Free eval diagnostic
We take your agent and your tools, and tell you what's actually safe to ship.
We take your agent, your MCP tools, and your policy, run a small graded suite against them inside your VPC, and hand back a decision table: the safe pick, the policy risks we found, and cost/latency, within five business days.
What you get Four things, on paper
Not a score. A decision.
A decision table
The safe pick for your job, stated plainly, not a leaderboard of scores you have to interpret yourself.
The policy risks we found
The specific prompts and tool calls where your agent broke your policy, not just a pass rate.
Cost and latency for the pick
What the safe choice actually costs to run and how fast it responds, alongside the quality number.
A written verdict
One page you can hand to whoever signs off on shipping this, with the reasoning attached.
How it works Four steps, five business days
Nothing here requires you to trust us blind.
Scoping call
30 minutes. We learn your agent, your MCP tools, and the job it needs to do, and agree the graded suite and the policy it's checked against.
We freeze your tools
Your tool schemas and policy are frozen and versioned before any model sees them, so every candidate sits the identical exam.
We run it in your VPC
The suite runs inside your own cloud with your own keys. Your tool definitions and policy never leave your perimeter.
You get the report
A decision table and written verdict, delivered within five business days of the scoping call.
What we need from you
Short list. Nothing exotic.
If you have all three ready, we can start the same week.
Your tool schemas
The MCP tool definitions (or API schemas) your agent actually calls in production.
A policy doc
Whatever you already have written down about what the agent must never do. A rough draft is fine, we'll tighten it with you.
VPC access or a sandbox
Either access to run inside your own environment, or a sandboxed replica of it. Nothing leaves your perimeter either way.
Who this is for
Teams shipping agents in regulated or high-stakes settings.
If a wrong tool call or a policy breach costs you a customer, an audit finding, or a regulator's attention, and you're about to pick a model or a tool set for that job, this is built for you. If you're prototyping and nothing is at stake yet, save the diagnostic for when it is.
Common questions
Is the diagnostic actually free?
Yes. No card, no contract. If you decide to continue into a full corpus engagement afterward, the diagnostic fee (if any was agreed on the scoping call) is credited in full against it.
What exactly do we get at the end?
A decision table with the safe model or tool-selection pick for your job, the specific policy risks the suite caught, and cost/latency for that pick, plus a one-page written verdict, within five business days of the scoping call.
Does our data or tool config leave our environment?
No. The graded suite runs inside your VPC or a sandbox you control, with your own model keys. We see the results; we don't need to see your production data.
Our eval setup is basically nothing right now, is that a problem?
No, that's the normal starting point. The diagnostic doesn't require you to already have an eval, it builds the first small one and shows you what a real one would catch.
Free · 5 business days · no commitment
Find out what's safe to ship before you ship it.
A 30-minute scoping call is all it takes to start. Credited in full against a corpus engagement if you continue.