ManePaw

Product Introduction

1. Definition: ManePaw is a native macOS application leveraging local artificial intelligence (AI) for private document retrieval and multimodal chat. It operates as an offline RAG (Retrieval-Augmented Generation) system, processing text, code, images, and audio entirely on-device.
2. Core Value Proposition: It eliminates cloud dependency for AI-powered knowledge management, ensuring zero data exposure, no mandatory accounts, and complete user privacy while enabling semantic search and contextual chat with personal files.

---

Main Features

1. Local AI Processing:

   • How it works: Integrates Ollama to run open-source LLMs (e.g., qwen2.5) locally. All AI computations—embedding generation, inference, and transcription—execute on the user’s Mac hardware.
   • Technologies: SwiftUI for native UI, Metal for GPU acceleration, Ollama for model orchestration.

2. Multimodal Indexing:

   • How it works: Automatically ingests and indexes diverse file types:
     • Text/Code: Chunks documents (PDF, Markdown, code) and extracts semantic embeddings using LanceDB.
     • Images: Generates AI captions via local vision models.
     • Audio: Transcribes speech to text using offline ASR (Automatic Speech Recognition).
   • Supported Formats: .txt, .md, .py, .js, .png, .jpg, .mp3, .wav, and 20+ others.

3. Project-Aware Semantic Search:

   • How it works: Detects codebases via manifest files (package.json, Cargo.toml) and indexes function/class signatures. Uses vector similarity search in LanceDB to retrieve results by contextual meaning, not just keywords.
   • Search Scope: Cross-file queries (e.g., "Find authentication middleware in my Node.js projects").

4. RAG-Powered Chat:

   • How it works: For each query, retrieves relevant document snippets using semantic search, then feeds context to the local LLM to generate sourced responses. Citations link to original files.
   • Use Case: Ask "Summarize my Rust API documentation" to get AI summaries with linked source files.

---

Problems Solved

1. Pain Point: Sensitive data exposure in cloud-based AI tools (e.g., ChatGPT, Gemini). ManePaw ensures confidential documents, proprietary code, or private media never leave the device.
2. Target Audience:
   • Developers needing private codebase Q&A.
   • Researchers handling confidential data.
   • Privacy-conscious professionals (legal, healthcare) requiring offline document analysis.
3. Use Cases:
   • Auditing code for security flaws without uploading to third parties.
   • Searching meeting recordings via transcribed audio.
   • Querying internal wikis on air-gapped networks.

---

Unique Advantages

1. Differentiation vs. Competitors:

   Feature	ManePaw	Cloud Tools (e.g., Dropbox AI)
   Data Location	On-device	Remote servers
   Internet Requirement	None	Mandatory
   Pricing	One-time purchase (free)	Subscription-based

2. Key Innovation: Hybrid native-local architecture—SwiftUI frontend + NestJS backend + LanceDB—enables complex RAG workflows entirely offline. Native macOS APIs enable file system integration and Metal-accelerated AI.

---

Frequently Asked Questions (FAQ)

1. Is ManePaw truly private?
   Yes. All data processing occurs locally—no telemetry, cloud uploads, or external servers. Files are stored in ~/Library/Application Support/ManePaw.

2. What file types can ManePaw search?
   It supports text (.txt, .md), code (.js, .py, .rs), images (.png, .jpg), and audio (.mp3, .wav). See README for full list.

3. Which macOS versions are supported?
   Requires macOS Sonoma (14+) or later due to SwiftUI 5 and Metal 3 dependencies for on-device AI.

4. How to use custom Ollama models with ManePaw?
   Pull any Ollama-supported model (e.g., llama3, mistral), then configure the backend config.json to point to your local model.

5. Can ManePaw index entire code repositories?
   Yes. It auto-detects projects via package.json (Node.js), Cargo.toml (Rust), and other manifests, indexing code structure for semantic queries like "Find database schema handlers."