§ 00

When the hospital asks
what your AI did, can you prove it?

For AI-native healthcare startups in active sales conversations with hospitals who keep hitting the “we need to complete our security review” wall.

your_app.py
import cyriscyris.init(api_key="...")
AUTO—INSTRUMENTED, NO CODE CHANGES
OpenAIAnthropicAzure OpenAIAWS BedrockGoogle GeminiVertex AIMistralCohereGroqTogether AIOllama (official client)MCP servers

Hospitals now require AI vendors to prove explainability, anomaly detection, and data lineage before they sign — and most startups can’t. Cyris wraps your agents in two lines of code and turns every decision into audit-ready evidence.

What you get back
✓Per-decision audit log — input, output, agent, timestamp
✓Real-time anomaly feed with severity classification
✓Patient-level workflow trace, end to end
✓Incident history with downstream impact tracked
✓Drift metrics per agent, exportable per window
✓PHI flagged before it ever leaves your process
§ 01 — WHAT HAPPENS AFTER TWO LINES

Your engineers can't instrument 14 agents before next Tuesday's hospital review. Cyris does it for them.

Install the SDK. Run your app. Every OpenAI, Anthropic, and Bedrock call your agents make is captured from that moment forward — no agent rewrites, no config files, no instrumentation sprint, no second job for your team.

Within a week Cyris has discovered every agent, mapped the handoffs between them, and drawn the workflow graph from real traffic — the same workflow graph the hospital’s compliance team is going to ask you for.

Day 1Agents discovered from live traffic
Day 3Connections mapped between agents automatically
Day 7Complete workflow graph — drawn from real data
Day 30Full audit trail — ready for any questionnaire
AGENT DISCOVERYLIVE
T
Triage
Claude 3.5
42 calls
C
Coordinator
GPT-4o
38 calls
R
Records
Claude 3.5
84 calls
B
Billing
GPT-4o-mini
91 calls
4 agents detected3 connections inferred$24.20 tracked
§ 02 — YOUR DAILY BRIEF

Catch the bad decision before the clinician does.

Your AI made 12,000 decisions overnight. Three of them failed. One agent's cost-per-call jumped 22% because someone bloated a system prompt at 11pm. Your CFO finds out next month. Your clinicians find out tomorrow morning. You find out in 60 seconds, over your first coffee — with the receipts ready to forward to whoever needs them.

Your agents made 215 decisions yesterday8:00 AM
AGENTCALLSCOSTLATENCYERRORS
Medical Scribe89$12.401.8s0
Insurance Check84$3.200.9s2
Triage Agent42$8.602.1s0
⚠ Triage cost-per-call up 22%. Input prompts grew 40%.
AUDIT TRAIL COVERAGE
67%
§ 03 — WHEN THE QUESTIONNAIRE ARRIVES

When the questionnaire arrives, the deal is on the line.

Last time, it took 3 weeks.

This time: 90 seconds.

Last time your team spent three weeks pulling logs, drafting answers, second-guessing the wording. The deal stalled. The hospital went with a competitor who'd done it once before. This time you drop the PDF in. Cyris reads the questions, fills the answers from your real operational data — real timestamps, real counts, real evidence — and flags the ones that need a human.

Vanta and Drata built compliance tooling for the hospital’s side of the review. Cyris built it for yours.

HIPAASOXEU AI Act
What hospitals askWhat Cyris generates
“Show us your audit log for every AI decision.”
Hash-chained per-decision log — input, output, agent, timestamp, immutable.
“How do you detect anomalies in model behavior?”
Real-time anomaly feed with severity classification, exportable as evidence.
“Can you trace a patient outcome back to the AI step that caused it?”
End-to-end patient-level workflow trace, every hop recorded.
“What’s your incident response history?”
Incident log with downstream impact tracked per event.
“How do you monitor model drift over time?”
Per-agent drift metrics, exportable per window with raw counts.
“How do you prevent PHI from leaving your environment?”
PHI detected and flagged before it ever leaves your process.
§ 04 — THE LIVE COMPLIANCE URL

Stop sending PDFs. Send a URL the hospital can check at any hour, on any day.

Compliance documentation used to be a one-time PDF you scrambled to assemble before every deal. Cyris turns it into a live data product. The same trust URL that satisfies your first hospital becomes the infrastructure that closes your next fifty — compounding by the day, not redrafted from scratch.

One click in your dashboard generates a private URL. Share it with the hospital’s compliance team. They see audit readiness, agent stats, hash chain integrity, anomaly counts, BAA status — in real time, with no login, no PDF, and no boilerplate. No invocation content, no PHI, no operational details ever leak.

“Vanta and Drata built the trust center for the hospital’s side of the review. We built it for yours.”

cyrisai.dev/trust/a8f2c1e4…LIVE
Northstar Health AIPOWERED BY CYRIS
AI AGENT COMPLIANCE TRUST CENTER
AUDIT READINESS94%
Audit readiness94%
Days continuously logging127
Total decisions on the record1,284,902
Hash chain integrityverified
PHI events logged8,412
Anomalies, last 30 days2
BAA on fileyes
No login required · refreshes every page load✓ CHAIN VERIFIED
§ 05 — WHEN SOMETHING BREAKS

When something breaks at 2 a.m., the trace is already there.

Your AI made 847 decisions today. If one of them was wrong — wrong dosage, wrong code, wrong patient, wrong claim — you have to know which one, why it happened, every record it touched, every downstream agent that acted on it. By the time the hospital’s safety officer is on the phone, the answer is already on your screen — not “we’ll get back to you next week.”

INCIDENT TRACESession: patient_admission_4821 · April 4, 2026
CHAIN VERIFIED ✓
Root cause: Insurance Verifier timeoutAgents affected: 3/6Duration: 19.1sPHI: 4 agents
14:42:12✓
Patient IntakeWebhook
Received intake form
PHI: name, dob, mrn
a8f2c1
14:42:13✓
Triage AgentClaude 3.5
ESI Level 1 — STEMI protocol
1.2s · $0.04 · PHI: diagnosis
3b9d2f
14:42:15✓
CoordinatorGPT-4o
Dispatched to 5 departments
1.8s · $0.06
e4c17a
14:42:16✗
Insurance VerifierCustom REST
TIMEOUT — payer API unresponsive
Retried 3x. Max retries exceeded.
7f2b8d
14:42:31⚠
Prior AuthGPT-4o
NOT TRIGGERED — dependency unresolved
Waiting on Insurance Verifier
——
14:42:32✓
Clinical NotesClaude 3.5
Note generated — missing coverage
2.1s · $0.08 · incomplete
9d2fb8
Export as incident report →Export JSONCopy link
§ 06 — WHERE YOUR MONEY GOES

One agent burns 60% of your budget for 12% of your output.

You’d never know unless you tracked every call. Cyris tracks every call — per agent, per model, per day. The morning your token spend jumps 22%, you find out at 8 a.m. with the smoking-gun prompt diff in hand — not when the invoice lands at the end of the month.

Monthly Spend by Agent
Triage
$210
5%
Coordinator
$420
10%
Research
$630
15%
Diagnostics
$2,520
60%
Summary
$420
10%

“$4,200/month on Claude calls. Agent D burns 60% for 12% of workflows. We'll show you.”

§ 07 — PRICING

The same audit trail that satisfies your first hospital becomes the infrastructure that closes your next 50.

Free to start. BAA when you need one. Pricing scales with what compliance asks of you, not with how many calls you make.

HOBBYIST
Free

Try Cyris on a side project.

1,000 agent calls / month
1 workflow
7-day audit retention
PHI detection
Hash-chained audit trail
Community support
Start free →
MOST POPULARTEAM
$99/mo

For startups before their first audit.

100,000 agent calls / month
5 workflows
30-day audit retention
Connection inference
CSV + JSON export
Email support
Start Team →
COMPLIANCE
Custom

For healthcare orgs that need a BAA.

Unlimited agent calls
Unlimited workflows
7-year HIPAA-compliant retention
Business Associate Agreement (BAA)
SSO / SAML
Audit log export + HIPAA reports
Dedicated Slack channel
Priority on roadmap
Contact sales →

Backed by Convoi Ventures