Open Source · MIT License · Zero Telemetry

Runtime Security for AI Agents

En Garde. Your AI agents
are exposed.

OnGarde is an open source runtime security proxy that sits between your AI agents and any LLM API. One config line. No code changes. Works with OpenClaw, CrewAI, LangChain, Agent Zero — or anything that calls an LLM over HTTP. Credential leaks, PII, prompt injection, dangerous commands: blocked in under 50ms. Fail-safe by default. Self-host the open source edition, or upgrade to Pro for managed infrastructure, expanded detection, and compliance certifications.

Star on GitHub See It Block a Threat
<50ms overhead
5 threat categories
1 line setup
100% open source

Watch It Work.

Select an attack scenario. Watch OnGarde intercept and block it in real time.

ongarde — interceptor — live demo
Click an attack scenario above to begin. ▋

What Gets Blocked

Five threat categories.
Every request. Both directions.

Block patterns grow over time as the community contributes — this is a living service, not a static ruleset. Every update protects all installations automatically.

What OnGarde Blocks — five threat categories: Credential Leaks, Dangerous Commands, PII & Personal Data, Sensitive Files, and Prompt Injection — all blocked in under 50ms
Living Service

Block categories expand continuously. Community-contributed patterns are reviewed and shipped to all installations — no updates required on your end.

Zero Code Changes

One config line.
Every request protected.

Set OnGarde as the baseUrl for your LLM provider. Works with any platform that calls an LLM over HTTP — no SDK changes, no middleware, no boilerplate.

.ongarde/config.yaml 
# That's it. One change. Any platform.
upstream:
  url: "https://api.openai.com"  # or Anthropic, Groq, any LLM
proxy:
  host: "127.0.0.1"
  port: 4242
scanner:
  mode: "full"  # fail-safe: blocks on error, never fails open
Zero code changes
Fail-safe: blocks on error
<50ms overhead
Any platform, any LLM
OnGarde architecture — AI Agent sends request through OnGarde proxy which scans and either blocks threats or forwards clean requests to the LLM provider

Integrations

Works where
you already run.

No custom SDKs. No vendor lock-in. No code changes.

Primary
OpenClaw
Native baseUrl proxy integration
Agent Zero
HTTP proxy layer
CrewAI
HTTP proxy or Python middleware
LangGraph
LangChain middleware or HTTP layer
LangChain
700+ integrations, HTTP proxy
OpenAI-Compatible
Drop-in proxy, zero code changes

If it calls an LLM over HTTP, OnGarde can protect it.

Open Source

Open source core.
Enterprise-ready when you need it.

The self-hosted open source edition gives your agents real protection out of the box — no subscription, no sign-up. When your team needs compliance certifications, managed infrastructure, and expanded detection coverage, Pro and Enterprise have you covered.

MIT Licensed

Use it, modify it, ship it. No restrictions. Read the source, verify the behavior, deploy with confidence.

Living Block Library

Detection patterns grow as the community reports and contributes new threat data. Your installation gets stronger every release — automatically.

Zero Telemetry

Nothing leaves your machine. No usage analytics, no call-home, no surprise data collection. Your agent traffic is yours, full stop.

Star on GitHub

MIT licensed · Self-hosted · No sign-up required

Quick Start

Up in three steps.

One command. One config change. Every LLM request protected.

Get Started in 3 Steps — Install with npx, get your API key from the dashboard, point your agent at localhost:4242
Get Started on GitHub QUICKSTART.md →

Pricing

Yours to run.
Yours to inspect.

Start open source. Upgrade when your business needs it. No telemetry at any tier.

Prime
Free MIT Licensed · Self-hosted

Everything you need to secure a self-hosted AI agent stack.

  • Regex scanner (lite mode)
  • Credential detection (12+ providers)
  • Basic PII patterns
  • SQLite audit trail
  • Streaming SSE protection
  • Community support
  • MIT licensed
Deploy in 5 minutes
Most Complete
Forte
Pro Coming Soon

Full NLP scanning with cloud audit and custom rules for production workloads.

  • Everything in Prime
  • Full Presidio NLP scanning
  • Supabase cloud audit trail
  • Custom detection rules
  • Webhook event delivery
  • Dashboard (coming soon)
  • Priority support
Octave
Team Coming Soon

Multi-instance fleet management for teams running AI at scale.

  • Everything in Forte
  • Multi-instance management
  • RBAC + SSO
  • SLA guarantee
  • Dedicated support channel
  • Custom onboarding
  • Enterprise audit export

All tiers are self-hosted. Your data never leaves your machine. OnGarde is open source under the MIT license.