Loading
AgentFox
Benchmark
Sign inDocs
BenchmarkSign in

Get started

  • What is AgentFox
  • Quickstart
  • Quickstart without an account
  • Core concepts

Connect your agent

  • Pick your stack
  • Python apps
  • LangGraph
  • LiteLLM proxy
  • MCP servers
  • Claude Code and Codex
  • Voice and hosted platforms
  • Any language (HTTP)

Discover

  • Map your agents

Govern

  • Tool access
  • Approvals and the kill switch
  • Who may see what

Protect

  • Guardrails
  • Custom rules
  • Fix false positives

Test

  • Attack tests
  • In CI and in production

Prove

  • Traces and issues
  • Audit evidence
  • Send traces elsewhere

Deploy

  • Self-host
  • Production checklist

Reference

  • CLI
  • HTTP API
  • Python SDK
  • Policy language
  • Detectors and findings
  • Configuration
  • Install options
  • OWASP Agentic Top 10
  • Coming from LLM Guard
  • Benchmarks
  • Limits

Help

  • Troubleshooting

Connect your agent

Pick your stack

Find the page for how your agent is built, and follow it.

A plain Python app needs nothing more: agentfox init in the Quickstart already adds the one line it needs.

  • Python appsOne line, agentfox.auto(), for OpenAI, Anthropic, LiteLLM, LangChain and CrewAI.
  • LangGraphGuard retrieval, model and tool nodes, and pause the graph for approval.
  • LiteLLM proxyAdd AgentFox to the proxy config. Every agent behind it is checked.
  • MCP serversScan the servers your agents load, pin their tools, and check each call.
  • Claude Code and CodexHooks that check what a coding agent runs and edits.
  • Voice and hosted platformsVapi, Retell, Copilot Studio and AgentCore.
  • Any language (HTTP)Point your model calls at the AgentFox proxy, or ask the guard API.
PreviousCore conceptsNextPython apps
AgentFox

Runtime guardrails for AI agents. Open source under Apache-2.0.

support@nometria.com
ProductPlaygroundLivePricing
BenchmarksBenchmarkLatency benchmark
ResourcesDocsBlogSupportGitHub
CompanySecurityPrivacyTerms
© 2026 Nometria, Inc.