<br/><br/>

🕯️ Wax

On-device memory for iOS & macOS AI agents.

No server. No cloud. One file.

Most iOS AI apps lose their memory the moment the user closes them. Wax fixes that — giving your agents persistent, searchable, private memory that lives entirely on-device in a single portable file.

let memory = try WaxMemory(url: .documentsDirectory.appending(path: "agent.wax"))

// Store a memory
try await memory.store("User prefers concise answers and hates bullet points.")

// Retrieve the most relevant context — semantically
let context = try await memory.search("communication preferences", limit: 5)

Why Wax

Building AI agents on Apple platforms means juggling Core Data for persistence, FAISS or Annoy for vector search, and a tokenizer for context budgets — none of which talk to each other. Or you spin up Chroma or Pinecone and suddenly your app has a server dependency, network calls, and a privacy story you can't tell users.

Wax packages all of it into one self-contained file:

| Capability | Without Wax | With Wax | |---|---|---| | Document storage | Core Data / SQLite | ✅ Built-in | | Semantic search | External FAISS / Annoy | ✅ Built-in (HNSW) | | Full-text search | Another index | ✅ Built-in (BM25) | | Token budgeting | Manual | ✅ Automatic | | Crash safety | You figure it out | ✅ WAL + dual headers | | Server required | Often | ✅ Never |

Features

• Hybrid retrieval — BM25 keyword search fused with HNSW vector similarity. Gets the right memory, even when wording differs.