langchain - 💡(How to fix) Fix Feature: Add AgentLock authorization middleware integration [2 comments, 3 participants]

Checked other resources

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

LangChain tools execute with no built-in permission validation. AgentLock is an open authorization standard (Apache 2.0) that adds declarative permission blocks to any tool definition: role-based access control, scope constraints, rate limiting, PII redaction, audit logging, and human approval gates.

The integration package already exists on PyPI: pip install agentlock[langchain]

Integration is a one-line wrapper:

from agentlock.integrations.langchain import AgentLockToolWrapper protected_tool = AgentLockToolWrapper(original_tool, permissions={...})

Requesting official listing in LangChain's integrations documentation so developers can discover it.

GitHub: https://github.com/webpro255/agentlock Docs: https://agentlock.dev PyPI: https://pypi.org/project/agentlock/

Use Case

We benchmarked a standard LangChain agent with 5 tools against 181 adversarial attacks:

Without AgentLock: 73 injection failures, PII leaking (SSNs, medical records), 13 YARA threat signatures With AgentLock: 12 failures, zero PII leaks, 2 threat signatures

The 12 remaining failures are model-layer information leakage that no middleware can fix.

Full benchmark: https://github.com/webpro255/agentlock/blob/main/docs/benchmark.md

Any LangChain developer deploying agents with tools that access databases, file systems, email, or APIs faces this problem. There is currently no standard way to enforce permissions on tool calls within LangChain.

Proposed Solution

List AgentLock as an official third-party integration in LangChain's integrations documentation. Happy to submit a docs PR following whatever format the team prefers.

The package is already published, tested (469 tests), and supports LangChain 1.2.x with direct tool calling via bind_tools().

Alternatives Considered

Manually wrapping every tool with custom permission checks. This works but is not standardized, not auditable, and every developer reinvents it differently. AgentLock provides a single standard that any LangChain project can adopt.

Additional Context

AgentLock aligns with OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, and EU AI Act requirements. Two NIST submissions have been made referencing the standard. The project has had engagement from security architects at Microsoft, Vanta, and Palo Alto Networks.

Apache 2.0 licensed. 469 tests. Framework-agnostic with integrations for LangChain, CrewAI, AutoGen, MCP, FastAPI, and Flask.