Knowledge Hub: Albus Chat

Last updated: May 4, 2026

How to use Knowledge Hub to improve Albus's accuracy for role mining, RBAC analysis, and general identity work.

Key Use Cases

  1. Role mining and RBAC analysis — Define which attributes matter (department, team, org level), what thresholds to use, and what apps or entitlements to include or exclude (e.g., "always ignore location-based groups in access policies").

  2. Attribute and org hierarchy context — Help Albus understand how your HR attributes relate to each other (e.g., "Org Level 3 = VP-level managers", "Supervisory Orgs map to cost centers").

  3. App-specific access nuances — E.g., "Salesforce has multiple instances — specify which one", "GitHub teams = permission groups."

  4. Contractor and employment status rules — Define how your organization identifies contractors, external users, and terminated employees so Albus can filter correctly (e.g., "employment status = Active means the employee is not terminated, but 'active' could also refer to recent login activity — ask for clarification").

  5. Birthright and policy definitions — Help Albus understand what constitutes birthright access, threshold rules, and which apps have already been designated as birthright.

How to Scope Context

  • Global — Org-wide rules like contractor identification, employment status definitions, birthright thresholds.

  • App-specific — Per-app nuances (e.g., "Salesforce has both Production and Sandbox — only analyze Production").

Tip: You can upload existing RBAC policies (from other tools or systems) as markdown files in Knowledge Hub, then ask Albus to build matrices using those policies. This is especially useful for migrations from existing IAM tools.

Best Practices for Managing Agent Context

Use these tips to get the most out of Knowledge Hub with Albus Chat.

Use trigger words

Use trigger words like "add to my memory" or "remember..." when teaching Albus new rules.

Include the scope of the memory

Be specific about whether a rule is general or scoped to a particular app or matrix:

  • General rule: "In the future, remember that generally..." — for top-level instructions that apply broadly.

  • Matrix-specific: "Remember that in my app matrix [name]..." — makes sure the rule is tied to the right context.

  • Common rules: "Try to apply [rule] as a common rule if possible" — helps Albus generalize.

  • App-specific: "Add a permission rule to my memory for app (Office 365) that permissions with [criteria]..." — always try to specify the app name.

To edit existing context

Say: "Edit my memory, replace... with..."

To show current context

Say: "Show me my current memories" or "Show me what you are currently remembering for me"

To delete context

Say: "Remove [rule] from my memory" or "Forget [rule] and remove it from my memory"

Avoid using the word "prompt"

Albus and other Lumos agents have protection guidelines that may flag the word "prompt" as a potential injection attempt. Use terms like "memory", "context", or "rule" instead.

Quality Check

After adding context:

  • [ ] Spot check permissions matrices across a few apps — do they reflect your new rules?

  • [ ] Verify Albus correctly excludes/includes apps and entitlements per your context

  • [ ] Test attribute hierarchy questions ("show me how my attributes relate")

  • [ ] Check that birthright thresholds are being applied correctly

  • [ ] Verify context doesn't cause unexpected results — Albus may occasionally misinterpret ambiguous rules