Modelpublic

FOUND-AI/found_protocol

sourceHugging Facemitupdated 1y agoView on Hugging Face

2likes

Model Card

<div align="center"> <img src="https://res.cloudinary.com/dykojggih/image/upload/v1753377308/IMG4287imd6zd.png" width="100px" alt="FOUND LABS Logo"> <h1>The FOUND Protocol</h1> <p><b>The Open-Source Engine for the Consciousness Economy</b></p>

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](https://foundprotocol.xyz), 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.

bash

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:

bash

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 foundconsciousnesslog, 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.

bibtex

@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}}
}