DAF — Deterministic Agentic Framework

The model proposes. The system governs.

Policy-Based Agentic Systems (PBAS) — A framework for building agentic AI applications where the LLM generates plans but a deterministic policy engine governs execution.

Python Open Source Apache 2.0

Documentation

Comprehensive guides and references for the DAF framework

📚

API Reference

Complete programmatic API documentation for all DAF components, classes, and methods.

Read Documentation →
🏗️

Design Philosophy

Eight core design principles behind DAF and how they influence the framework architecture.

Read Documentation →
👨‍💻

Developer Guide

Step-by-step guide to extending DAF, implementing custom tools, agents, and policies.

Read Documentation →

Quick Start

Get up and running with DAF in 2 minutes with mock examples and boilerplates.

View Quick Start →
📁

Repository Structure

Understand the organization of the DAF repository and locate key components.

View Structure →
🚀

Seven Boilerplates

Ready-to-run templates for common use cases: contract review, report generation, compliance, and more.

Explore Boilerplates →

Core Features

🎯 Deterministic Governance

No LLM calls in policy evaluation. Either the plan conforms to your PolicyMatrix or it doesn't.

🔐 Separation of Concerns

LLM handles cognition (planning), policy engine handles governance, execution runs in scoped context.

📋 Policy-Based Control

Write governance rules in YAML. Change policies without modifying code.

🔄 Self-Correcting Loops

When plans violate policies, the system provides context for re-planning without manual intervention.

🚀 Mock-First Testing

Develop and test with mock LLM clients. No API key required. Seamless switch to production.

📦 Ready-to-Run Boilerplates

Seven production-grade templates: contract review, report generation, support triage, and more.

Repository Overview

561+

Tests (Unit, Adversarial, Integration)

7

Production Boilerplates

99.9%

Python Codebase

Apache 2.0

Open Source License

Ready to Explore DAF?

Clone the repository and run your first governed agentic loop in minutes.