Runnable examples showing how to integrate Ancilis SDK in common AI agent patterns.
| Example | Framework | Description | Difficulty |
|---|---|---|---|
| minimal-quickstart | Python | Fastest path to first scan — 10 lines | Beginner |
| certification-driven | Python | One config line to AIUC-1 readiness | Beginner |
| data-classification | Python | Declare data types, get HIPAA/GDPR/SOC 2 controls | Beginner |
| cover-mcp-gap-assessment | MCP | Map business language to Ancilis setup and evidence gaps | Beginner |
| auto-register | Any | auto_register — instantiate one producer per installed SDK |
Beginner |
| openai-assistant | OpenAI | Wrap chat.completions.create with OpenAIActionProducer.wrap_create |
Intermediate |
| langchain-chatbot | LangChain / LangGraph | Drop-in LangChainCallbackHandler for any Runnable, Chain, or LLM |
Intermediate |
| crewai-research-crew | CrewAI | step_callback / task_callback / crew_callback — per-agent attribution |
Intermediate |
| autogen-group-chat | AutoGen / AG2 | producer.attach(agent) — auto-wire send + receive hooks |
Intermediate |
| mcp-middleware | MCP | Intercept MCP tool calls with Ancilis middleware | Intermediate |
| cli-agent | Python | CLI agent with HTTP producer | Intermediate |
pip install ancilis
cd examples/minimal-quickstart
python main.py
ancilis scan --config ancilis.yamlAll Python examples require:
pip install ancilisFramework examples (LangChain, CrewAI, AutoGen, OpenAI) drive the producers directly with framework-shape arguments, so they run without an API key or any of the framework SDKs installed — the producers are duck-typed. Once you wire a real Runnable/Crew/Agent/Client into the same producer instance, the same observations are produced for free.
Set OPENAI_API_KEY for the OpenAI example to swap the fake transport for a real chat.completions.create call.
AGPL-3.0-or-later — see LICENSE.