Compliance-agent

Focus on this project is to built something that creates value using as much AI tooling as possible.

Usage

Installation and setup

• Install uvx by running curl -LsSf https://astral.sh/uv/install.sh | sh
• cp ./backend/.env.example ./backend/.env
• fill in .env file

Usage

• pnpm install
• pnpm dev
• open localhost:5173 in the browser

Features

User features

• Interact through conversations
• Point to your repo
• The agent will help you identify:
  • Relevant laws and regulations to your repo
  • Data points
  • CIA of each data point (rated 1-4)
    • Confidentiality (1-4)
    • Integrity (1-4)
    • Availability (1-4)
  • Threats and vulnerabilities
  • Mitigations
• Outputs
  • Structured JSON output (source of truth) for content
  • Frontend renders the JSON output and can be interacted with directly
  • Save to markdown

Dev features

• Render zod schema as frontend and fill with JSON content
• Render any zod schema as frontend and make the JSON content editable by user
• JSON report content and current conversation state stored in browser localstorage for now
• One pnpm dev command in root to run entire application locally

Frontend view

• Chat sidebar always visible on the right, just one conversation (list of messages)
• repo url at the top
• zod schema display on the left with structured JSON content as input, zod structure as static

Technologies

• typescript
• pnpm
• trpc
• react vite webpage frontend
• tailwind css
• node express backend
• Langchain for LLM interaction
• MCP servers as langchain tools

LLM Capabilities

LLM conversation:

• Together with user form the JSON content fields

MCP capabilities:

• Web crawl
• Read repo files