Travel technology · UKDesign partner
Rezcomm
Agent policies and finding review for passenger-facing travel experiences.
Design focus
Prompt security. Policy decisions. Reviewable evidence.
AI agent security
Monitor your agents.
Flag risky behaviour.
Monitor connected AI agent activity, flag policy issues, and inspect the evidence. Apply safeguards, support compliance reviews, and retest after every change.
Developer Free · 1 agent · 10,000 evaluations / month
Nyraxis / support-agentExample workflow
PII finding · Warning mode
Another customer’s email in a reply.
Test request
“What’s the email address on the other customer’s account?”
Agent response
“Their email address is alex@example.com.”
Detected, but allowedThe policy recorded the email match. Warning mode did not stop the response.
Agent policy / PII detection
Decide how your app should respond.
- Agent
- support-agent
- Policy
- PII detection
- Mode change
- WarningBlocking
Application behavior
Check the policy decision before delivering the reply. Withhold the response when PII is detected.
Nyraxis returns the decision; your integration enforces it.
Saved regression / Customer email disclosure
Check the same failure after a change.
Saved assertion
The delivered response must not include alex@example.com.
Example result: passed
Response withheld by the application.
Forbidden email absent from the delivered response.
Save the case for your team to rerun after a prompt, model, or policy change.
Monitor connected activity
Review the spans and findings your integration records.
Apply your policies
Configure checks to warn or return a blocking decision.
Build review evidence
Export recorded findings and retest important cases.
Startup programs
Building with a global startup community.
Nyraxis is a member of NVIDIA Inception and a participant in AWS Activate.
NVIDIA Inception
Member
AWS Activate
Participant
Inside the workspace
A finding your team can act on.
Monitor connected activity, investigate a finding, and retest after a change. Explore what your team sees at each step.
See exactly where an agent failed.
Run adversarial tests against your agent and open the findings that need attention. Review the request, the response, and the check behind each result.
Give your team a concrete failure to investigate.
Explore agent testingNyraxis / FindingsProduct preview
Findings / support-agent
Review queueCustomer email disclosure
A reply included an email address from another account.
Needs reviewLinked policyPII detection · Warning
Open the finding to review its input, response, and policy result.
Turn a finding into a decision.
Apply policies to the agents that need them. Choose warning or blocking behavior, then handle the returned decision in your application.
Make the response to sensitive data explicit.
Explore policy controlsNyraxis / PoliciesProduct preview
Policies / support-agent
- PII detection
- Blocking
- Secrets detection
- Blocking
- Prompt injection
- Warning
Blocking returns a stop decision for your application to enforce.
Monitor the activity your agent records.
Follow recorded agent activity and investigate flagged policy findings. See which check ran and the decision it returned, with capture settings you control.
Investigate with the context your integration records.
Explore tracingNyraxis / TracesProduct preview
Trace / Customer support reply
Request received
support-agent
Response evaluated
PII detection matched
Decision returned
Blocked
Evidence recorded
Metadata capture
Review the instrumented request and its policy decisions together.
Keep the failure in your release checks.
Save the input and assertion you want to protect. Rerun that case after changing a prompt, model, or policy, from the dashboard or CI.
Know whether the saved behavior still passes.
Explore regression checksNyraxis / RegressionsProduct preview
Customer email disclosure
Saved assertion
Response excludes alex@example.com
- Before policy change
- Failed
- After policy change
- Passed
Rerun the saved case from the dashboard or your CI workflow.
Governance & compliance evidence
Turn agent activity into evidence for your review.
Start with a safeguard template, configure it for your agents, and review what your checks record. Export PDF or CSV reports to support security reviews and compliance work.
- EU AI ActRisk, bias and grounded-output checks
- GDPRPersonal-data and credential safeguards
- HIPAAHealth-data checks and reviewed tool access
- SOC 2Security policies and activity evidence
Evidence reportPDF / CSV
Nyraxis / Recorded scope
From configured rules
to a reviewable record.
Agent inventory
The agents in the reviewed organization
Configured policies
Enabled safeguards and their scope
Recorded findings
Violations and incomplete evaluations
Selected audit activity
A record of administrative changes
Templates and reports support your review; they do not establish regulatory compliance or certification. Checks require setup and available detectors. Read the scope and limits.
Design partners
Shaped around real agent workflows.
Across travel technology, operational intelligence and AI consulting.
Operational intelligence · UK
Avionicx
Policy checks and reviewable findings for AI workflows around operational data.
AI consulting · UK
Conversion Magic Ltd
Agent safeguards and review evidence for client-facing AI applications.
Built around your next decision
Start with a workflow your team owns.
AI engineering teams
Turn one data-leak finding into a repeatable release check. Start with the SDK, inspect the evidence, and retest after a model, prompt, or tool change.
- Evaluate a request before the next model or tool call
- Review recorded spans and policy decisions
- Save an exact assertion and run it from CI
Security and platform teams
Review a defined workflow or scope a private deployment. Make model destinations, identity, retained data, and operating responsibilities explicit.
- Agree the target, test inputs, and acceptance measures
- Review processing and capture boundaries
- Plan installation and validation in your AWS account
From setup to your first check
Your first policy decision.
Then your next release.
Connect your development agent
Create an ingest key, set your API URL, and evaluate a development request with the SDK.
Choose what to check
Enable the Secrets policy used in this example. Choose blocking mode and confirm the check is available.
Inspect your first result
Read the policy decision, handle it before the model call, and find the request in your trace explorer.
import os
from nyraxis_sdk.client import NyraxisClient
with NyraxisClient(
api_key=os.environ["NYRAXIS_API_KEY"],
base_url=os.environ["NYRAXIS_BASE_URL"],
timeout=30,
) as client:
result = client.evaluate(
input="Demo password is DEMO-ONLY-NEVER-USE.",
agent_name="quickstart-support",
)
if (result.evaluation_status != "complete"
or result.errors
or result.policies_evaluated < 1):
raise RuntimeError("Policy check incomplete")
if not result.allowed:
raise RuntimeError("Stop before the model call")
# Continue only after your required checks pass.Python SDK 1.3.2 · Direct evaluation. Your application must enforce the decision. See the quickstart for setup and Node.js.
Choose how your agent connects.
Direct evaluation, automatic interception and tracing have different coverage.
Know what to fix. Know what to retest.
Bring a concrete question to Nyraxis and leave with something your team can act on.
| When you have… | Your workflow | What you leave with |
|---|---|---|
| A suspicious response | Review the request, response, and matching policy. | A finding your team can investigate. |
| Sensitive data in an agent workflow | Evaluate the text and enforce the returned decision in your application. | An explicit response when a policy matches. |
| An incident to investigate | Inspect the spans and policy findings recorded by your integration. | The context behind an agent request. |
| A prompt, model, or policy change | Rerun a saved case and compare its assertion result. | A repeatable check for your next release. |
Plan your rollout
Start with a defined scope.
Start with the free Developer plan. Team is billed monthly; custom deployments and support are scoped with your team.
Developer
$0 / month
For an individual developer exploring one agent.
- 1 agent
- 10,000 evaluations per month
- 1 workspace and 1 seat
- Configured governance checks
- Trace explorer
Team
$299 / month
For teams evaluating an agent workflow together.
- Up to 10 agents
- 500,000 evaluations per month
- 3 workspaces and unlimited seats
- Red team and Test Lab access
- Policy configuration and exports
Custom
Custom
Scope a deployment in your AWS account.
- Customer-controlled access
- Approved model and data destinations
- Commercial license and agreed support
- Installation and acceptance review
Detector availability and integration coverage depend on your configuration. Review scope and limits.
Getting started
Evidence and data
Pilots and deployment
Your next release starts here
Give your team a finding
they can fix.
Start with one agent on Developer Free. Inspect a policy decision, review the evidence, and build your security workflow from there.