AI in Feith · Custom prompts

Custom AI prompts, run where your records live.

Prompt Runner takes a plain-English instruction and runs it across a whole record set, extracting metadata, classifying content, reasoning over policy, and triaging what comes in. The model runs inside your security boundary, so the records never leave to be read.

Request a walkthrough All six capabilities ›

1 Prompt, thousands of recordsPlain English, no code; it behaves like a classifier you write yourself

0 Records copied outThe model runs inside your boundary; bring your own

Any Document layoutReads whole records, not fixed form fields

1 Taxonomy, normalized toEvery value lands in your fields: dates, currencies, names

What it does

One prompt. Thousands of records.

Through a low-code prompt builder, your team writes plain-English instructions that behave like modular classifiers and extractors. Prompt Runner reads complete documents of any layout and normalizes every value it returns to your taxonomy, writing the result straight back as record metadata.

Reads complete documents, not fixed form positions, so it works on contracts, invoices, and forms whose layout varies.
Normalizes what it extracts: dates land in one format, currencies parse, names match your taxonomy.
Runs policy logic from a paragraph of plain English, no rules to code.
Triages what comes in by urgency, topic, and target team, then routes it to a queue, a dashboard, or a reviewer.
Works retroactively: surfaces metadata that was locked inside records captured years ago.

Prompt Runner runs on its own, on a schedule, or as one optional AI step inside a Workflow iQ workflow. It's the AI step inside that workflow.

Worked example

One prompt

“Extract renewal date, ceiling value, FAR clauses, and party names from each contract.”

Prompt Runner runs in your boundary

Reads each record in full

Whole documents, any layout, and normalizes every value to your taxonomy.

Structured output

renewal: 2026-09-30 ceiling: $4.2M FAR 52.217-9 party: ACME LLC

Written back as record metadata; renewal alerts fire automatically.

normalized to your taxonomy audit entry written

How it works

The model comes to where the records live.

Prompt Runner runs your prompt against records already in the vault, with a language model you've authorized running inside the same boundary. Nothing is exported to be read; the structured result is written straight back.

Records in your vault

Prompt Runner

Your prompt runs on each record

Your model, inside your boundary (bring your own)

Normalized fields written back

Why it's different

Your model, your boundary

Bring your own LLM. Prompt Runner runs it where your records already live; nothing is copied out to be read.

Reads whole records

It works on documents whose layout varies, and normalizes every value to your taxonomy.

Auditable by design

Every run writes to the audit trail, and you can route low-confidence results to a person before anything is committed.

How you run it

One step, three ways to run it.

Prompt Runner is the AI step: run it by itself or drop it into a larger Workflow iQ process.

On demand

Point Prompt Runner at a record set, write a prompt, and run it once to backfill the metadata you wish you'd captured.

On a schedule

Run the same prompt automatically as new records arrive, so fresh records are enriched the moment they land.

As one step in a workflow

Drop Prompt Runner in as the AI step inside a Workflow iQ workflow. It reads, classifies, or extracts, then hands the result to the next step.

See Workflow iQ ›

In practice

Metadata extraction, use case by use case.

The same engine, pointed at a different record set with a different prompt. A few of the jobs agencies run it on:

Contract analysis

Renewal date, ceiling value, FAR clauses, party names

Renewal alerts fire with almost no clerical effort.

Service-desk triage

Issue type, urgency, location

Tickets route to the right queue; SLA compliance climbs.

Safety incident reports

Risk indicators, personnel involved, measures taken

Speeds compliance reporting and root-cause analysis.

Historical archives

Meaning read from handwritten field notes

Researchers get searchable access far sooner.

Asset capture

Serial numbers read from equipment photos

The asset database updates in near real time.

NARA attributes at capture

Creator, creation date, identifier, record schedule

Zero-click NARA compliance the moment a record lands.

The rest of the set

Five more capabilities, one boundary.

Every capability shares the same permission model, the same audit trail, and your choice of large language model.

Back to AI in Feith ›

Get started

Turn the metadata you wish you had into metadata you have.

Bring a record set and a plain-English prompt. We'll run Prompt Runner across it live and show the normalized fields written back to your records, inside your boundary.