Use case

Every incoming document, checked on arrival.

An agent watches intake folders, classifies what arrives, flags policy violations and drafts redactions, then queues its findings for human review.

Deploy LM-Kit One Build with LM-Kit.NET
Watches intake folders Drafts, never publishes Human approves changes Runs fully local

Findings at audit time are the expensive kind.

Documents arrive all day; checks happen quarterly, on a sample. Everything between those two facts is accepted risk.

  • Sampling blindness. Manual spot checks read a sliver of intake and certify the rest by hope.
  • Late discovery. A violation found at audit time has already been a liability for months.
  • Reviewer fatigue. People skim the fortieth document of the day; software reads it like the first.
  • No standing evidence. Without per-document records, every audit reconstructs history instead of reading it.

Checked when it arrives, not when the audit does.

Compliance found at intake costs a correction; compliance found at audit time costs a finding.

01

Watch

New files in intake locations wake the agent.

02

Classify

Each document gets a type and routes to its checks.

03

Check

Policy rules and PII detection run against the content.

04

Queue

Findings and drafted redactions wait for a person to approve.

Autonomous watching, governed acting.

The agent's autonomy ends exactly where the document changes: at a person.

Always on

Coverage without headcount

Every arrival gets the same checks at any hour; nothing depends on who was in that day.

Draft-only

Proposes, never publishes

Flags and drafted redactions go to a review queue; a change applies only after sign-off.

Recorded

Built for the audit to come

Every check, finding and decision is logged, so the process itself is evidence.

Detection to queue, nothing applied.

The check that runs on every new file: detect, record, queue. Approved removals run through the same redaction call as the review workflow.

IntakeCheck.cs 
using LMKit.Model;
using LMKit.TextAnalysis;

// Runs for every new file the intake watcher reports.
var model = LM.LoadFromModelID("lmkit-tasks:4b-preview");
var pii   = new PiiExtraction(model);

var findings = pii.Extract(documentText);

foreach (var f in findings)
    reviewQueue.Add(file, f.Value, f.Confidence);

// Nothing is redacted until a person approves the queue.

The technology underneath.

The watcher composes documented capabilities you can inspect on their own.

Detect

PII extraction

Typed entities with value, occurrences and confidence, page by page.

How PII detection works

Route

Document classification

Type detection that sends each arrival to the right checks.

How classification works

Draft

Smart redaction

Findings become reviewable regions; approved removals are permanent.

How redaction works

Record

Observability

Every check, finding and decision logged, so the process itself is the audit evidence.

How observability works

Private Document Intelligence

Put a watcher on your intake.

Deploy LM-Kit One The review workflow