Collateral Description Agent

Agentic solution for describing an asset from images and estimating its value.

Versions:

• v1: Linear flow with minimal agent behavior
• v2: More agentic, flexible flow

Prerequisites

• Python 3.10+ (recommended)
• Docker (optional, for containerized run)
• An OpenAI API key and a Tavily API key

Quickstart

1. Copy the .env template and fill required variables

• Copy .env.sample to .env
• Fill the required values listed in Env vars

2. Navigate to the project root

cd collateral-description-agent

Run with Python

3. (Recommended) Create and activate a virtual environment
4. Install dependencies

pip install -r requirements.txt

5. Place images for a single asset into the data/ folder
6. Start telemetry

python -m phoenix.server.main serve

7. Start the app (change version if needed)

python app-v2/app.py loadenv   # or: python app-v1/app.py loadenv

8. Open telemetry at http://localhost:6006

Run with Docker

3. Build the image

docker build --tag collateral-agent .

4. Run the container, passing .env and mapping the images folder

docker run -p 6006:6006 --env-file .env \
  -v <folder-containing-images>:/data:ro \
  collateral-agent

3. Open http://localhost:6006

Tip

The Dockerfile is set up for v2 by default. To use v1, update the version in Dockerfile and start.sh.

Env vars

Variable name	Description
OPENAI_API_KEY	OpenAI API key
TAVILY_API_KEY	Tavily API key

Place these in .env before running the app.

Architecture

V1

V2