examples

DeepintShield examples

Minimal, runnable examples for the deepintshield Python SDK against a DeepintShield gateway. Every script is standalone, uses DeepintShield.from_env(), and shows one concept — copy/paste friendly.

Setup

1. Run a gateway (pointed at a real provider, e.g. OPENAI_API_KEY set):

   npx -y @deepintshield/ai-security        # gateway on :8080

2. Set the two env vars every example reads via from_env():

   export DEEPINTSHIELD_BASE_URL=http://localhost:8080   # default is the hosted cloud
   export DEEPINTSHIELD_VIRTUAL_KEY=sk-...               # create one in the UI (Virtual Keys)

3. Install the SDK with the extra(s) for the example you want to run. Only the deepintshield SDK is installed — the extras are its own optional provider/framework clients:

   Folder	Install
   sdk/ (core: OpenAI/Anthropic)	pip install 'deepintshield[openai,anthropic]'
   openai/ (stock OpenAI SDK, drop-in)	pip install openai httpx
   langchain/	pip install 'deepintshield[langchain]'
   langgraph/	pip install 'deepintshield[langgraph]'
   crewai/	pip install 'deepintshield[crewai]'
   pydanticai/	pip install 'deepintshield[pydanticai]'
   openai-agents/	pip install 'deepintshield[openai-agents]'
   llamaindex/	pip install 'deepintshield[llamaindex]'
   autogen/	pip install 'deepintshield[autogen]'
   litellm/	pip install 'deepintshield[litellm]'

   python sdk/01_chat_completions.py     # run any single example
   python sdk/run_all.py                 # run the whole core suite

Core SDK (sdk/)

Provider-native usage through the gateway (pip install 'deepintshield[openai,anthropic]'):

Script	Shows
01_chat_completions.py	.openai() chat completions
02_streaming.py	streaming / SSE chat
03_embeddings.py	.openai() embeddings
04_native_chat_helper.py	.chat() helper
05_multi_provider.py	.openai() / .anthropic() surfaces
06_guardrails.py	.evaluate_guardrail() (deterministic regex + PII)
07_agentic_decide.py	.agentic.decide() policy decision point
08_shield_tool.py	@shield.agentic.tool tool-call authorization
09_mcp_tools.py	shield.mcp list / call tools
10_rag_grounding.py	build_chunk + shield.rag.filter
11_virtual_keys.py	virtual-key governance headers
12_error_handling.py	typed exceptions (GuardrailDenied, DeepintShieldError)

Frameworks

Keep your framework code; just get the model/tooling from the SDK binder so traffic flows through the gateway (guardrails, RAG filtering, agentic tool-gating, identity).

Folder	Examples
openai/ (stock OpenAI SDK - no DeepintShield SDK, just base_url + a virtual key)	chat.py, rag.py, mcp.py, agentic.py
langchain/	chat.py, rag.py, mcp.py, agentic.py
langgraph/	chat.py, agentic.py (auto-governed graph), mcp.py
crewai/	chat.py, agentic.py (tool gating), mcp.py
pydanticai/	chat.py, agentic.py, mcp.py
openai-agents/	chat.py, agentic.py, mcp.py
llamaindex/	chat.py, rag.py, mcp.py
autogen/	chat.py, mcp.py
litellm/	chat.py, mcp.py

Scope

The deterministic/free tier runs against the open-source gateway. Advanced ML/LLM tiers (ML guardrail suite, agentic supply-chain, governed/OAuth MCP, RAG-security re-ranking) use the same SDK but require DeepintShield Cloud / Enterprise.