Capture layer · framework adapters · inspectable

One SDK across
every agent framework.

Issued to customers on request and inspectable end-to-end, with framework adapters for the runtimes your team actually runs. One line instruments OpenAI Agents SDK, Claude Agent SDK, MCP, LangChain or CrewAI — and captures not just what the agent did, but the authority it was acting under at that instant.

Book a 30-minute walkthrough Read the docs
OpenAI Agents SDK Claude Agent SDK MCP servers LangChain · CrewAI Python · TypeScript (beta)

In this section

🐍

Python SDK

Hash-chain integrity, inline PII redaction, non-blocking transport, offline buffer. Source review under NDA at the paid tier.

⚡

TypeScript SDK

Identical receipt format, identical canonical-JSON hash (parity-tested against the Python SDK). Adapters for LangChain, OpenAI Agents, Vercel AI SDK, MCP, Claude Agent SDK and Mastra. Beta.

🎫

Authority capture

Every receipt pins the authority, scope, and delegation in force at the moment of the action — recorded then, not reconstructed later. The record shows what the agent was permitted to do, not just what it did.

🤖

OpenAI Agents SDK adapter

One-line wrap on the OpenAI Agents SDK runtime. Every tool call, model call, and sub-agent spawn captured as a receipt — each pinned to the authority in force at that instant.

🧠

Claude Agent SDK adapter

Same primitive for Anthropic's Claude Agent SDK. Receipts capture model selection, tool routing, and sub-agent hierarchy — each pinned to the authority in force at that instant.

🔌

MCP server adapter

Protocol-level capture. Wraps any MCP server so every tool call routed through it is automatically receipted, regardless of the agent host — each pinned to the authority in force at that instant.

🦜

LangChain & CrewAI

Callback-based adapters for LangChain and CrewAI. Captures chain steps, agent decisions, and tool invocations at the framework boundary — each pinned to the authority in force at that instant.

Five-minute quickstart.

# 1. Install the SDK

$ pip install agentaudits



# 2. Initialise in your agent code

import agentaudit

agentaudit.init(api_key="aa_live_...", agent_id="my_agent")



# 3. Wrap your framework — choose one:

agentaudit.instrument_openai_agents()

agentaudit.instrument_claude_agent()

agentaudit.instrument_mcp_server()



# 4. Your agent runs as normal — receipts flow automatically

agent.run(customer_query)



# 5. Verify chain integrity any time

$ agentaudit verify --agent my_agent --from 2026-06-01 --to 2026-06-07

Want help integrating?

We pair-program the integration with your team. Half a day, then your receipts are flowing in production.

Book a 30-minute walkthrough