These reference integrations show the smallest useful architecture for independently verifiable AI decisions: assessment, authorization, execution, receipt issuance, and third-party verification. They are designed for technical due diligence, research environments, and regulated pilots.
Evidence pipeline
The goal is not to impress with UI. The goal is to prove that a partner can generate a Decision Receipt inside its own workflow, then verify it later without trusting the original system.
Capture the policy route, risk level, model choice and execution conditions before a consequential action happens.
Record whether a human or a system policy authorized execution, and seal that information into the same evidence trail.
Hash content locally, issue an Ed25519-signed Decision Receipt, and append it to the existing Signatrust hash chain.
Use only the receipt and public verification endpoints to prove integrity, signature validity and ledger linkage.
Available examples
Minimal Node.js reference flow that registers an agent, seals a healthcare decision, and verifies the receipt immediately.
Standard-library example for teams that want zero third-party dependencies in the base proof-of-concept.
Raw HTTP flow for buyers that want protocol-level evaluation before touching an SDK.
A lightweight authorization service showing Signatrust as the evidence layer inside an application endpoint.
Graph-based orchestration example with step-level sealing under one shared trace id.
Workflow-native example for operational teams that want low-friction sealing inside automation flows.
Verified PyPI adapters
Give any LangChain, OpenAI Agents SDK, CrewAI or AutoGen agent the ability to seal its own decisions into independently verifiable receipts. Each adapter is a first-party package under the same GitHub organisation as the Signatrust core, published under a sole-owner account on PyPI.
pip install langchain-signatrust
Three LangChain tools: signatrust_generate_receipt, signatrust_verify_receipt, signatrust_get_receipt. Compatible with LangGraph create_react_agent.
pip install openai-agents-signatrust
Tools for the OpenAI Agents SDK. Seal each tool call individually or seal the final decision — configurable per agent.
pip install crewai-signatrust
CrewAI-native tools. Attach a receipt to every consequential crew task and preserve the chain of custody across agents.
pip install autogen-signatrust
Microsoft AutoGen (autogen-core ≥ 0.4) tools. Async-first API, receipt hooks around agent handoffs.
pip install signatrust (v1.3.1 on PyPI). All four framework adapters call the same public Signatrust API.
For research and assurance teams
Start with the reference integrations repository, then adapt the same evidence pipeline to your domain: healthcare, public-sector decisions, finance, or operational agents.