Inspiration
Cloud-based AI is convenient but fundamentally flawed: it hallucinates, and it compromises user privacy by sending sensitive data to remote servers. Meanwhile, our local NAS devices have become massive digital graveyards of scattered, multi-modal files (PDFs, whiteboard photos, text notes) with no way to connect the dots. I wanted to build a system that turns local storage into a secure, intelligent brain. The inspiration was to create a "Zero-Trust AI" that acts like a forensic detective—synthesizing information but cryptographically proving its sources, ensuring 100% privacy and 0% hallucinations.
What it does
Synapse Local Detective is an autonomous multi-agent Skill for UGREEN NAS. It scans a chaotic folder and clusters related files across different formats (images, PDFs, text). It then extracts the data and generates a unified project report. The core innovation is its "Zero-Trust Synthesis": every claim generated in the final report is tied to a specific source file via a verifiable SHA-256 hash. If a claim cannot be verified by a local file, it is flagged as [UNVERIFIED] instead of being invented.
How we built it
We designed a lightweight orchestration architecture using Markdown (SKILL.md) that chains together UGREEN's native agents:
- File Agent catalogs and hashes the directory (SHA-256).
- Photo Agent performs local OCR on receipts/whiteboards.
- Document Agent extracts text from PDFs.
- Orchestrator clusters the data using a local embedding model and applies the "Zero-Trust" verification logic.
- Knowledge Agent saves the final, cited report back to the NAS. ## Challenges we ran into The biggest challenge was designing a robust synthesis logic without relying on a massive cloud LLM. Ensuring the agent can seamlessly cross-reference an OCR text from a photo with a PDF contract locally required careful orchestration of the local agents. Additionally, implementing the "Zero-Trust" cryptographic verification logic in a way that is lightweight enough to run on a NAS environment required multiple architectural iterations. ## Accomplishments that we're proud of We are incredibly proud of successfully designing a "Hallucination-Free" AI architecture. Creating a system that prioritizes data sovereignty and cryptographic truth over generative "magic" is a major step forward for local AI. We are also proud of the SKILL.md architecture, which proves that complex multi-agent forensic analysis can be orchestrated efficiently using standard local NAS capabilities. ## What we learned We learned that the future of AI is not just about generating content, but about verifying it. The integration of cryptographic hashing (SHA-256) with AI agent orchestration is a powerful way to build trust in machine-generated reports. We also learned how to optimize multi-modal data extraction and clustering for resource-constrained local environments like a personal NAS. ## What's next for Synapse Local Detective Win
