Arcova Docs

Iris AI

Iris

Iris is an AI assistant that answers your team's questions using your own policies, post orders, and procedures, so officers and supervisors can find the right answer without paging someone. Iris answers are AI-generated suggestions, not authoritative — verify high-stakes decisions with a supervisor.

Last updated February 2026

Overview

When to Use This

  • When officers need to look up a post order in the field instead of paging a supervisor.
  • When supervisors want to publish company policies and SOPs in one searchable place.
  • When new hires need quick answers about site-specific procedures during training.
  • When you want to see which topics your team asks about most so you can improve documentation.

Model

Core Concepts

Iris uses the following objects and relationships as its operating model.

Knowledge Source

An uploaded document Iris reads from — like a post order, SOP, or employee handbook.

Persisted Data
Document file, who can see it, version history.
State Changes
Draft, published, archived.

Question

Something a user types into Iris chat.

Persisted Data
The question text, who asked it, when it was asked.
State Changes
Submitted, answered.

Answer

What Iris returns, including citations to the source paragraph and a confidence indicator.

Persisted Data
Answer text, source links, confidence level (how relevant the source material was).
State Changes
Generated, rated, escalated.

Feedback

A thumbs up or thumbs down a user gives an answer.

Persisted Data
Rating, optional reason, who left it.
State Changes
Open, reviewed, resolved.

Role Requirements

  • Anyone signed in to your company account can ask Iris a question.
  • Knowledge owners (usually a supervisor or training lead) upload and approve documents Iris is allowed to read.
  • Admins control which documents are published and who can see them.

Side Effects

  • A newly published document becomes searchable a few minutes after upload, once Iris finishes reading it.
  • A thumbs-down rating creates a review task for the knowledge owner so they can fix or replace the source.
  • Every question and answer is saved so you can review what your team is asking about.

Notifications

  • Knowledge owners get a weekly digest of questions that received negative feedback.
  • Admins are alerted if a document fails to upload or process.
  • Admins are notified when usage approaches their plan limits.

Automation Hooks

  • New documents are processed in the background; you can keep working while Iris reads them.
  • Low-rated answers automatically rise to the top of the knowledge owner's review list.
  • Admins can set a usage cap to prevent runaway questions on a busy day.

Flow

How It Works

  1. 1

    Upload your post orders and policies as PDF or DOCX.

  2. 2

    Iris reads them and builds a searchable index.

  3. 3

    Your team asks a question in chat.

  4. 4

    Iris returns an answer with links to the source paragraph and a confidence indicator.

  5. 5

    Users can rate the answer; low-rated answers create a review task for your knowledge owner.

Walkthrough

Example

Scenario

A new officer at the Westfield site opens Iris on his phone and asks, "What is the dress code for the Westfield site?"

Walkthrough

  1. 1

    Iris searches the published Westfield post orders and finds the uniform section.

  2. 2

    It returns a short answer summarizing the requirement and a link to the exact paragraph in the source document.

  3. 3

    The officer taps the link, confirms the source, and gives the answer a thumbs up.

  4. 4

    The supervisor sees in the dashboard that the dress-code question is being asked often and decides to add a poster at the guard booth.

Outcome

The officer gets a verified answer in seconds and the supervisor learns where documentation could be reinforced.

Watch Out

Edge Cases / Gotchas

  • Unpublished and archived sources are filtered from search by default. Audit Iris responses regularly; do not rely on archival as a security control.
  • Iris only sees documents your company has uploaded — it does not read the public internet.
  • After uploading a large document, give Iris a few minutes to index it before expecting answers from it.
  • The confidence indicator reflects how closely Iris matched your source material — not whether the answer is factually correct. Always verify high-stakes decisions.
  • Iris can mask sensitive values (like SSNs or birth dates) in answers if you turn on PII redaction.