Agent framework integrations¶

Full guides: integrations/README.md — per-framework wiring, extras, and the 72-angle security matrix.

Beginners: GETTING_STARTED.md · REVIEW vs BLOCK: AGENT_ACTIONS.md

Unplug ships framework-agnostic hooks first. Install only the extra for your stack:

pip install unplug-ai                      # core — no agent SDK deps
pip install "unplug-ai[langgraph]"         # LangGraph
pip install "unplug-ai[crewai]"            # CrewAI
pip install "unplug-ai[integrations]"      # all documented agent/RAG extras

Core: `AgentHooks`¶

from unplug import Guard
from unplug.integrations.hooks import AgentHooks

hooks = AgentHooks(Guard())  # or Guard(mode="server") for hosted API

decision = hooks.scan_user_input("user message")
if not decision.allowed:
    raise RuntimeError(decision.message)

wrapped, rag_decision = hooks.wrap_retrieved_content(webpage_html)
tool_decision = hooks.before_tool_call("send_email", {"to": "...", "body": "..."})
if tool_decision.needs_review:
    pause_for_operator(tool_decision)  # see AGENT_ACTIONS.md
elif not tool_decision.allowed:
    raise RuntimeError(tool_decision.message)

Hook	When to call
`scan_user_input`	Before LLM turn
`wrap_retrieved_content`	After RAG / web fetch
`before_tool_call`	Before every tool (always local)
`scan_agent_output`	Before returning to user
`scan_request_isolated`	Eval / probes (no session bleed)

Framework modules¶

Framework	Extra	Module
LangGraph	`langgraph`	`unplug.integrations.langgraph`
OpenAI Agents SDK	`openai-agents`	`unplug.integrations.openai_agents`
LangChain	`langchain`	`unplug.integrations.langchain`
Google ADK	`google-adk`	`unplug.integrations.google_adk`
smolagents	`smolagents`	`unplug.integrations.smolagents`
DSPy	`dspy`	`unplug.integrations.dspy`
Strands Agents	`strands`	`unplug.integrations.strands`
Letta	`letta`	`unplug.integrations.letta`
Griptape	`griptape`	`unplug.integrations.griptape`
AG2	`ag2`	`unplug.integrations.ag2`
Atomic Agents	`atomic-agents`	`unplug.integrations.atomic_agents`
Agno	`agno`	`unplug.integrations.agno`
CrewAI	`crewai`	`unplug.integrations.crewai`
AutoGen	`autogen`	`unplug.integrations.autogen`
Haystack	`haystack`	`unplug.integrations.haystack`
LlamaIndex	`llama-index`	`unplug.integrations.llama_index`
Pydantic AI	`pydantic-ai`	`unplug.integrations.pydantic_ai`
Semantic Kernel	`semantic-kernel`	`unplug.integrations.semantic_kernel`
Custom loop	(none)	`unplug.integrations.hooks`

LangGraph¶

from unplug.integrations.langgraph import langgraph_input_node, langgraph_tool_guard

graph.add_node("unplug_gate", langgraph_input_node(hooks))
tool_guard = langgraph_tool_guard(hooks)

Demo: python examples/langgraph_hooks_demo.py

Haystack (RAG)¶

from unplug.integrations.haystack import UnplugDocumentGuard, scan_for_ingestion

Install: pip install unplug-ai[haystack]. See RAG_DEFENSE.md.

Hosted vs local¶

Setup	Guard init
Hosted API	`Guard(mode="server")` + `UNPLUG_API_KEY`
Sidecar	`Guard(mode="server", server_url="http://127.0.0.1:8000")`
Embedded	`Guard()`

Tool enforcement always runs locally. See DEPLOYMENT.md.

Security testing¶

uv run pytest tests/security/test_agent_integration_matrix.py -v

See integrations/TESTING.md for all 72 angles.

Docker E2E (sidecar + examples)¶

cd sdk && make docker-e2e

MCP¶

For Claude / Cursor MCP server, use unplug-mcp — see integrations/mcp/README.md.