- Home
- AI Guardrail Software
- OpenAI Guardrails
OpenAI Guardrails
San Francisco, California, United States
What it is
OpenAI Guardrails adds configurable safety and compliance checks to LLM applications by validating inputs and outputs. Its checks include PII masking, moderation, jailbreak detection, off-topic prompt checks, URL filtering, and hallucination detection; prompt-injection detection for agent tool calls is experimental. The Python package provides wrapped OpenAI clients for Chat Completions and Responses API calls and integrates with the OpenAI Agents SDK. Its quickstart also documents Azure OpenAI clients and compatibility with OpenAI-compatible APIs, including a local Ollama example. Teams can create pipeline configurations with the Guardrails Wizard or define them in JSON, then evaluate performance on labeled datasets. SDKs are available in Python and TypeScript/JavaScript. The Python project is Preview, available under the MIT license, and requires Python 3.11 or newer. Configurations using Contains PII also require the spaCy model during build or deployment. Guardrails is free and open source, but calls to paid OpenAI APIs incur charges for developers. Third-party services such as Presidio may also be used; OpenAI says it does not develop or verify them.
For and against
For
- Checks inputs and outputs for several safety concerns.
- Integrates with OpenAI clients and the Agents SDK.
- Supports Azure and compatible APIs, including local Ollama.
- Offers JSON configuration and labeled-dataset evaluation.
- Available under the MIT license.
Against
- Python package is labeled Preview.
- Python requires version 3.11 or newer.
- Contains PII configurations require a spaCy model.
- Paid OpenAI API calls incur developer charges.
The verdict
OpenAI Guardrails provides configurable checks, client integrations, and evaluation tooling for LLM applications. Consider its Preview status, Python requirements, third-party components, and separate paid API charges when deciding whether it fits your deployment.
Who it suits
It suits developers adding configurable safety checks to LLM applications, including teams using OpenAI clients, the Agents SDK, Azure OpenAI, or compatible APIs. Developers should be prepared to manage the Python requirements and any paid API charges.
On the record
- Agents integration
- The Python SDK integrates with the OpenAI Agents SDK through GuardrailAgent.github.com · read 4 Oct 2026
- Availability
- The Python project is labeled Preview and is available under the MIT license.github.com · read 4 Oct 2026
- Checks
- Available checks include moderation, jailbreak detection, PII detection, URL filtering, secret key detection, hallucination detection, prompt injection detection, and custom prompt checks.github.com · read 4 Oct 2026
- Client integration
- The Python package provides drop-in replacements for OpenAI clients and integrates with the OpenAI Agents SDK.github.com · read 4 Oct 2026
- Configuration
- The Guardrails Wizard creates pipeline configurations that can be exported for use with the SDK.github.com · read 4 Oct 2026
- Evaluation
- Guardrail performance can be measured on labeled datasets using the exported configuration and the evaluation dependencies.github.com · read 4 Oct 2026
- Evaluations
- The Python package includes evaluation tooling for measuring guardrail performance on labeled datasets.github.com · read 4 Oct 2026
- Maker
- OpenAI describes itself as an AI research and deployment company; it was founded in 2015 and is based in San Francisco.openai.com · read 4 Oct 2026
- Maturity
- The Python repository labels OpenAI Guardrails as a preview.github.com · read 4 Oct 2026
- OpenAI clients
- The Python SDK provides wrapped OpenAI clients for Chat Completions and Responses API calls.github.com · read 4 Oct 2026
- Other providers
- The quickstart documents Azure OpenAI clients and compatibility with OpenAI-compatible APIs, including a local Ollama example.github.com · read 4 Oct 2026
- Platform support
- The maker provides Python and TypeScript/JavaScript Guardrails SDKs for use in applications.openai.github.io · read 4 Oct 2026
- Provider compatibility
- The quickstart documents Azure OpenAI clients and compatibility with OpenAI-compatible APIs, including a local Ollama example.github.com · read 4 Oct 2026
- Purpose
- OpenAI Guardrails adds configurable safety and compliance checks to LLM applications by validating inputs and outputs.github.com · read 4 Oct 2026
- Requirements
- The Python package requires Python 3.11 or newer; configurations using Contains PII also require its spaCy model during build or deployment.github.com · read 4 Oct 2026
- Third-party component
- Guardrails may use third-party services such as the Presidio open-source framework, which OpenAI says it does not develop or verify.github.com · read 4 Oct 2026
- Third-party services
- Guardrails may use third-party services such as the Presidio open-source framework, which is subject to its own terms and is not developed or verified by OpenAI.npmjs.com · read 4 Oct 2026
- Usage costs
- Guardrails calls paid OpenAI APIs, and developers are responsible for the associated charges.github.com · read 4 Oct 2026
