Skip to main content
GuardAI
For Developers

Add a policy layer to your existing model integration

Route model calls through GuardAI's OpenAI-compatible API. Configure your policy set, then evaluate allowed, blocked, and rewritten responses with your existing application.

Try Interactive DemoRequest Demo

Connect your application.

Set the GuardAI base URL and API key. Keep the chat-completion call structure and select your configured model.

app.py
1 from openai import OpenAI
2  
3 client = OpenAI(
4−    api_key="your_openai_key"
5+    base_url="https://api.guard4ai.org/v1",
6+    api_key="your_guardai_key"
7 )
8  
9 response = client.chat.completions.create(
10     model="gpt-6-astra",
11     messages=[{"role": "user", "content": prompt}]
12 )
13  
14 # That's it. Same SDK, same models, same code.

Built for your workflow.

Keep policy configuration separate from model calls. Evaluate enforcement, response quality, and latency with your application’s requests.

01

Separate policy evaluation

Check the model’s draft response against your rules before delivery. Measure the added latency with your application’s requests.

02

Policies independent of the model

Keep policy definitions across supported models. Test model changes against your existing evaluation cases.

03

Evaluate before rollout

Test permitted requests, off-policy responses, and instruction overrides. Use failed cases to refine your policies and measure false refusals.

04

Full visibility and control

Review policy-decision metadata and configure content logging, storage access, and retention for your deployment.

05

RAG-ready protection

Apply disclosure rules to responses drawn from your knowledge base, including confidential information in the model’s context.

OpenAI-compatible integrations.

Connect applications that support a configurable OpenAI-compatible endpoint. Check model support and request features against your integration.

Supported model providers

Keep model selection separate from policy definitions. Verify the endpoint, model, and request features for each provider you connect.

Cloud ModelsOpen-SourceLocal ModelsFine-tuned

Popular Frameworks

Configure LangChain, LlamaIndex, or Haystack to call GuardAI’s endpoint with a supported model.

LangChainLlamaIndexCustom RAG

Choose your deployment.

Use the managed API or run GuardAI containers in your own infrastructure. Both require outbound access to Vertex AI for policy evaluation.

API INTEGRATION

Managed Proxy

Configure your application or SDK with GuardAI's endpoint, API key, and model. GuardAI checks the model's draft response against your policies before delivery.

Existing apps that support custom backends
Your code using the OpenAI SDK
LangChain, LlamaIndex, and other frameworks
ENTERPRISE

Private Deployment

Run GuardAI containers in your own VPC or on-prem infrastructure. You control network boundaries and data residency.

Full control over your data environment
Custom model configurations
Your infrastructure, your rules

Add GuardAI to your stack.

See the same-base-model benchmark results on the whitepaper.

Try Interactive DemoContact Sales