Policy Foundry

AI-powered firewall policy analysis and recommendation engine. Feed it traffic data and get back validated, risk-assessed firewall rule proposals with change request exports.

Input: Excel traffic exports or VPC Flow Logs (local/S3) Output: Terminal display, JSON, or change request forms (xlsx/pdf)

  Traffic Data  ──▶  AI Analysis Pipeline  ──▶  Change Request
  (Excel/Logs)       Analyze → Assess →         (xlsx / pdf /
                     Generate → Validate →       terminal / JSON)
                     Decide

Installation

Option 1: Install from PyPI (recommended)

pip install policy-foundry

Option 2: Install with pipx (isolated environment)

pipx install policy-foundry

Option 3: Install with uv

uv tool install policy-foundry

Verify installation

policyfoundry --help

Setup

PolicyFoundry uses an LLM to analyze traffic. You need one of these providers configured:

Using Ollama (free, local, default)

1. Install Ollama
2. Pull a model:

   ollama pull llama3.2
   

3. That's it — PolicyFoundry uses Ollama by default.

Using OpenAI

Set your API key as an environment variable:

export POLICYFOUNDRY_LLM__PROVIDER=openai
export POLICYFOUNDRY_LLM__MODEL=gpt-4o
export POLICYFOUNDRY_LLM__API_KEY=sk-your-key-here

Using AWS Bedrock

export POLICYFOUNDRY_LLM__PROVIDER=bedrock
export POLICYFOUNDRY_LLM__MODEL=anthropic.claude-3-sonnet-20240229-v1:0

Requires AWS credentials configured via aws configure or environment variables.

Using a config file

Create .policyfoundry.yaml in your working directory:

llm:
  provider: ollama
  model: llama3.2
  temperature: 0.1
  max_tokens: 4096
  timeout: 120

Usage

Analyze an Excel traffic export

policyfoundry analyze --source excel --file traffic.xlsx

Export a change request form

# Excel change request
policyfoundry analyze --source excel --file traffic.xlsx --export xlsx

# PDF change request
policyfoundry analyze --source excel --file traffic.xlsx --export pdf

# Both
policyfoundry analyze --source excel --file traffic.xlsx --export xlsx,pdf

Use a custom change request template

policyfoundry analyze --source excel --file traffic.xlsx \
  --export xlsx --template my-template.xlsx

A sample template is included in the repo.

Analyze VPC Flow Logs

# From local files
policyfoundry analyze --source local --sg-ids sg-0123456789abcdef0

# From S3
policyfoundry analyze --source s3 --sg-ids sg-0123456789abcdef0

Get JSON output

policyfoundry analyze --source excel --file traffic.xlsx --format json

View current firewall rules

policyfoundry rules --sg-id sg-0123456789abcdef0

Show resolved configuration

policyfoundry config

CLI Reference

policyfoundry analyze

policyfoundry rules

policyfoundry config

Configuration

PolicyFoundry merges configuration from multiple sources (highest priority wins):

1. CLI flags
2. Environment variables (POLICYFOUNDRY_ prefix, __ for nesting)
3. Local YAML (.policyfoundry.yaml in current directory)
4. Global YAML (~/.policyfoundry/config.yaml)

Full config example

llm:
  provider: ollama          # ollama | openai | bedrock
  model: llama3.2
  temperature: 0.1
  max_tokens: 4096
  timeout: 120

sources:
  log_paths:
    - /var/log/vpc-flow/*.log
  # s3_bucket: my-vpc-logs-bucket
  # s3_prefix: vpc-flow-logs/

targets:
  security_group_ids:
    - sg-0123456789abcdef0

excel:
  # sheet_name: null
  # header_row: 1

output:
  format: rich
  data_dir: ~/.policyfoundry/data

Environment variable examples

export POLICYFOUNDRY_LLM__PROVIDER=openai
export POLICYFOUNDRY_LLM__MODEL=gpt-4o
export POLICYFOUNDRY_LLM__API_KEY=sk-...
export POLICYFOUNDRY_SOURCES__S3_BUCKET=my-vpc-logs
export POLICYFOUNDRY_TARGETS__SECURITY_GROUP_IDS=sg-abc123,sg-def456

Docker

Run PolicyFoundry with an Ollama sidecar:

docker compose up -d
docker compose exec ollama ollama pull llama3.2
docker compose run policyfoundry analyze --source excel --file /path/to/traffic.xlsx

Contributing

Build from source

git clone https://github.com/vahagn-madatyan/PolicyFoundry.git
cd PolicyFoundry

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install with dev dependencies
uv sync --group dev

# Run from source
uv run policyfoundry --help

Run tests

uv run pytest
uv run pytest --cov=policyfoundry
uv run pytest -k "test_analyze"

Project structure

src/policyfoundry/
├── main.py                  # CLI app (analyze, rules, config)
├── adapters/                # Firewall vendor adapters (AWS SG, etc.)
├── analysis/                # Traffic analysis & aggregation
├── config/                  # Configuration management
├── ingestion/               # Data ingestion (local, S3, Excel)
├── pipeline/                # AI analysis pipeline (LangGraph)
├── storage/                 # Parquet persistence + DuckDB queries
├── output/                  # Terminal & JSON formatters
└── export/                  # Change request export (xlsx/pdf)

Architecture

PolicyFoundry runs a 5-stage AI pipeline built on LangGraph:

LLM calls use Instructor + LiteLLM for structured output — every call returns a validated Pydantic model, not free-form text.

Adapter system

Firewall adapters are loaded via Python entry points. The included AWS Security Group adapter validates against AWS-specific constraints (allow-only rules, 60-rule limit, CIDR validation). All adapters are wrapped in a read-only safety layer — PolicyFoundry never modifies live firewall rules.

Infrastructure (optional)

The infra/terraform/ directory contains Terraform for a test environment (VPC, security groups, S3 bucket, flow logs):

cd infra/terraform
terraform init
terraform plan -var="name_prefix=policyfoundry-dev"
terraform apply

License

Apache License 2.0