shieldpi 0.2.0

Runtime EDR for AI agents — zero-code-change monitoring for LangChain, LangGraph, OpenAI Assistants, and the Anthropic SDK

ShieldPi — Runtime EDR for AI Agents

Zero-code-change monitoring for AI agents. Every tool call, LLM call, file read, and outbound request your agent makes streams to ShieldPi's detectors in real time — the same way EDR agents stream endpoint events to a SIEM.

The 60-second setup

pip install "shieldpi[all]"
export SHIELDPI_SDK_KEY=shpi_live_...   # get one at https://shieldpi.io/dashboard

# Add one import at the top of your agent process:
import shieldpi.auto

# Use LangChain / LangGraph / OpenAI Assistants / Anthropic SDK normally.
from langchain.agents import AgentExecutor
agent = AgentExecutor(...)
agent.invoke({"input": "..."})   # every tool call + LLM call is captured

That's it. Open https://shieldpi.io/dashboard/agent-monitor to see the live event stream and any alerts the detectors fire.

What gets captured

Agent action	What ShieldPi sees	Detector that scores it
User sends a message	user_message event	pattern_match scans for jailbreak strings
Agent calls a tool	tool_call event with args	analyzer categorizes (destructive / exfil / credential_access)
Agent reads a file	tool_call args with path	pattern_match catches path traversal
Agent makes HTTP request	tool_call with URL	trajectory catches read→exfil kill chains
Agent writes to memory	memory_write event	memory_integrity detects tampering
Agent responds to user	final_response event	response_leak_scanner checks for leaked secrets

Six detectors run in parallel on every event: pattern_match, trajectory, analyzer, memory_correlation, memory_integrity, response_leak_scanner.

Supported frameworks

Framework	Auto-patch	Manual handler	Optional dep
LangChain (AgentExecutor)	✅	✅	pip install shieldpi[langchain]
LangChain (LCEL tools)	✅ (BaseTool hook)	✅	pip install shieldpi[langchain]
LangGraph	✅	✅	pip install shieldpi[langgraph]
OpenAI Assistants API	✅	—	pip install shieldpi[openai]
OpenAI Chat Completions w/ tools	✅	—	pip install shieldpi[openai]
Anthropic SDK (tool use)	✅	✅	pip install shieldpi[anthropic]
Custom agents (no framework)	—	✅	base install

Explicit configuration (optional)

If you want to override env-var defaults:

import shieldpi.auto

shieldpi.auto.configure(
    sdk_key="shpi_live_...",           # or SHIELDPI_SDK_KEY env
    agent_name="invoice-bot",           # or SHIELDPI_AGENT_NAME env
    stated_goal="help users file invoices",
    frameworks=["langchain", "anthropic"],  # or SHIELDPI_AUTO_INSTRUMENT=langchain,anthropic
)

Environment variables:

Variable	Purpose
SHIELDPI_SDK_KEY	Your SDK key (starts with shpi_live_). Required.
SHIELDPI_BASE_URL	Override the API base URL (defaults to production).
SHIELDPI_AGENT_NAME	Logical name for this agent. Default: unnamed-agent.
SHIELDPI_AGENT_GOAL	Stated goal (helps detectors flag off-goal behavior).
SHIELDPI_AUTO_INSTRUMENT	Comma-separated framework allowlist. Default: all installed.

Manual integration

For custom agents or full control:

from shieldpi import Monitor

monitor = Monitor(sdk_key="shpi_live_...")
with monitor.start_session(
    agent_name="invoice-bot",
    stated_goal="help users file invoices",
) as session:
    session.log_user_message("How do I file a Q1 invoice?")
    session.log_tool_call("search_docs", {"query": "Q1 invoice filing"})
    session.log_tool_result("search_docs", {"results": [...]})
    session.log_final_response("Here's how to file a Q1 invoice...")

LangChain (manual)

from shieldpi import Monitor
from shieldpi.hooks.langchain import ShieldPiCallbackHandler

monitor = Monitor(sdk_key="shpi_live_...")
handler = ShieldPiCallbackHandler(monitor, agent_name="my-agent")
agent.invoke({"input": "..."}, config={"callbacks": [handler]})

Anthropic SDK (manual)

from anthropic import Anthropic
from shieldpi import Monitor
from shieldpi.hooks.anthropic import monitored_tool_use

anth = Anthropic()
monitor = Monitor(sdk_key="shpi_live_...")

with monitored_tool_use(monitor, agent_name="invoice-bot") as session:
    session.log_user_message("Help me file an invoice")
    response = anth.messages.create(
        model="claude-opus-4-20250514",
        messages=[{"role": "user", "content": "Help me file an invoice"}],
        tools=[...],
    )
    session.observe_anthropic_response(response)

Safety guarantees

The SDK is built to never crash your agent:

• Every HTTP call is fire-and-forget (background worker, bounded queue).
• Monitoring failures are logged at WARNING level; they never raise to your agent.
• If SHIELDPI_SDK_KEY is missing, import shieldpi.auto logs once and does nothing.
• All patches are idempotent — double-import won't double-wrap.
• Patches gracefully no-op when the target framework isn't installed.

What's new in 0.2.0

• shieldpi.auto — one-line zero-code-change monitoring for LangChain, LangGraph, OpenAI, and Anthropic.
• LangGraph support — first-class patching of Pregel.invoke/ainvoke/stream/astream.
• OpenAI Assistants support — new patch for client.beta.threads.runs.create_and_poll and Chat Completions with tools.
• Import hook — frameworks imported AFTER shieldpi.auto are still patched.
• Idempotent patches — safe to re-import or reconfigure at runtime.

Links

• Dashboard — https://shieldpi.io/dashboard/agent-monitor
• Docs — https://shieldpi.io/docs/live-agent-monitor
• Scanner package — shieldpi-mcp (MCP server for Claude Desktop / Cursor / Continue)

License

Apache-2.0