FormKiQ KnowledgeBase: AI Retrieval in Your AWS Account

Natural-language retrieval across governed document collections, powered by Amazon Bedrock

Published Updated

FormKiQ KnowledgeBase. Answers connected to source records. A book-like source collection connects to quoted passages and an answer card beside a magnifier.

This article began in April 2024 as a roadmap concept for an AI-powered knowledge base. FormKiQ KnowledgeBase is now available as an add-on module for Advanced and Enterprise, using Amazon Bedrock to support natural-language questions over selected document collections.

The enduring design question is how to keep answers connected to approved, current, accessible source material. Adding a chat interface does not settle collection scope, permissions, review, or source retirement.

Product information reviewed: November 12, 2026. Current scope is described on the KnowledgeBase module page. The original concept used OpenAI’s API; the current module uses Amazon Bedrock.

Use retrieval to make answers inspectable

Retrieval-augmented generation combines selected source passages with a model-generated answer. Source references help a person inspect the supporting records. They do not guarantee that the answer is complete, current, or correct.

Choose a bounded use case, such as finding the approved policy for a case or preparing research from a controlled collection. Define when the system should decline to answer, show uncertainty, or send the result for review before it drives a consequential action.

Configure the knowledge lifecycle

A document can remain a valid historical record while becoming unsuitable for current answers. Manage eligibility for retrieval separately from repository retention. Define the integration steps and verify the actual delay between a source change and its effect on answers.

These are lifecycle requirements to configure and test for the chosen retrieval setup.
ChangeRequired behaviour
New document versionIngest the approved replacement, retire superseded retrieval entries where required, and keep citations tied to the exact source version.
Periodic reviewAssign an owner and due date, notify reviewers, and apply the approved eligibility rule if review is overdue.
Expiry or withdrawalRemove the source from the applicable retrieval collection or filters, confirm the removal, and preserve the repository record when retention requires it.
Permission changeUpdate applicable access filters or collection membership and verify that restricted passages, citations, and cached answers are no longer disclosed.

Automatic replacement, review scheduling, expiry handling, and permission propagation depend on the deployed configuration and integrations. A repository update alone is not evidence that every retrieval index or answer cache has changed.

Keep retrieval inside the user’s authority

Define who can query each collection and which source records may contribute to an answer. Enforce the intended access boundary before disclosing passages to the user or model. Test with users who have different document permissions, including after a role change or revoked grant.

For an external knowledge service, select content approved for that audience. A document’s presence in an internal collection does not make it appropriate for a public website or customer portal.

Choose model geography and evaluation criteria

In customer-managed deployments, Bedrock is accessed through the customer’s AWS account. The selected model, endpoint, and inference profile determine processing geography. Review Bedrock’s regional routing behavior; document storage in one region does not establish where every inference request runs.

Evaluate representative questions for retrieval relevance, supported claims, citation accuracy, obsolete sources, denied access, and cases where no answer should be produced. Repeat the evaluation after changing the model, prompt, collection, or retrieval configuration.

Example: a policy is replaced while a review is open

  1. Approve the new source

    The policy owner publishes version 3 and records when it becomes effective. Version 2 remains in the repository as a historical record.

  2. Update the retrieval collection

    The configured process ingests version 3 and removes or excludes version 2 from current-policy answers. Failed refresh work enters a visible retry or review path.

  3. Verify the answer boundary

    Test that current questions cite version 3, historical questions follow the approved scope, and unauthorised users receive neither policy text nor revealing citations.

Scope the module around the application

KnowledgeBase can sit alongside AI Processing and Analysis, Enhanced Search, document reviews, and business integrations. Agree collection scope, refresh and removal behaviour, permissions, model access, source attribution, monitoring, and review responsibility for the implementation.

Explore KnowledgeBase and the search architecture guide, or talk to FormKiQ about the questions and document collection your application needs.

Practical checklist

  • Define collection eligibility and permitted audiences.
  • Test new versions, withdrawals, and permission propagation.
  • Check citations and questions with no supported answer.
  • Review the selected model, routing, and refresh monitoring.

Make the workflow and scope explicit

Keep answers tied to approved, current sources and verify the removal and access boundaries as content changes.

Plan the next step

Discuss the workflow with FormKiQ using your documents, permissions, integrations, and acceptance criteria.

      Start with FormKiQ Core

      The open-source foundation — API-first, deployable into your own AWS account, and free to use. Right for architecture validation and early implementation.

      Get Started Free

      Deploy FormKiQ Essentials or Advanced

      Production-ready editions for departments and complex workflows. Start with a Proof-of-Value deployment or go straight to production.

      Explore Options

      Plan an Enterprise Rollout

      For governance-heavy environments with residency, sovereignty, assurance, and multi-jurisdiction requirements. Talk to us about the right deployment model.

      Book a Call