Examples

Built with K9-AIF

Two applications running live that you can try today, and focused examples that each teach one idea: routing, governance, Zero Trust, retrieval, or wrapping an existing CrewAI or LangGraph application.

Start with DAS Start Here (docs)

Live applications

Complete applications on K9-AIF, running in public. Try them, then read the code.

LiveFlagship

DAS: Defense Acquisition System

AI agents prepare the defense acquisition package; a person decides at every gate. A Service capability requirement goes from the sponsor's documents through Service validation, the Materiel Development Decision, Milestone A and the System Requirements Review, following the process in force today.

  • A versioned process model: every stage and gate cites its policy source
  • Every section screened by k9x Shield and Granite Guardian; red-team documents included
  • A verdict per criterion, with the score computed in code; an independent grader from a different model family
  • Every gate decided by a person in K9X HIL; undecided gates fail closed
  • A sample AWS GovCloud deployment guideline
Try it Code ▶ In depth (7 min) ▶ Tour (4 min)
LiveAssistant

K9Chat

A governed chat assistant, open source, on your own models. It answers from a grounded knowledge base, gives each visitor their own document projects, and protects the GPU it runs on.

  • Every message through a rate limit, k9x Shield, Guardian and the router
  • Knowledge grounding and per-visitor Projects with their own documents
  • A concurrency and GPU-thermal admission guard
  • An LLM-as-judge toggle, streaming, and an Ask API for other applications
Try it Code

Focused examples

Smaller applications in the framework repository. Each one teaches one idea you can reuse.

Starter

Weather Assist (CrewAI)

The same small app built twice: as plain CrewAI, and wrapped by K9-AIF.

Shows: keeping an existing CrewAI crew unchanged and adding K9-AIF governance, routing and observability at its boundary with the adapter.

Code
Starter

Weather Assist (LangGraph)

The sibling of Weather Assist: same app and governance, with a LangGraph graph inside.

Shows: the same adapter pattern for a different agent framework.

Code
Security

Zero Trust Execution

Zero Trust applied to execution, not only access: every agent action is evaluated before it runs.

Shows: identity, data sensitivity, destination and context risk; allow, conditional or deny decisions at agent boundaries, set by policy, not code.

Code
Pattern

ACME Support Center

A customer support workflow where one architecture hosts several agent behaviour strategies.

Shows: switching reasoning patterns by configuration (ReAct, agentic RAG, plan and execute, reflection) without changing the architecture.

Code
Domain

ACME Health Insurance

Cooperating agents for plan lookup, provider search, eligibility checks and claims support.

Shows: multiple agents, connectors and configuration-driven workflows in one business domain.

Code
Retrieval

RFP Analyzer

A large RFP document processed through the framework's built-in retrieval pipeline.

Shows: a document squad that chunks, embeds into a vector store and answers by retrieval, all governed.

Code

Reference

Full application

K9X Enterprise Insurance Operations Center

An end-to-end enterprise insurance operations platform: events on Kafka, domain orchestrators, squads of agents for PII inspection, fraud, adjudication and compliance, Zero Trust, a custom model router, a human-in-the-loop queue and audit trails. Three processes (app and UI, router, orchestrator) in a Podman pod.

Code

Small code samples

SampleWhat it contains
k9routingIntent routing: an intent orchestrator and intent agents driven by configuration (run.py, config.yaml)
myappThe smallest shape of an agent: a class, its YAML and a test
prompt_evaluationEvaluating prompts with a configuration and a test

Where to start

A path from small to complete:

  1. Weather Assist: the smallest app, and how K9-AIF wraps an agent framework you already use.
  2. Zero Trust Execution: how every agent action is governed before it runs.
  3. ACME Support Center: one architecture, several reasoning strategies, chosen by configuration.
  4. RFP Analyzer: retrieval over a large document.
  5. DAS: a complete application with a process model, screening, grading and human gates. Watch the 7-minute walkthrough, then run it.

How to run them

Each example's README has its own setup. From the root of the k9-aif-framework repository:

./run_acme_support_center.sh       # ACME Support Center
bash build.sh && bash run_eoc_pod.sh   # Enterprise Insurance Ops Center (Podman)

DAS and K9Chat live in their own repositories: k9aif/dow-k9x (with k9aif/k9x-hil for the gates) and k9aif/examples. Their READMEs list the services and settings.

The examples are practical demonstrations, not finished products: they show how K9-AIF structures governed, modular AI applications that you can extend for your own needs.