Runtime guardrails that protect your Mosaic AI agents

Wrap your agent in the WonderFence SDK to enforce personalized guardrails trained on your policies, or configure out-of-the-box domain-specific policies for your business. Runtime guardrails inspect every turn, blocking multilingual and multi-turn attacks to keep agents on task.

How it works

Runtime Guardrails for Agents You Can Trust

WonderFence wraps Databricks Mosaic agents with runtime guardrails — intercepting prompts, enforcing policies, and monitoring outputs in real time.

Compliance evidence that keeps pace with AI

Find and fix what internal testing misses

Lifecycle-wide AI risk visibility

Technical Documentation

    mosaic_integration.py 
 from wonderfence import Guardrails
 from mosaic_agents import Agent
 
 guardrails = Guardrails(
     policies=["prompt_injection", "pii_detection", "tool_misuse"],
     mode="enforce",
     log_violations=True
 )
 
 agent = Agent("customer-support-mosaic")
 
 response = guardrails.run(
     agent=agent,
     input=user_message,
     context={"user_role": "customer", "channel": "web_chat"}
 )
 
 return response.safe_output 

Stay on your toes

What WonderFence adds to Mosaic Agents

Mosaic Agents gives you the agent. WonderFence gives you the controls — dynamic guardrails, granular policies, and full runtime visibility on top of what Mosaic Agents ships by default.

Explore the WonderSuite

Policies That Match Your Business

Mosaic Agents 's default guardrails are generic. WonderFence lets you enforce policies that reflect your compliance rules, brand voice, and risk tolerance — preconfigured options out of the box, or write your own. Apply them organization-wide or per-agent.

Defense Against Prompt Injection and Tool Misuse

Stop jailbreaks, data exfiltration, and tool abuse before they hit your databases, APIs, or customers. WonderFence catches malicious patterns that get past Mosaic Agents 's default guardrails.

Visibility Into What Agents Actually Do

Mosaic Agents shows you what you built. WonderFence shows you what your agents are doing in production — every prompt, tool call, and output, plus drift, policy violations, and unexpected behavior as it happens.

Trusted by the world’s biggest enterprises & foundation model labs

Trusted by teams shipping agents in finance, healthcare, and government

Integration FAQ’s

Do our existing platform guardrails cover us for HIPAA compliance?

Platform guardrails don't generate HIPAA audit evidence or enforce institution-specific PHI controls. WonderFence logs every interaction and enforces customizable policies at the interaction layer — built for what regulators actually require.

What happens to our compliance posture when the model gets updated?

WonderCheck supports ongoing adversarial evaluations at scheduled intervals, catching behavioral drift and PHI exposure regressions as models change. Your compliance posture doesn't reset after every update.

How do we validate clinical AI before go-live without slowing down our deployment timeline?

WonderBuild integrates into existing CI/CD workflows and returns prioritized findings against your specific clinical use cases so your team knows what to fix before launch, not after.

Who owns AI governance when it sits across clinical, IT, and compliance teams?

WonderSuite surfaces prioritized clinical risk findings for product teams, evidentiary logs for compliance, and lifecycle-wide visibility for the CISO. So, governance doesn't have to live in one place to work.