AI-NUX

Inspiration

As developers in the age of generative AI, our intelligence is scattered. We have brilliant conversations locked in ChatGPT history, system designs in Claude, codebase context in GitHub Copilot, and fleeting thoughts in local markdown files.

There is no single source of truth. We are constantly context-switching, losing valuable prompts and reasoning loops. We built this project to solve this "AI fragmentation" problem—creating a local-first, self-hosted Personal Knowledge Operating System that aggregates all AI chat logs, developer tools, and note-taking apps into a single, unified "Second Brain."

What it does

Our project acts as a secure, decentralized intelligence hub. It continuously ingests and parses heterogeneous data—including official ChatGPT/Gemini chat exports, local Markdown vault notes, and API call logs.

Once ingested, the system embeds and indexes this data, allowing users to perform blazing-fast semantic searches across their entire digital footprint, ask questions to their historical self, and recall exactly how a specific bug was resolved three months ago in an AI chat.

How we built it

We architected a robust, privacy-centric stack using the following open-source and infrastructure components:

• Everos Core & Docker: Containerized, local-first database architecture ensuring absolute ownership of personal data.
• Vector Search Engine (ChromaDB): Text chunks are vectorized using state-of-the-art embedding models. Semantic retrieval is powered by cosine similarity calculations: $$\text{Similarity} = \cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{|\mathbf{A}| |\mathbf{B}|}$$
• FastAPI & NiceGUI: High-performance asynchronous backend with a sleek, responsive Python-based local web UI.
• Cloudflare Zero Trust & Custom SAML SSO: To make our workspace securely accessible on any device without exposing open ports, we implemented custom Single Sign-On (SSO) routed through a Cloudflare Tunnel, complete with domain validation.

Challenges we faced

• Heterogeneous Data Normalization: Chat exports from different AI platforms use wildly different JSON/HTML schemas. We had to build an extensible pipeline of parsers to recursively clean and format threads into standardized Markdown chunks without losing the conversational hierarchy (User vs. System).
• Secure Multi-Device Synchronization: Balancing convenience (remote access) with absolute privacy (zero-trust). We solved this not by hosting a vulnerable cloud database, but by shielding our local server behind a Cloudflare Tunnel with strict identity validation.

Accomplishments that we're proud of

• Zero Data Leakage: Successfully built an ultra-smart second brain where all indexing, vector storage, and search queries happen 100% locally or via secure private tunnels.
• Semantic Recall: We can now query our historical AI chat logs with natural language and get precise, contextual references within milliseconds.

What's next

We plan to build a local background daemon that auto-syncs chat histories from browser extensions and local IDEs, making the ingestion process completely invisible and real-time.

Built With

• chromadb
• cloudflare
• docker
• everos
• fastapi
• knowledge-graph
• nicegui
• python
• saml-sso
• vector-search
• zero-trust