OWASP LLM Top 10 — #1 Covered

Stop prompt injection
in your AI app.
One line change.

CollieAI is a drop-in security proxy for LLM applications. Point your OpenAI client at CollieAI and block prompt injection, jailbreaks, PII leaks, and more — before they reach your users.

Get started free See the docs

Free up to 15,000 API calls/month · No credit card · 5 min to integrate

protect_chatbot.py

# Before — direct call to OpenAI
from openai import OpenAI
client = OpenAI(api_key="sk-...")

# After — change base_url, full protection
from openai import OpenAI
client = OpenAI(
    base_url="https://api.collieai.com/v1",
    api_key="clai_your_project_key",
)

# Same code, now protected
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{ "role": "user", "content": user_msg }]
)

✓Ready. Input / output filtering · 9 rule types · full OpenAI compatibility

Live Demo

See CollieAI catch
a real injection — right now

Type any message and watch how CollieAI classifies it in milliseconds. Try a benign request or a prompt injection attempt.

Runtime detection

Every prompt scanned before it reaches your model

Full observability

Every request logged with triggered rules, tokens, and latency

Input / output

Filter both user prompts and model responses with configurable policies

CollieAI Proxy — Live Inspector● LIVE

User prompt

Waiting for input

Click "Analyze prompt" to inspect

Policy Engine

Configure rules visually

Set up input and output filtering policies with an intuitive rule builder — no code required.

app.collieai.io/app/analytics
app.collieai.io/app/playground
app.collieai.io/app/policies
app.collieai.io/app/logs
app.collieai.io/app/analytics
app.collieai.io/app/playground
app.collieai.io/app/policies
app.collieai.io/app/logs

Drop-in proxy

One line change. Works with:

47M+

API calls protected

1.8M

Attacks blocked

99.2%

Detection accuracy

~10ms

Avg. rule latency

Based on internal testing and early deployments. ML model rules: 10–50ms; regex/pattern rules: <5ms.

Quick start

How CollieAI keeps your AI safe

3 simple steps — integrate in minutes, see blocked attacks immediately.

Point your client at CollieAI

# Change two config values
base_url = "https://api.collieai.com/v1"
api_key  = "clai_your_project_key"

# Free: 20K tokens/month

→

Your code stays the same

CollieAI is a transparent proxy. Use the standard OpenAI SDK — every request is filtered by your policy rules automatically.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.collieai.com/v1",
    api_key="clai_...",
)
client.chat.completions.create(...)

→

Threats blocked, full audit trail

Only safe, compliant responses reach your users. Dashboard shows every blocked request with triggered rules and context.

// Blocked response (OpenAI-compatible)
{
  "error": {
    "message": "Content blocked by policy",
    "type":    "content_blocked",
    "code":    400
  }
}

Incoming threats

Prompt Injection

Malicious instructions

Jailbreak

Bypass safety rules

PII Exposure

Personal data leaks

Policy Violation

Off-topic, harmful content

Firewall

AI Firewall for LLM Applications

GuardrailsBlock injectionsi

Policy EngineContent rulesi

ML-rulesAI detectioni

EvaluationScore qualityi

LoggingFull audit traili

Safe output

Safe for RAG

Clean retrieval

Safe for Agents

Trusted actions

Safe for Apps

Compliant responses

Prices and plans

Start free on our cloud. Scale when you're ready.

Every plan covers the OWASP LLM Top 10 — prompt injection, jailbreaks, PII exposure, and policy violations. Full coverage from day one.

Free

$0forever

Evaluate CollieAI on our cloud. Perfect for testing and prototyping.

AI Guardrails

15,000 API calls / month

Maximum prompt size 10k tokens

All filtering functionality

Bidirectional (input + output)

Providers

OpenAI, Claude, DeepSeek, Gemini

Platform

OpenAI-compatible proxy

1 project

Dashboard & logs

Custom policies & rules

Async jobs & webhooks

Community support (GitHub Issues)

Get started free

Stop prompt injectionin your AI app.One line change.