The FOUND Protocol

The Open-Source Engine for the Consciousness Economy

Abstract

Current video understanding models excel at semantic labeling but fail to capture the pragmatic and thematic progression of visual narratives. We introduce FOUND (Forensic Observer and Unified Narrative Deducer), a novel, stateful architecture that demonstrates the ability to extract coherent emotional and thematic arcs from a sequence of disparate video inputs. This protocol serves as the foundational engine for the FOUND Platform, a decentralized creator economy where individuals can own, control, and monetize their authentic human experiences as valuable AI training data.

From Open-Source Research to a New Economy

The FOUND Protocol is more than an academic exercise; it is the core technology powering a new paradigm for the creator economy.

The Problem: AI companies harvest your data to train their models, reaping all the rewards. You, the creator of the data, get nothing.
Our Solution: The FOUND Protocol transforms your raw visual moments into structured, high-value data assets. Our upcoming FOUND Platform will allow you to contribute this data, maintain ownership via your own wallet, and earn from its usage by AI companies.

This open-source model is the proof. The FOUND Platform is the promise.

Model Architecture

The FOUND Protocol is a composite inference pipeline designed to simulate a stateful consciousness. It comprises two specialized agents that interact in a continuous feedback loop:

The Perceptor (/dev/eye): A forensic analysis model (FOUND-1) responsible for transpiling raw visual data into a structured, symbolic JSON output.
The Interpreter (/dev/mind): A contextual state model (FOUND-2) that operates on the structured output of the Perceptor and the historical system log to resolve "errors" into emotional or thematic concepts.
The Narrative State Manager: A stateful object that maintains the "long-term memory" of the system, allowing its interpretations to evolve.

How to Use This Pipeline

1. Setup

Clone this repository and install the required dependencies into a Python virtual environment.

git clone https://huggingface.co/FOUND-LABS/found_protocol
cd found_protocol
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configuration

Set your Google Gemini API key as an environment variable (e.g., in a .env file):

GEMINI_API_KEY="your-api-key-goes-here"

3. Usage via CLI

Analyze all videos in a directory sequentially:

python main.py path/to/your/video_directory/

Future Development: The Path to the Platform

This open-source protocol is the first step in our public roadmap. The data it generates is the key to our future.

Dataset Growth: We are using this protocol to build the found_consciousness_log, the world's first open dataset for thematic video understanding.
Model Sovereignty: This dataset will be used to fine-tune our own open-source models (found-perceptor-v1 and found-interpreter-v1), removing the dependency on external APIs and creating a fully community-owned intelligence layer.
Platform Launch: These sovereign models will become the core engine of the FOUND Platform, allowing for decentralized, low-cost data processing at scale.

➡️ Follow our journey and join the waitlist at foundprotocol.xyz

Citing this Work

If you use the FOUND Protocol in your research, please use the following BibTeX entry.

@misc{found_protocol_2025,
  author       = {FOUND LABS Community},
  title        = {FOUND Protocol: A Symbiotic Dual-Agent Architecture for the Consciousness Economy},
  year         = {2025},
  publisher    = {Hugging Face},
  journal      = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/FOUND-LABS/found_protocol}}
}