STALWART AI ASSURANCE

The product suite

Five products behind one API key, for the teams who build AI agents and the teams who have to sign them off.

Guardrail Bug Hunter

Find every way past your AI guardrails before an attacker or an accident does.

Scan your own guardrail policy and router and get back every route to a forbidden action. Each one is a finding with a severity, a way to reproduce it and a fix that has already been checked.

What it does

  • Every finding comes with a severity, a reproduction and a checked fix
  • Scans grouped into projects, each one compared with the last
  • A quality gate that can block a release
  • Triage as confirmed, false positive or accepted risk. An accepted risk needs a justification and an expiry date
  • A history for every finding: raised, seen again, fixed, back

Built for: AI platform and security teams, and risk owners

Guardrail Verification

A yes-or-no answer, with evidence, to “can this agent ever do X without Y?”

Ask whether one action can fire without its safety context, or check a whole set of rules at once. A bypass comes back with a concrete example and, where one exists, a repair.

What it does

  • Check a single rule or a whole rule set, with a certificate for every rule
  • Every bypass comes with a concrete example, and a repair where one exists
  • Certificates that you or your auditor can confirm independently
  • Every certificate download is logged, so your evidence has an access record

Built for: Risk and compliance teams, auditors and platform teams

Guardrail Writing

Write a rule in plain English and get back a policy that has already been checked.

Describe a rule in prose and name your own model. You get back a draft policy with a verdict on it, and a repaired draft when the first one is wrong and a repair is found.

What it does

  • Turns a plain-language rule into a draft policy, using your own model
  • Verifies every draft, and repairs it when it is wrong
  • Your key and your prompt are never stored

Built for: Policy authors and platform teams

MCP Tool Selector

Predictable, explainable tool choice for AI agents.

Choose the tool a user's request asks for by what each tool declares, with no AI model in the decision. Learn new words from your users' feedback and keep the vocabulary versioned.

What it does

  • Deterministic: the same request picks the same tool, every time
  • Every step names the rule that produced it
  • Learns new words from user feedback, with versioned vocabularies and rollback
  • Feedback is never used for training and never read across customers
  • Handles the calculations that no tool covers

Built for: Teams building agents on the Model Context Protocol (MCP)

Random Forest Verification

Show that your random-forest model behaves as required, or get the exact case where it does not.

Check a trained random forest against robustness, monotonicity, fairness and reachability requirements, or search around real records for counterexamples.

What it does

  • Checks robustness, monotonicity, fairness and reachability
  • Every violation comes with a concrete example, checked on the model itself
  • The search around real records is exhaustive, not sampled

Built for: Model risk, data science and compliance teams, for example in lending or insurance

Bot Builder

Coming next

Declare a domain agent that acts with all of the above built in.

  • One API key reaches every Stalwart product.
  • Test keys let you try calls out before you go live.
  • A web console manages your keys, usage, jobs and findings.
  • New customers receive their first key from our team: register your interest, or email newaccounts@stalwart.it.