Security & AI safety

AI guardrails & agent security

3,625 agents · most starred first

AI guardrails & agent security, page 36 of 37

bedrock-guardrail-firewall

Privacy-first policy enforcement for generative AI traffic

byteplussdk-llmshield

Byteplus llmshield SDK for Python

call4agents

One yes, one payment, even after a crash: once-only actions and approvals bound to exact arguments for Python AI agents

capsule-litellm

Capsule plugin for the LiteLLM gateway — forwards completion requests to the Capsule agentsecurity service for centralized policy enforcement and audit.

claude-agent-acgs

ACGS rule-based tool-call governance for the Claude Agent SDK, via a PreToolUse hook

collieai-litellm-guardrail

CollieAi guardrail for the LiteLLM proxy: checks every chat completion against your CollieAi policy, observing it or blocking what the policy blocks.

conduct-agent-guard

Runtime governance for AI agents. Conduct Guard on every tool call in Claude Agent SDK, OpenAI Agents SDK, LangChain, Google ADK and CrewAI agents.

crewai-acgs

ACGS rule-based tool-call governance for CrewAI, as native tool call hooks

crewai-aer1

AER-1 verifiable execution receipts for CrewAI crews: one checkable receipt per tool call, via CrewAI's native event bus.

crewai-alex-evidence

Verify signed ALEX evidence bundles as a CrewAI tool. Thin adapter, no second verifier.

crewai-nti

Full 5-pillar NTI post-quantum security for CrewAI agents.

datalayer-agent-shield

Datalayer Agent Shield: a runtime firewall for every agent

datumline-agent-guard

Independent verifier for AI agent completion claims (VERIFIED, FAILED, FALSE_DONE, UNVERIFIABLE): CLI, Python library, MCP server, GitHub Action and LangChain tool.

haikeilabs-agentware

Policy enforcement and audit middleware for LLM agent tool calls

langchain-acgs

ACGS tool-call governance middleware for LangChain agents

langchain-alex

Verify signed ALEX evidence bundles in LangChain and LangGraph workflows.

langchain-nti

Full 5-pillar NTI post-quantum security for LangChain agents.

langchain-right-rudder

LangChain agent middleware that reads an agent's committed state, and repairs the step where a later action would contradict an earlier one.

langchain-sync-monitors

AI-control monitors for LangChain agents and Deep Agents: every step an agent proposes is judged before any of its own tools run, and a control protocol decides what runs.

lightguard-agent

A Lightweight Defense Architecture Against Indirect Prompt Injection in Tool-Calling AI Agents and MCP Workflows

llm-guardrails-middleware

Drop-in guardrails wrapper for LLM API calls: PII redaction, prompt-injection detection, output schema validation, and usage/latency logging.

llm-redteam-evidence

Turn LLM red-team results (garak, promptfoo) into audit evidence, OSCAL and a model release gate

llm-stream-guards

Deterministic output guards for a streamed LLM reply: untrusted links, made-up emails, phone numbers and prices, quoted system prompts and code, stopped before they reach the screen.

llmfirewall-core

A lightweight, provider-agnostic security and policy enforcement layer for LLM, RAG, and AI-agent applications.

llmrivotril

A lightweight framework to reduce LLM hallucinations, enforce guardrails, manage stateful memory, and monitor performance via local dashboard.

llmsentry-ai

Drop-in prompt-injection firewall for LLM apps

loopgrid-claude-agent

Signed decision, policy, approval and tool-action evidence for Claude Agent SDK

maigp-bedrock-agentcore

AIGP governance adapter for AWS Bedrock AgentCore Runtime

maigp-pydantic-ai

AIGP governance adapter for PydanticAI

nemo-agent-toolkit-tenuo

Task-scoped Tenuo authorization for NVIDIA NeMo Agent Toolkit

nishanttyagi-agenttrust

Thin evidence chain across policy, sealed effect, witness, and CI regression.

openai-agents-acgs

ACGS rule-based tool-call governance for OpenAI Agents SDK function tools, via tool guardrails and approvals

orderstream-agent-guard-proxy

Enterprise reverse proxy and firewall middleware for Model Context Protocol (MCP) JSON-RPC tool calls

orderstream-agent-sandbox

Deterministic Zero-Trust Security Sandbox & Cryptographic Audit DAG for Autonomous AI Agents in CI/CD

provared-langchain

A LangChain middleware that checks each tool call of an agent against the permission a person signed, and records it with Provared. Evidence, not a verdict.

provellm-verify

Verify ProveLLM agent credentials and signed receipts — offline, with the issuer's public key only.

pydantic-ai-acgs

ACGS rule-based tool-call governance for Pydantic AI agents, as a capability

pydantic-ai-loopgrid

Native Pydantic AI capability for signed, verifiable consequential agent action evidence

pydantic-ai-provared

A Pydantic AI capability that checks each tool call of an agent against the permission a person signed, and records it with Provared. Evidence, not a verdict.

pyhroff-ragsentry

Static scanner for injected instructions and retrieval-manipulation in RAG source documents.

rag-security-protocol

RAG Security Protocol — where security tools hook into a RAG pipeline, and what they return

ragleap-security

Small, auditable security primitives for the ragleap libraries: immutable tenant scopes, a tamper-evident hash-chained audit log, secret handling and redaction, hard input limits, rate limits, keys with rotation, file integrity manifests and optional authenticated encryption. Standard library only, except optional AES-256-GCM encryption (extra: crypto). Not independently audited.

readystack-agentforce-action-audit

13 rules on a Salesforce Agentforce topic file, before the deploy and before Article 50 (bundled Node.js runtime)

reprobe-agents

Turn an agent security failure into a reproducible regression test.

sphragis

Cryptographic clone-detection engine for NXP NTAG 424 DNA (SUN/SDM)

sweetai-agent-firewall

Pre-transaction trust firewall for AI agents: should-i-pay decisions, sybil forensics, liveness SLA (SweetAI x402 / ERC-8004)

vamp-rag-audit

RAG and agentic AI security auditor: prompt injection, context poisoning, data exfiltration

weard-agenticguard

AgenticGuard SDK for Python: tamper-evident records of what AI agents do. First release coming soon.

whiteout-litellm

Whiteout guardrail for the LiteLLM Proxy — govern every app behind the gateway

whiteroom-bedrock

Attaches WhiteRoom session attestation to a boto3 bedrock-runtime client

ghostcorpnet

ghostcorpnet review: compare features, pricing, use cases, access model, and alternatives for this AI Security agent in 2026. Governed Agent Mesh Playbook: agent identity, spend limits, approval gates, and audit trails.

agentguard-eval

Guardrails and evals for AI agents: a policy gate for tool calls, a checker for invented facts, and a benchmark for shell-command gates. Deterministic, no API keys.

governed-agent

Cryptographic capability sealing, scope-bounded planning, and independent audit for LLM agents

langgraph-secure-agents

Deterministic authorization for Python callables, LangGraph nodes, agent tools, and MCP tools.

vegaduta-agent

Governance for AI agents in Python: tool risk classification, human approval, budgets, idempotent effects, guardrails, hooks, audit and telemetry.

@hynaweb/kavlr

The AI Agent Action Gate.

sello

Reference implementation of the Sello protocol for service-signed AI agent receipts.

@muxoai/cli

Muxo — Governed AI Agent Execution, Zero-Trust Credential Vault, Policy Enforcement & Cryptographic Receipts across 870+ Tools and 28 Cloud Providers

@tenova/swt3-mcp

MCP server for AI compliance audit trail and governance. 75 tools, 93 frameworks, 47 provider adapters. EU AI Act, NIST AI RMF, GDPR, CMMC, ISO 42001, OWASP Agentic Top 10, OWASP MCP Top 10, SR 11-7, NIS-2, EU Battery Reg, FERC. Adapters for OpenAI, Anthr

principles-disciple

Principles Disciple is an AI Agent Governance System. Stop correcting the same AI behavior across sessions. Turn repeated Agent corrections into Owner-approved, observable, reversible behavior principles.

agentlatch-security

Local-first AI agent security scanner: checks agent code, tool manifests, and prompts for prompt injection, over-privileged tools, and secrets.

@bydtech/agent-guard

A small policy layer between an LLM agent and its tools: allow, deny or ask a human before each tool call runs, with argument checks, rate limits and an audit log.

@safeprompt.dev/langchain

LangChain integration for SafePrompt: prompt-injection detection as a callback handler. Validates every prompt flowing through a chain before it reaches the LLM.

@atlasauth/pca-openai

OpenAI adapter for Proof-Carrying Authority: turn a model tool_call into a PCActn and run the handler with the proof attached.

configure

Identity layer SDK for AI agents

agent-ledger-guard

Tamper-evident flight recorder and policy gate for AI coding agents: Claude Code, Codex, OpenCode, Cursor, Gemini, Cline, Goose and more.

agentcap-cli

Local-first quota and cost guard for coding agents: per-job share of Claude Code/Codex/Gemini subscription quota, loop breaker, evidence receipt.

agentdetox

Check risky writes to coding agents' memory, instructions, skills and config. Review changes and source evidence, then restore earlier versions.

codesec-agentguard

Official Python SDK for the AgentGuard AI Agent Security Gateway

fullcourtdefense-agent

Control what your AI agents do: check every tool call against your company's FullCourtDefense policy (Monitor or Enforce).

invaris-agentauth

Self-certifying, rotatable identities for autonomous AI agents

llm-app-security-scanner

Static security scanner for LLM/agent repos: shell, file-write, secret, and untrusted-input risks as a CI gate, with SARIF output.

nemo-agent-toolkit-hol-guard

HOL Guard runtime security middleware for NVIDIA NeMo Agent Toolkit

tba-storage-access-governance

Owned access-governance identity database tables and migrations.

@authentify2026/langchain

LangChain tool wrapper for the Authentify authorization API

firstwatch

First line of defence for AI agents: spots attacks hidden in pages, files and tool results, records each one as evidence in an OpenTelemetry-ready format, and asks before risky actions. Claude Code hooks plus a skill for any agent. No dependencies, no net

@prismor/mastra

Prismor for Mastra — policy-controlled tool calls via a wrapped execute function and the Prismor eval-server

@oked/langgraph

OKed for LangGraph and LangChain. Wraps a tool so a risky call waits for a human approval before it runs.

@oked/sdk

OKed SDK - human approval layer for AI agents

agent-firewall-cascade

A risk-aware firewall for LLM agent tool calls (prompt-injection defence)

agentguard-scan

Detecta tool poisoning y rug pulls en skills (SKILL.md) y conectores MCP de agentes de IA

agentonboard-sdk

Universal identity layer for AI agents — Python Partner SDK for token verification and account creation

agentprov

AgentPROV: an execution-provenance framework for LLM-agent security research — source-preserving, evidence-qualified execution records and defensible W3C PROV, with benchmark/framework adapters

agentprov-capture

AgentPROV capture/0.1 raw-evidence event log: model, writer, hash-chain verifier, redaction. Stdlib + rfc8785 only; shared by the AgentPROV core (reader) and the isolated live backends (writer).

agentscope-acgs

ACGS rule-based tool-call governance for AgentScope agents, via agent middleware

agentsec-agentpolicy

A pre-execution, deterministic policy decision point for tool-using AI agents. Evaluate every proposed tool call against declarative policy and return allow / audit / escalate / deny.

autogen-surface

Surface for AutoGen: the link-safety layer between an AI agent and the open web. One tool, surface_preflight, that tells your agent whether a link is worth fetching.

ci-agent-audit

Find prompt injection and unsafe AI agent setups in GitHub Actions and GitLab CI.

coding-agent-audit

Scan your machine for known AI-coding-assistant supply-chain backdoor indicators (Claude Code / Cursor / Windsurf extensions, MCP servers) plus risky hook/permission config patterns. Usable as a CLI or as an MCP server.

complyee-agent-guard

Run agent tools only on a verified Agent Control Room grant (LangGraph, CrewAI, LlamaIndex, LangChain and plain Python).

crewai-surface

Surface for crewai: the link-safety layer between an AI agent and the open web. One tool, surface_preflight, that tells your agent whether a link is worth fetching.

dspy-surface

Surface for DSPy: the link-safety layer between an AI agent and the open web. One tool, surface_preflight, that tells your agent whether a link is worth fetching.

eternity-agentshield

ETERNITY : AgentShield 2.0 — Enterprise Security Middleware & Guardrails SDK for Autonomous AI Agents

langchain-security-extension

Enterprise RBAC and Attribute-Based Access Control (ABAC) security framework for LangChain AI applications.

langchain-surface

Surface checks a short link before your agent opens it. LangChain tool surface_preflight. Fetch or skip; it never returns the page. Free for the first million requests a month.

llama-index-tools-surface

Surface checks a short link before your agent opens it. LlamaIndex tool surface_preflight. Fetch or skip; it never returns the page. Free for the first million requests a month.

llm-layered-defense

Layered defense framework (verifier, metadata-filtered RAG, negative prompting, alignment) for mitigating LLM hallucinations, with a reproducible ablation harness.

openai-agents-surface

Surface for OpenAI Agents SDK: the link-safety layer between an AI agent and the open web. One tool, surface_preflight, that tells your agent whether a link is worth fetching.

pydantic-ai-surface

Surface for Pydantic AI: the link-safety layer between an AI agent and the open web. One tool, surface_preflight, that tells your agent whether a link is worth fetching.

rogue-llm

Adversarial LLM proxy for testing whether agent security is structural or behavioral

More in Security & AI safety