We'll evaluate your current guardrails, and build you custom ones.
Generic guardrails block real customers, miss real violations, and add latency. We assess your current guardrails, build custom policies for your agent, and give you the report. Zero cost, zero obligation.
Claim my free assessment
Claim your
free assessment
We assess your current guardrails, build custom policies for your agent, and run both on the same traffic. The report shows the difference. Nothing to buy, nothing to install.
One agent. Custom guardrails. One report you keep.
Tell us what your agent may and may not do, plus a few examples. This helps us understand the context.
We score your current guardrails, then build a custom one for your agent, ready within as little as 24 hours.
Receive a report showcasing your current guardrail quality vs our custom guardrails.
What you get, free
No credit card. No sales deck. Just one call and a free assessment.
Your guardrails are already making calls. Find out how many they get wrong.
Generic guardrails miss what matters
and block what doesn't
A fixed list that isn’t written for your agent lets your violations through, turns away real customers, and adds a round trip to every call. Alice trains a policy on your rules and your traffic, and runs it fast enough that nobody notices.
<120 ms
P95 latency at runtime
99%
Accurate once calibrated to your traffic
<24 hrs
To a live custom policy
A policy trained on your rules, in your words
We train a small, fast model on your rules, so it catches what matters to you and lets your customers through.
Your rules, in plain language
You describe what the agent may and may not do, with a few examples. We help you write it, or choose from our calibrated policies list.
Frontier accuracy, small-model speed
A large model learns your policy and generates examples, which you review. A small, fast model is then trained on the result, and is live within 24 hours.
Precise on both sides, fast at runtime
Each policy only catches real violations. When the small model flags something, a larger model carries out an additional check, ensuring high accuracy at all times.
Calibrated to your traffic, or built from your rules
Both types of policies are fine-tuned to fit your agent.
Trained for the risks every agent faces, tuned on your traffic. Pre-built, 99% accurate once calibrated.
Prompt injection and jailbreaks · data leakage and PII · hate, harassment and violence · sexual and self-harm content · illegal activities · impersonation + more
Rules only you can define, built with the teacher-student flow above. Live within 24 hours.
Example: "never quote a delivery date", "never recommend a competitor"
Rabbit Hole, Alice's adversarial intelligence engine
Hundreds
8
120+
Multimodal
Generic vs custom, measured on your own traffic
How your guardrails
perform on real violations
How often you turn
away legitimate requests
Latency, current
vs custom, side by side
Which policy to ship,
and what to tune
What's included
Scope
One agent, one policy: your current guardrails vs a custom one
Compared against
Provider-native, off-the-shelf, or homegrown guardrails
What we need
Traffic sample or test set, a few allowed and blocked examples, your agents description
Your effort
One intro call, one calibration round, a walkthrough call
Turnaround
Policy live within 24 hours of calibration. Report in 1 business day
Cost
None. No purchase, no obligation