Your AI is out of control
User behavior is unpredictable, on the edge – or even well beyond – your intended use cases. Without proper guardrails, your AI will respond to anything, even when it definitely shouldn't.
Poor Performance
When queried off-topic, your AI is likely to behave subpar, leading to dissatisfaction and loss of trust.
Compliance and Safety Risk
Responses to out-of-scope queries can breach your controls, with consequences ranging from policy violations to full-blown PR nightmares.
Security Vulnerability
Out-of-scope and restricted queries create an attack surface that can be exploited to leak secrets or bypass safeguards.
Higher Cost
You are paying for those tokens, especially those output tokens, driving up costs without delivering any business value.
Enter ScopeGuard
A lightweight, high-performance guardrail that sits between your users and your Agents, ensuring every conversation stays within bounds.
Define what your assistant is meant to do
Start by defining what your assistant is meant to do. This specification — your AI Service Description — captures the intended scope, role, and constraints of the assistant.
Start with your system prompt and refine it as needed.
Classify the scope of each user query
On every request, the guardrail runs first.
It classifies the user query against your AI Service Description and returns a scope signal.
Act on the scope signal
Decide how to act on the scope identified.
Let legitimate requests through, block and monitor restricted requests, or design your own workflow.
Multilingual
Native support for English, Italian, French, Spanish and German with consistent accuracy.
Hosted
Fully managed infrastructure with 99.99% uptime SLA and automatic scaling to handle any load.
Open-Source
Core classification engine available on GitHub for transparency and on-premise deployment needs.
Production ready, with no integration hassle
Add ScopeGuard to your AI in minutes, with no code changes required.
You are a helpful assistant. Your functionalities are:
Configurable
Simply describe your AI service in plain language to configure ScopeGuard to your exact needs.
Response Time < 1 Second
ScopeGuard processes requests in less than one second on consumer-grade hardware. Your users won't notice any delay.
State-of-the-Art
ScopeGuard accurately identifies the scope of your users queries, outperforming leading competitors in benchmark tests.
Configurable, Fast, State-of-the-Art
Deploy Your Way
Choose the deployment model that fits your organization's needs, from open-source freedom to enterprise-grade managed services.
Open-Source
ScopeGuard is open-source and open-weight, in line with our philosophy.
- Full source code access on GitHub
- Community support and active development
- Apache 2.0 License for maximum flexibility
Hosted Access
Production-ready API with enterprise SLA and better performances.
- Improved models and guardrail performances
- 99.99% uptime SLA
- Auto-scaling to handle any load
On Premise
Self-host within your own infrastructure with guaranteed data sovereignty.
- Complete data isolation and sovereignty
- Access to improved models
- Dedicated technical account