Adrian

Summary

Adrian is an open-source runtime security layer for AI agents, free to use under an Apache-2.0 licence, that watches an agent's actions and reasoning as it runs and can alert, block, or route decisions to a human reviewer before they execute. It ships as an SDK you wrap around existing code (a few lines around a LangChain or Anthropic-based agent, or a Claude Code plugin via hooks) with events streamed to a hosted or self-hosted dashboard, making it a fit for teams building or securing agentic AI systems rather than traditional application security teams. The README frames it as a runtime counterpart to static analysis and network monitoring, catching prompt-injection and out-of-remit behaviour those miss, and includes PII redaction for agent transcripts. Development is active, with recent releases adding reasoning capture across SDKs and deeper Claude Code integration.

Open-source runtime AI agent security tool - monitors and controls AI agents, catching malicious tool use, prompt injection, and policy drift in real time, before the agent acts.

Command line

Values are placeholders taken from each option’s declared default. Nothing is executed here — the output shown is a recording of a run that already happened.

Release history

1. python-sdk-v1.2.0 Aug 18, 2026 · issue 005

   Adrian Python SDK v1.2.0 adds reasoning capture for LangChain and OpenAI-compatible models, including a new reasoning field in LLM events.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch python-sdk-v1.2.0 https://github.com/secureagentics/Adrian.git
   # already have the repo? check out this version:
   $ git checkout python-sdk-v1.2.0

   • ›Adds a reasoning field to LLM end events in the Python SDK, populated via _extract_reasoning() from LangChain ChatGeneration messages, capturing chain-of-thought output from reasoning models.
   • ›Extends reasoning capture to support three provider layouts: a summary list of {type: 'summary_text', text: ...} entries (OpenAI Responses / LangChain default), a bare reasoning string (LangChain output_version='v1'), and Anthropic's thinking string.
   • ›Adds reasoning capture to the TypeScript OpenAI SDK integration, reading delta.reasoning_content from streamed chat completion chunks for OpenAI-compatible servers that surface chain-of-thought in that field.
2. python-sdk-v1.1.0 Aug 11, 2026 · issue 005

   Adrian Python SDK v1.1.0 adds a Claude Code plugin with HITL verdicts, PII filtering, and Anthropic SDK integration.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch python-sdk-v1.1.0 https://github.com/secureagentics/Adrian.git
   # already have the repo? check out this version:
   $ git checkout python-sdk-v1.1.0

   
   • ›New Claude Code plugin (integrations/claude-code/adrian_cc/agent.py) connects Claude Code hooks to the Adrian backend via WebSocket, supporting three server-driven execution modes: MODE_ALERT (fire-and-forget), MODE_BLOCK (wait for verdict, block on policy), and MODE_HITL (wait for human approval on the dashboard).
   • ›Adds PII filtering to the Claude Code plugin (integrations/claude-code/adrian_cc/pii/_patterns.py), detecting 12 PII categories — EMAIL, PHONE, SSN, CREDIT_CARD, IP_ADDRESS, DATE_OF_BIRTH, IBAN, PASSPORT, STREET_ADDRESS, POSTAL_CODE, DRIVER_LICENSE, and AWS_KEY — in agent conversations before they are forwarded.
   • ›HITL verdicts are now forwarded to Claude Code for native inline approval, enabling human-in-the-loop decisions to flow directly back to a running Claude Code session.
   • ›Adds Anthropic SDK integration, allowing Adrian to instrument Anthropic API calls alongside the existing LangChain/LangGraph support.
   • ›LangChain/LangGraph instrumentation is extracted into langchain_handler.py, isolating the handler for cleaner integration management.
   +1 moreshow less

   • ›Severity badges now appear in the Events table and mobile cards, based on MAD codes, and the event list API response adds latest verdict metadata while omitting full payload and token fields from list rows.