christopherkarani/Wax — Single-file memory layer for AI agents, sub mili-second RAG on Apple Silicon.

Description and use

**Brief Summary (2‑3 sentences)**
Wax is a single‑file, pure‑Swift memory layer that enables sub‑millisecond Retrieval‑Augmented Generation (RAG) for AI agents on Apple‑Silicon devices, running entirely on‑device with Metal acceleration and no external servers or APIs. It provides a lightweight, standardized protocol for connecting AI assistants to tools, data sources, and Model Context Protocol (MCP) servers, making it easy to embed real‑world knowledge and functionality directly into iOS/macOS apps. With 730 ★, recent commits, and strong community interest, Wax is positioned as a production‑ready OSS component for on‑device AI workloads.

**Value**

• **Speed & Privacy:** Metal‑optimized, sub‑ms RAG eliminates network latency and keeps user data on the device, satisfying privacy‑by‑design requirements.
• **Simplicity:** A single Swift file means minimal integration overhead—no build‑system gymnastics, no external dependencies, and no need to manage server infrastructure.
• **Standardization:** Implements the Model Context Protocol, giving developers a common contract for tool and data integration that can be reused across different agents and services.

**Practical Adoption Path**

1. **Evaluate:** Clone the repo, run the provided CLI or Swift SDK demo, and benchmark retrieval latency on your target Apple‑Silicon hardware.
2. **Integrate:** Add the single Swift file to your Xcode project, configure the memory layer with your data embeddings, and invoke the RAG API from your AI assistant code.
3. **Extend:** Use the exposed protocol to register custom tool adapters or MCP servers, enabling the assistant to call external services while staying on‑device.
4. **Deploy:** Ship the app through the App Store; the entire stack remains self‑contained, requiring no backend changes.

**Production Readiness**

• **Activity & Adoption:** Recent commits (as of 2026‑05‑11), 730 ★, 42 forks, and active issue discussions indicate a healthy maintainer community.
• **Stability:** The single‑file architecture reduces build‑time bugs and dependency conflicts; Metal acceleration is well‑tested on Apple Silicon.
• **Risk Considerations:** License compliance, security audit, and maintainer continuity should be verified before large‑scale rollout, but no major red flags are evident.

Overall, Wax offers a fast, private, and low‑friction way to equip on‑device AI agents with real‑time knowledge and tool access, making it a strong candidate for production pilots in iOS/macOS AI products.

Problem

christopherkarani/Wax helps connect AI assistants to real tools and data through a standard protocol.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• connect AI agents to tools
• ship Model Context Protocol servers
• standardize integrations

Evidence

Single-file memory layer for AI agents, sub mili-second RAG on Apple Silicon. Metal Optimized On-Device. No Server. No API. One File. Pure Swift

Source