for AI agents that connects to various data sources and records session transcripts to maintain context across runs.

  • Project
    memvid/memvid

    A portable, single-file memory layer for AI agents that provides persistent long-term memory and instant retrieval without requiring a database.

  • Project
    AVIDS2/memorix

    Memorix is a Node.js library that adds a local‑first, searchable memory store to any AI coding assistant. It persists facts, design rationales, and Git‑derived insights in a SQLite database, exposing them via MCP, a CLI, and an optional web UI. By installing a single `memorix setup --agent <name>` command, dozens of agents (Claude Code, Codex, Copilot CLI, Cursor, Gemini CLI, etc.) can share the same project‑scoped memory, enabling smoother handoffs, better context retention, and multi‑agent orchestration.

  • Project
    FerroxLabs/wayland

    A local-first desktop AI agent command center that orchestrates multiple AI CLIs and agents from one interface with shared persistent memory.