aport-agent-guardrails-langchain

APort is an AI agent passport and control plane for agentic work. It is described as authorization infrastructure for the AI agent economy, and it focuses on issuing passports, authorizing tool calls before execution, and recording guardrail decisions for audit.

Its core pieces are agent passports, pre-action authorization, guardrails, an audit terminal, and proofs. The page says passports use verifiable identity via W3C DID/VC and are provided as a JSON file with no signup. Policy checks can enforce command allowlists, spending caps, file size limits, and sector concentration rules. It also describes deliverable contracts for deterministic quality gates such as test coverage thresholds, review requirements, and acceptance criteria, along with cryptographic records that are Ed25519 signed and tamper-evident. The same system is shown blocking destructive commands before execution and verifying task completion before an agent can mark work done.

APort is presented for developers and enterprises. The examples and use cases include AI engineering teams working with multi-agent workflows, enforced quality, agent handoffs, and blocking secret exposure or path traversal, as well as regulated industries using pre-trade risk checks, data export controls, and signed attestations. The page also mentions support for audit requirements including SOC 2, HIPAA, SOX, and IIROC, and references compliance contexts such as PIPEDA and OSFI in its use-case text.

It works with Claude Code, Cursor, MCP tools, LangChain, CrewAI, OpenAI, FastAPI, Express, Next.js, Python, and Node.js, and the page also lists Claude Code, LangGraph, Vercel, OpenClaw, Anthropic, Salesforce, AWS, Django, and Gemini among the frameworks, tools, and ecosystems it works with. One installation example shows a PreToolUse hook for Claude Code, and another shows an install script.