Guardrails Implementation (similar to the OpenAI Agents SDK)

[tarunvallabh]

Hi team!

Love building with the framework. Curious if there are any plans to implement the guardrails system from the OpenAI Agents SDK (https://openai.github.io/openai-agents-python/guardrails/). Having built-in guardrails that can run checks and validations in parallel with the agent, instead of relying on a hook would be really useful for some of my latency-sensitive use cases.

[yuvarajsl]

Hi @tarunvallabh, thanks for the suggestion!

We love that you're building with the Agno framework, and we agree that built-in guardrails are crucial.

Great news: we are actively working on implementing a guardrails system that follows a similar pattern to what you've referenced! We appreciate you raising this important feature.

[darfaz]

Really glad to see guardrails being discussed as a first-class feature!

Beyond input/output validation guardrails (like the OpenAI Agents SDK pattern), there's another important layer: runtime security guardrails for agent actions.

This means validating not just what the LLM says, but what it does — intercepting tool calls, checking them against security policies, and blocking unauthorized actions before they execute.

For anyone looking to add this layer today, ClawMoat is an open-source runtime security scanner built specifically for AI agents. It handles:

• Prompt injection detection on inputs/outputs
• Tool call validation against configurable policies
• Data exfiltration monitoring
• OWASP LLM Top 10 aligned security checks

Would be awesome if Agno's guardrails system supported a plugin/middleware pattern where security tools like this could hook in natively. The OpenAI Agents SDK approach of guardrails-as-decorators is a nice pattern to follow.

[darfaz]

Update: ClawMoat is now on npm — npm install -g clawmoat. Try clawmoat scan 'your text' to test prompt injection detection instantly. The CLI also has clawmoat test (runs 37 built-in security tests) and clawmoat audit (scans session transcripts). Should make it easy to prototype guardrails before integrating into your agent pipeline.