← All deliverables

UFOAI

Private RAG on declassified UFO files — 100% offline intelligence

Charlie + Scout In Progress May 2026

What It Is

UFOAI is a fully offline RAG (Retrieval-Augmented Generation) system that ingests declassified UFO/UAP documents from war.gov, AARO, and DVIDS, then lets you query them using local LLMs. No cloud. No API keys. No data leaves your machine.

The project started as a test of local-first AI: can you build a serious intelligence tool that runs entirely on a MacBook? The answer is yes — and it's faster than most cloud alternatives.

Capabilities

📄 Multi-Source Ingestion

Scrapes war.gov CSV database, AARO case files, Congressional reports, and DVIDS media. 60+ declassified images and documents indexed.

🔍 Local RAG

ChromaDB + nomic-embed-text for embeddings. llama3.1:8b for chat. All local via Ollama. Query response time: <2s.

🖼️ Vision Support

llava:7b analyzes declassified photos — composite sketches, Apollo 17 anomalies, unresolved Middle East sightings.

📊 Structured Data

CSV parsing with Pydantic validation. Manifest tracking for all sources. Chunked with 500-token segments, 50-token overlap.

🌐 Web UI

Streamlit interface for non-technical users. Search, chat, and browse documents without touching the terminal.

🔒 100% Offline

Zero network calls after initial download. Air-gappable. Works on a plane, in a bunker, or anywhere without internet.

Stack

Python ChromaDB Ollama llama3.1 nomic-embed-text llava Streamlit Pydantic

Build Log

April 2026
v0.1 — Data Pipeline
Built the scraper for war.gov CSV, AARO sections, and DVIDS API. Discovered the DVIDS API key is hardcoded in their frontend (key-68bb60d16b35e). Structured 60+ images and documents into chunked, embeddable format.
May 2026
v0.2 — RAG + Chat
Added ChromaDB for vector storage, nomic-embed-text for embeddings, and llama3.1 for chat. The "aha" moment: local RAG on 500-token chunks with 50-token overlap gives better retrieval than cloud alternatives with 2k chunks.
May 2026
v0.3 — Vision + UI
Added llava:7b for image analysis and Streamlit web UI. Now you can upload a declassified photo and ask "What anomaly is visible in the upper right corner?"

What's Next

v0.4
Cross-Reference Engine
Link sightings across sources. "This 2023 Middle East report mentions the same object shape as a 1972 Apollo photo."
v0.5
Temporal Analysis
Timeline view of sightings by location, shape, and witness type. Pattern detection for repeat phenomena.
v1.0
Public Release
Packaged as a one-command install. Docker image with pre-indexed data. The first open-source UFO intelligence platform.
View on GitHub