Autonomous Marketing AI Agent

An autonomous multi-agent system engineered to plan, execute, and inspect digital marketing campaigns without constant human intervention.

This project implements a self-correcting cognitive architecture using LangGraph, where specialized agents collaborate to achieve high-level business goals. It moves beyond simple "chains" to a stateful, cyclic workflow capable of error recovery and iterative optimization.

System Architecture

The core logic is decoupled into three distinct agentic roles, ensuring separation of concerns and modular scalability:

1. Planner Agent (Strategy)

• Responsibility: Deconstructs broad user goals (e.g., "Launch Q3 efficiency campaign") into tactical steps.
• Logic: Utilizes LLMs to generate valid JSON-based campaign structures, selecting appropriate platforms and budget allocations based on historical performance data.

2. Executor Agent (Action)

• Responsibility: The side-effect handler. Interacts with external APIs (Twitter/X, LinkedIn) and databases.
• Logic: Executes the Planner's step. If an API call fails, it can self-correct or request an updated plan, preventing total system failure from transient errors.

3. Evaluator Agent (Optimization)

• Responsibility: Closes the feedback loop.
• Logic: Analyzes output against the initial goal. It writes "Lessons Learned" into a Vector Memory store, allowing the system to improve its planning logic over time (e.g., learning that "Threads do better on Tuesdays").

Technical Stack

• Orchestration: LangGraph - for building stateful, cyclic agent workflows.
• Memory: Vector Database integration for semantic retrieval of past campaign performance.
• API: FastAPI - providing a robust, async interface for the frontend or external triggers.
• Database: PostgreSQL & SQLModel - for persistent meaningful state (campaigns, ads, users).
• Runtime: Python 3.11+

Quickstart

1. Environment Setup

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt  # (Ensure you generate this from pyproject.toml)

2. Configuration Copy .env.example to .env and configure your API keys (Anthropic, OpenAI, Twitter/X, etc.).

3. Run the System

   uvicorn src.marketing_agent.api:app --reload

4. Trigger a Campaign

   curl -X POST http://localhost:8000/campaign/create \
        -H "Content-Type: application/json" \
        -d '{"goal": "Promote the new open-source vector DB features", "platforms": ["twitter", "linkedin"]}'

Development

The specific agent logic can be found in src/marketing_agent/nodes.

• planner.py: Strategy generation prompting.
• executor.py: Tool definitions and API wrappers.
• evaluator.py: Analysis logic and memory commits.

---

Build with ❤️ by Kaftrandev