Your AI agents decide in milliseconds. Prove every decision was safe.
MastGuard records every action your agents take, flags the risky ones in real time, routes the high-stakes ones to a human, and turns the whole trail into a compliance report your auditor will accept.
Monitoring AI agents across healthcare, fintech, and enterprise teams.
THE GAP
AI agents are already touching your most sensitive workflows.
Agents now read patient records, approve transactions, and make decisions that used to need a person. Most teams deploy them with almost no visibility into what the agent actually did once it was live.
When an auditor asks what your agent decided last Tuesday, or whether it ever stepped outside its scope, most companies cannot answer. That is the gap MastGuard closes.
No visibility
You cannot see what an agent did, only what it was supposed to do.
No proof
No tamper-proof record to show a regulator or your board.
No early warning
No alert when an agent starts acting outside its intended scope.
HOW IT WORKS
Three steps to full visibility.
Instrument your agent
Wrap your agent with the AgentShield SDK. Three lines of TypeScript. No infrastructure changes, no downtime.
npm install @auxdynamics/mastguard-agent-sdkMonitor every action
Each action becomes a structured event scored against your policies in real time. Step loops, scope violations, and injection attempts show up on the dashboard as they happen.
Generate reports on demand
Pick a framework and a date range, then export a PDF. GDPR, HIPAA, EU AI Act, and SOX, formatted the way your auditor expects.
THE PLATFORM
Six capabilities. One governance layer.
Everything you need to monitor, control, and prove the behavior of your AI agents.
Real-time agent monitoring
MastGuard captures every agent action as a structured event: what the agent did, what data it touched, and what it decided. Every event lands in a tamper-proof audit log, SHA-256 chained and append-only, so the record can be verified and never quietly edited.
MAST failure detection
We classify agent failures with the MAST taxonomy from UC Berkeley research presented at NeurIPS 2025: Memory, Action, Space, and Time. That covers prompt injection, scope violations, data exfiltration attempts, and multi-step attack chains, caught in real time.
Human-in-the-Loop
Route high-risk decisions to a human before the agent acts. You set the policy. Reviewers approve, reject, or escalate from a queue, and every decision is logged and time-stamped.
Automated compliance reports
Generate GDPR, HIPAA, EU AI Act, and SOX reports straight from your audit data. One click, PDF ready for regulators.
RedScan red-teaming
Run 468 adversarial test cases against your agent and get an AI Risk Score from 0 to 100, mapped to NIST AI RMF, ISO 42001, and the EU AI Act.
ProvenanceGuard
Detect poisoned training data with absolute-count cluster analysis, and generate an AI-BOM for every dataset, aligned to EU AI Act Article 11.
REDSCAN
Red-team your agent before an attacker does.
RedScan runs 468 adversarial test cases against your agent, drawn from the AgentHarm benchmark and driven by our Boundary Point Jailbreaking engine. Prompt injection, jailbreaks, and boundary violations, all in one automated run.
You get an AI Risk Score from 0 to 100 and a board-ready PDF report, with results mapped to NIST AI RMF, ISO 42001, and the EU AI Act.
AI Risk Score
Low risk · 468 cases run · 6 findings resolved
COMPLIANCE
Mapped to the controls your auditor will ask for.
MastGuard maps your agent activity directly to the frameworks that govern regulated work.
PRICING
Start free. Upgrade when you scale.
All prices in USD. No credit card to start.
Free
Free forever. No credit card required.
- 1 agent
- 50,000 events per month
- 1 team member
- 5 policy rules
- 7-day audit retention
- 1 RedScan per month (AI Risk Score, no PDF)
Pro
Billed monthly. Cancel anytime.
- Up to 10 agents
- 2,000,000 events per month, then metered overage
- 10 team members
- Unlimited policy rules
- 90-day audit retention
- 10 RedScans per month with full PDF report
- 5 ProvenanceGuard scans per month
- Threat Intelligence feed (opt-in)
- Full HITL review workflow
- GDPR and SOX reports, webhooks, TypeScript SDK
- Priority support
Enterprise
HIPAA tier with BAA included.
- Unlimited agents, events, and team members
- Unlimited RedScan and ProvenanceGuard
- All frameworks: GDPR, HIPAA, EU AI Act, SOX, MiFID II
- Private threat signatures and SIEM export
- 99.9% uptime SLA, 7-year audit retention
- Dedicated Azure VNet infrastructure (coming soon)
- Dual-approval workflows (coming soon)
- Dedicated customer success manager
Need more scans? Buy scan credits.
Prepaid credits cover RedScan and ProvenanceGuard beyond your monthly quota. Credits roll over and never reset.
INTEGRATION
Works with the models you already use.
Wrap OpenAI, Anthropic, or Azure OpenAI in three lines. The AgentShield SDK intercepts each call, scores it, and forwards it. No rewrites.
wrapOpenAI()Drop-in wrapper for the OpenAI SDK. Every completion is governed.
wrapAnthropic()Same pattern for the Anthropic SDK and Claude models.
wrapAzureOpenAI()Governs Azure OpenAI deployments without changing your call sites.
WHAT TEAMS TELL US
The value shows up the first time someone asks for proof.
We turned a week of audit prep into a single PDF export. The reviewers stopped asking how we know what the agent did, because now we can show them.
Compliance Officer, Fintech
RedScan found a prompt injection path we had no idea was open. We fixed it before the agent went anywhere near production data.
AI Engineering Lead
Human-in-the-Loop let us put a clinician in front of the decisions that mattered without slowing down the ones that didn't.
Head of Platform, Digital Health
FAQ
Common questions.
Yes. Free is $0 per month with no credit card and no expiry. You get 1 agent, 50,000 events per month, 1 team member, 5 policy rules, 7-day audit retention, and 1 RedScan per month with the AI Risk Score.
No. The AgentShield SDK wraps your existing OpenAI, Anthropic, or Azure OpenAI calls in three lines. Your call sites stay the same.
RedScan runs 468 adversarial test cases against your agent, covering prompt injection, jailbreaks, and boundary violations. It returns an AI Risk Score from 0 to 100, and on Pro and Enterprise a board-ready PDF mapped to NIST AI RMF, ISO 42001, and the EU AI Act.
ProvenanceGuard detects poisoned training data using absolute-count cluster analysis, and generates an AI-BOM for each dataset. It is available on Pro (5 scans per month) and Enterprise (unlimited). It is coming soon on Free.
MastGuard runs on Microsoft Azure, with data stored in Azure Canada or Azure US East regions. Audit logs are encrypted, tamper-proof, and append-only. We do not use your agent event data to train AI models.
Pro includes 2,000,000 events per month. Beyond that, usage is metered as overage. RedScan and ProvenanceGuard run on prepaid scan credits, sold in packs of 5, 10, or 25 that roll over and never reset.
Enterprise customers can request a Business Associate Agreement. Email info@auxdynamics.com. Healthcare teams should not store PHI on Free or Pro without a signed BAA.