Marketing Agent

This repository contains a full-stack “marketing research agent” app.

Users create chat sessions (“projects”) via the FastAPI backend. For each session, the backend runs an OpenRouter/OpenAI-compatible LLM in a tool-using loop (e.g. web_search, get_content, simulate_behavior), streams progress as text/tool blocks to the frontend via Server-Sent Events (SSE), and persists the resulting blocks and project metadata in Supabase.

Capabilities

Use combined web search tool to access public data, social media and the internet for general information, and initiate sub-agents to simulate the behaviors of consumers of different profiles.

Setup

pip install -r requirements.txt

Backend environment variables

The backend reads API keys from the process environment (backend/credentials.py has no hardcoded defaults). Set these before starting the API:

Variable	Purpose
EXA_API_KEY	Exa web search
OPENROUTER_API_KEY	LLM calls via OpenRouter
SUPABASE_SECRET_KEY	Supabase server client (service role / secret key)
SUPABASE_PUBLIC_KEY	Publishable key if your code paths need it

Example:

export EXA_API_KEY="..."
export OPENROUTER_API_KEY="..."
export SUPABASE_SECRET_KEY="..."
export SUPABASE_PUBLIC_KEY="..."

API Endpoints

Method	Path	Description
GET	/api/chats	List recent chat sessions
GET	/api/work/{id}	Get work process (blocks + metadata) for a chat
GET	/api/work/{id}/stream	Stream work-process blocks via SSE
POST	/api/chats	Create a new chat session

Launch

1. Backend: cd backend && source run.txt
2. Frontend: cd agent-app && npm run dev

Note

marketing agent folder is a testing workspace, not part of the project in production