โšก DORA enforcement is live Now in Private Beta

Your AI is ready.
Your auditors aren't.

Phronesis captures every decision your AI agent makes โ€” so when the regulator calls, you have the forensic evidence ready in seconds, not days.

// high concept "The Black Box Flight Recorder for Agentic AI"
Built for ๐Ÿฆ EU DORA ยท ๐Ÿ‡บ๐Ÿ‡ธ SR 26-2 ยท ๐Ÿ‡ฎ๐Ÿ‡ณ RBI Master Directions ยท ๐Ÿ”’ OTel-Native ยท ๐Ÿ›ก๏ธ Air-Gapped Ready
DORA Incident Window ACTIVE
Time remaining to file initial report
3:47:22
Article 5(4) โ€” Initial notification deadline
14:03:18 โš  Agent hallucination detected โ€” FX routing decision
14:03:19 Searching raw OTel logs... 847,293 records
Phronesis Flight Recorder PROTECTED
โœ“
Evidence Pack Ready
Generated in 3.2 seconds
๐ŸŽฌ Forensic Replay โ€” 47 agent steps traced
๐Ÿ“‹ DORA Initial Report โ€” Structured & ready
๐Ÿ”’ Ledger Hash โ€” Cryptographically verified

Three crises. One frozen AI project.

Banks are trapped between engineering teams pushing forward and risk teams holding them back. The result is millions in stranded investment.

$5M+
average AI initiative frozen in sandbox

"Pilot Purgatory"

The Head of AI Platform built working agentic workflows. Then the Risk Committee vetoed production deployment. The AI sits unused โ€” costing millions in missed savings every quarter.

"We can't deploy it โ€” the auditors need to understand every decision the AI makes, and we have no way to show them." โ€” Head of AI Platform, Tier 1.5 Bank
4 hours
DORA initial incident reporting window

The DORA Clock

Under DORA Article 5, when an AI agent hallucinates an unauthorized action it's an ICT incident. Banks have 4 hours to file an initial report โ€” but engineers need 3 days to reconstruct what happened.

"If this AI makes a mistake and I can't explain why to the ESA within the window, that's my personal liability." โ€” Chief Risk Officer, ECB-supervised institution
โšก Up to โ‚ฌ1M personal fine under DORA Article 5(4)
0
existing tools designed for this problem

The Governance Vacuum

Datadog tells you the system crashed. ServiceNow logs traditional model metadata. Neither can map the business context of why a probabilistic AI agent made a specific, multi-step decision.

Datadog / Splunk Shows latency, not decisions
ServiceNow GRC Static inventory, not runtime traces
Manual + Word Doc Days of work, legally indefensible
โœˆ
The Black Box Metaphor

A flight recorder doesn't prevent crashes โ€” it survives them and provides the exact, undeniable chronology of what the pilot did. Banks need the same thing for their AI agents.

Your AI's Black Box.
Finally readable.

Phronesis sits passively alongside your AI agents, ingesting every decision as it happens and transforming raw telemetry into structured, legally defensible evidence.

// Unique Value Proposition
"Safely unblock your Agentic AI deployments. We translate unreadable AI telemetry into structured, chronological audit trails โ€” giving you the forensic evidence required to confidently file a DORA or RBI incident report in under 4 hours."
3s Evidence pack generation
4h DORA/RBI window โ€” covered
0 Code changes to your AI
01
๐ŸŽฌ

Forensic Reasoning Replay

Ingests standard OpenTelemetry traces and automatically translates them into a human-readable, chronological decision tree โ€” step by step, in plain English.

14:01:03 Agent started โ€” FX Routing Workflow #4821
14:01:04 Read: sanctions_list.pdf (cached, stale)
14:01:05 โš  Decision: Approved transfer based on stale data Confidence: Moderate
SR 26-2 โ€” Effective Challenge Evidence
02
โšก

1-Click DORA Evidence Pack

When an incident occurs, your CRO opens the dashboard, clicks one button, and receives a perfectly structured, regulator-ready PDF โ€” not a frantic Slack thread.

โœ“ Chronological incident timeline
โœ“ Root cause analysis โ€” automated
โœ“ DORA Article 19 required headers
โœ“ Cryptographic integrity hash
DORA Art. 19 / RBI Cyber Resilience
03
๐Ÿ”’

Context-Aware Append-Only Ledger

A high-velocity event-streaming backend that chains scattered agent traces into a single, tamper-evident system of record. The logs are immutable โ€” just like a real flight recorder.

Event Depot
โ†’
Topology
โ†’
PailState
Hash-chained ยท Append-only ยท Tamper-evident
SR 26-2 โ€” System of Record
04
๐ŸŒ

Deploy Anywhere: Hosted or Air-Gapped

Your data, your rules. Choose a fully managed cloud for convenience, or deploy 100% on-premise behind your firewall. In our air-gapped mode, your sensitive telemetry never leaves your infrastructure.

Cloud Fully managed hosted option for rapid deployment
In-House 100% air-gapped on-premise deployment
24x7 Support We are always here to assist you, because you never know when you'll need us
DPDPA & GDPR Data Sovereignty

From chaos to clarity
in four steps.

Phronesis uses a "passive interceptor" model. Your AI agents don't change. We sit alongside them and do the translation work.

01

Connect

Point your existing OpenTelemetry exporter to our ingestion endpoint. One config change. No SDK to install. No code changes to your AI agents.

otel-config.yaml
exporters:
  otlp:
    endpoint: ingest.phronesis.ai
    headers:
      x-phronesis-key: "your-key"
02

Capture

Every agent decision undergoes Double-Gated Semantic Validation before being locked into our append-only, hash-chained ledger in real time. The machine complexity is invisible to you.

<50ms Ingestion latency
100% Trace capture rate
โˆž Retention (mirrored)
03

Monitor

Your CRO's board-level dashboard shows continuous oversight: agent workflows governed, anomaly counts, and compliance posture โ€” all in plain English.

14 Agent Workflows โ€” Governed
0 Anomalous Incidents โ€” Last 30 days
DORA & RBI Posture โ€” Ready
04

Report

When an incident occurs, click "Generate Evidence Pack." Receive a perfectly structured, legally defensible incident report mapped to DORA or RBI guidelines in seconds. Not days.

๐Ÿ“‹
RBI_Cyber_Incident_Report_20250609.pdf
Generated in 3.2s ยท Hash: a8f2c1...
Ready
โš™
๐Ÿ“ก
๐Ÿ”’
๐Ÿ“‹
// Phronesis Forensic Replay โ— LIVE
14:01:03.241 AGENT_START FX Routing Agent โ€” initiated workflow #4821
14:01:03.489 TOOL_CALL ReadDocument("sanctions_list_v2.pdf") โ†’ 847KB
14:01:04.112 TOOL_CALL CheckBalance(acct: "FX-EU-0029") โ†’ โ‚ฌ2.4M
14:01:04.891 LLM_INFERENCE Decision: "Counterparty cleared. Proceed with transfer."
14:01:05.033 ANOMALY Sanctions list was cached (stale 4h 12m) โ€” not live API

Built for institutions where
the stakes are real.

We're not selling to every bank. We're solving a specific problem for institutions caught between AI innovation and regulatory accountability.

โญ Primary Market Years 1โ€“3
๐Ÿฆ

Tier 1.5 & Tier 2 Banks

$50B โ€“ $500B in assets

EU โ€” DORA / ECB US โ€” SR 26-2 / Fed UK โ€” PRA CA โ€” OSFI

Face the same punitive regulatory pressure as Mega-Banks, but lack the 10,000-person engineering army to build governance infrastructure in-house. Structurally forced to buy.

~314 Total addressable institutions
$3M+ Target ARR by Year 3
Decision-Making Unit
๐Ÿ”ฅ Champion Head of AI Platform / VP of Risk Engineering Their AI is built and blocked
๐Ÿ’ฐ Economic Buyer VP of Enterprise Architecture Holds infrastructure budget
๐Ÿ›‘ Veto Chief Risk Officer (CRO) Personal DORA liability โ€” the real blocker
Secondary Market Year 2+
๐Ÿš€

Regulated FinTechs

The "Speedboats"

Cloud-Native API-First OTel-Native

Modern, heavily funded FinTechs, payments processors, and digital-only banks. They buy with engineering-driven POCs. Fast decisions, minimal bureaucracy.

Digital banks Payments processors Embedded finance
Decision-Making Unit
๐Ÿ’ฐ Buyer CTO / VP Engineering Signs off after successful POC
๐Ÿ”ฅ Champion Lead ML Engineer / Head of Data Platform Runs the integration POC
POC-to-contract: 3โ€“6 months vs. 9โ€“18 months for banks
๐Ÿšซ
Explicitly Out of Scope: Quant Hedge Funds

Extreme data sovereignty paranoia, 100% air-gapped deployments, and no off-the-shelf SaaS adoption. A distraction for a seed-stage startup โ€” revisit in Year 4+.

Land and expand.
No surprises.

Priced per agentic workflow governed โ€” so your cost scales with the value you're extracting. Start with a staging pilot, expand as your AI goes live.

// Value metric: Per Agentic Workflow Governed Aligns cost to operational ROI โ€” not arbitrary seat counts
The Land

Design Partner Pilot

$25,000 flat fee
60-day engagement on staging data
  • โœ“ OpenTelemetry ingestion setup
  • โœ“ Forensic Reasoning Replay (1 workflow)
  • โœ“ Sample DORA Evidence Pack generation
  • โœ“ Architecture review session
  • โœ“ Staging data only โ€” no SOC-2 required
Designed to stay within VP discretionary signing authority
Start Pilot Discussion
The Shield

Enterprise Compliance Shield

$350,000+ / year ARR
Unlimited agents ยท Board-level dashboards ยท SLA
  • โœ“ Everything in Core License
  • โœ“ Unlimited agentic workflows governed
  • โœ“ Centralized Board-Level Risk Dashboard
  • โœ“ Multi-business-line coverage
  • โœ“ Dedicated SLA guarantees (99.9% uptime)
  • โœ“ Custom on-premise / VPC deployment
  • โœ“ SR 26-2 Model Risk validation support
When you govern 50+ agents, we become core infrastructure
Contact for Enterprise
// Target LTV: >$360,000 per institution over 3 years ยท LTV:CAC ratio target >3x

AI should be accountable.
Not a black box that gets banks fined.

The financial system is undergoing a once-in-a-generation transformation. Agentic AI has the power to save billions in operational costs and eliminate human error in critical workflows. But that promise collapses the moment a regulator asks "why did the AI do that?" and no one can answer.

We built Phronesis because we believe the gap is not a technical one โ€” it's an infrastructure one. The tools to safely deploy AI at enterprise scale already exist. What was missing was the accountability layer that makes those deployments legally defensible.

Just like the aviation industry didn't abandon flight when crashes happened โ€” they built better recorders โ€” the financial industry doesn't have to choose between AI innovation and regulatory compliance. Phronesis is that recorder.

M
Madan
Founder ยท QuirkyLabs.ai
๐Ÿ“ Bengaluru, India
madan@quirkylabs.ai
// Technical Architecture
โš™
High-Velocity Backend
Event-stream architecture, purpose-built for ledger semantics
๐Ÿ“ก
OpenTelemetry Native
Ingests CNCF standard traces โ€” no custom SDK, zero agent changes
๐Ÿ”’
Cryptographic Integrity
Hash-chained append-only ledger โ€” tamper-evident by design
๐ŸŒ
DORA/SR 26-2 Alignment
Structured output maps directly to regulatory reporting requirements
Stage Private Beta
SOC-2 In Progress
Looking for Design Partners
โšก Private Beta โ€” Limited Design Partners

Get your AI out of Purgatory.

Join a small cohort of banks and FinTechs who are first to deploy Phronesis. Design Partners get hands-on onboarding, early pricing, and direct influence over the product roadmap.

โœ‰
Reach out directly
madan@quirkylabs.ai
Response within 24 hours
// Email template โ€” copy and send
Design Partner Benefits
๐ŸŽฏ
Hands-on onboarding
Direct setup support with Madan
๐Ÿ’ฐ
Founder pricing
Locked-in early access rate ($25k pilot)
๐Ÿ—บ
Roadmap influence
Your use case shapes the product
๐Ÿ“‹
Case study co-authorship
Joint thought leadership on DORA compliance
โœ“ Staging data only โ€” no production exposure in pilot
โœ“ SOC-2 in progress โ€” architecture review available
โœ“ Data sovereignty guaranteed โ€” your data stays yours
โœ“ No lock-in โ€” full data export at any time