Patty | Verifiable enterprise AI

Patty connects models, GPU infrastructure, and agents through one governance layer.

PattyExecution receipt

Evidence sealed

Execution ID: EX-02481
Executed at: 2026-08-26 · 09:42:18 KST
Approved scope: production / configuration.write

Before AI can act inside an enterprise

Example action

Change a production environment configuration

Verification statusEvidence sealed

The approval scope, execution result, and provenance remain in one verifiable record.

Cryptographic sealDARI · COSE-SIGN1

What AI is authorized to do matters more than what it can do

Patty separates proposal, approval, execution, and recordkeeping. A conversation is not authority, and a log is not verifiable evidence.

Propose

AI proposes the next action within the work context and policy boundary.

intent captured

Approve

An authorized person reviews the scope and impact before granting authority.

authority verified

Execute

The control kernel runs only approved work through isolated tools and infrastructure.

policy enforced

Record

Action, rationale, result, and provenance remain as a verifiable receipt.

evidence sealed

One control layer, six execution surfaces

Each product handles different work while managing identity, approval, execution, and evidence through the same operating model.

Technical depth organized as verification paths

Inspect the mechanisms we build across model training, distributed inference, and agent execution.

Open in Technology →

All capabilities · 21 capabilities

We publish only verifiable claims

Measurements include conditions. Protocol claims link to specifications and conformance evidence. Unverified customers, certifications, and awards are excluded.

Agent browser finished-path measurementDebug build · semantic snapshot and governed cycle~14–16 / <25 msDARI protocolOpen specification with conformance testingApache-2.0 / v1.0Deployment profiles including sovereign and closed-networkMirr Enterprise deployment architecture3 profiles

Security and governance are operating architecture

Policy cannot remain in documents. Identity, data boundaries, approval, execution, and audit must work together at runtime.

Enterprise review questionDo you know what your employees are doing with AI right now?Control outcomeAudit & Provenance

Most companies cannot. AI used through personal accounts leaves no record at all. Every human and AI execution needs to leave a receipt. Patty keeps DARI receipts and line-level provenance for every output in an encrypted vault.

Enterprise review questionYou wish employees would stop pasting company data into AI — how do you actually stop them?Control outcomeLLM Gateway & Policy Engine

Training and warnings will not stop it. Policy has to decide on every request before data touches the model or infrastructure. Patty's LLM gateway makes that decision first and never lets unauthorized access through.

Enterprise review questionThink personal data gets filtered automatically when someone types it into an AI?Control outcomeDLP & Privacy

It is not. Public AI services take input as-is. Personal data must be detected and masked before it reaches retrieval or the model. Patty DLP works as pre-emptive blocking, not after-the-fact audit — with the same scrubbing discipline in training pipelines.

Enterprise review questionDoes an AI really follow "don't do that"? Can you trust it in production?Control outcomeAI Action Monitoring

Assume it will not. Oversight beats faith. Patty registers agents as first-class identities, keeps behavioral logs, detects anomalies, and can pause or terminate them immediately.

Enterprise review questionSo where does company data typed into ChatGPT or Claude actually end up?Control outcomeData Residency & Sovereignty

If you cannot check, it can go anywhere. The environment your data lives in has to be decided at the system level. Patty provides Korean data residency and closed-network deployment profiles as standard.

Enterprise review questionIs there any way to pull up the record of an AI accessing your systems?Control outcomeIdentity & Access

No record, no review. AI must not borrow human accounts — it needs its own signed identity. Patty separates human and AI identities with scoped permissions, so "whose action was this?" always has an answer.

Complete control coverage

DARI Protocol Security Model Responsible AI Threat Defense Model Security Regulatory Readiness Compliance & Certifications

Research & papers9/13/2026

From Matching Models to Recruiting Agents: AI Hiring Systems, Evaluation, and Governance

Tracing how AI hiring moved from scoring profile pairs to multi-stage recruiting workflows, and what a systematized review of 40 works says about evaluation evidence and governance.

Research & papers9/13/2026

EXAONE Forecast for Finance: A Linear Foundation Model Designed Apart for Finance

Tracing how EXAONE Finance pairs an attention-free linear backbone with causal convolution, group-aware pooling MLP, and masked context augmentation to cover long many-channel panels and rank first on all three FinVerse tiers.

Research & papers9/13/2026

Tau Tau Bench: A Benchmark That Scores the Whole Job of Building Agents

A new evaluation setting that turns building a customer service agent into the task for coding agents. A close read of its realistic starting conditions, held-out user scoring, low pass rates, and what they mean.

We build systems, not model wrappers

From Korean modeling and GPU serving to agent runtimes and organizational control, Patty builds the full enterprise AI system.

About Patty

Start an enterprise evaluation

Review product adoption, technical validation, security requirements, and sovereign deployment with our team.

Flagships

Agent workspace

Models & Training

Evaluation & Benchmarks

Inference & Infrastructure

Applied AI

Before AI can act inside an enterprise