CRM Lead Agent

CRM Lead Agent is an AI assistant for lead retrieval, scoring, and prospecting email generation.
It uses an agent-based architecture combined with RAG to search a local vector database of leads, rank them, and provide ready-to-send email suggestions.

The system runs entirely locally using Ollama for LLM inference — ensuring privacy and security with no external API calls.

Project uses langchaingo library.

---

Features

• Fuzzy lead search with vector embeddings
• Automated lead filtering and ranking
• AI-generated prospecting emails
• Local LLM execution — no API keys required
• CLI and optional web interface
• Fully self-contained using Docker for dependencies

---

How it Works

1. Markdown source data of leads and company product is embedded and stored in a Chroma vector store.
2. The agent pipeline:
   • Filters relevant leads based on a query
   • Searches the vector store for matches
   • Scores and ranks the best lead
   • Generates a suggested email for outreach
3. The LLM and embedding models are served locally through Ollama.

---

Requirements

• Docker & Docker Compose
• Go 1.21+
• Node.js & npm (optional, for web UI)

---

Installation & Setup

Clone the repository:

git clone https://github.com/nebojsaj1726/crm-agent.git
cd crm-agent

Create .env file

Build and setup the environment:

make setup

This will:

• Start Docker containers for Ollama and Chroma

• Pull the LLM and embedding models

• Seed the vector store with sample leads

---

Usage

make query

This will:

• Prompt for a fuzzy lead description

• Return the top match, a lead score with reasoning, and a draft prospecting email

Optionally, you can open web UI in browser on localhost with:

make web

Performance Notes

• On CPU-only systems, responses may take up to 1–2 minutes for large models.

• For <30 second responses, a GPU with sufficient VRAM is recommended.