First Class Modules

AI Insights

Runs permission-aware AI analysis in the background each day to create prioritized, evidence-backed insights and governed next actions from meaningful changes in authorized business records.

Use AI Insights when the system should look for important changes without waiting for a user to ask. The daily background cycle turns permitted record and knowledge context into a small, prioritized insight feed for the people entitled to see and act on it.

What it does

Runs permission-aware AI analysis in the background each day to create prioritized, evidence-backed insights and governed next actions from meaningful changes in authorized business records.

Key capabilities

• Run a durable daily background cycle for configured insight types and source scopes.
• Read only AI-eligible records and evidence through tenant, DataRole, field, location, and CanAiReadRecords controls.
• Create structured insights with source record, type/category, title, summary, recommended action, priority, confidence, severity, evidence, model/profile provenance, and lifecycle dates.
• Use source version/context fingerprints and current-state checks to avoid repeatedly presenting stale or duplicate findings.
• Deliver each insight only to users who can currently access its source record; losing source access removes the insight from their secured feed.
• Let each user view, pin, snooze, hide, dismiss, or advance an insight without changing another user's personal feed state.
• Turn a recommended next step into a governed action that still requires the current user's action permission and configured policy or approval.

Common uses

• Daily account risk, opportunity, renewal, follow-up, and customer-health findings.
• Operational exceptions, overdue work, recurring issues, quality changes, and material KPI movement.
• Proactive company-knowledge, case, service, sales, support, or asset insights with evidence links.

How it connects

The background cycle writes canonical AiInsights linked to their source Records and stores supporting evidence separately. The secured Insights feed presents those findings with per-user state. Recommended work may create a governed insight action, workflow, Inbox item, or GoClaw suggestion, but the insight itself is a finding—not authorization to mutate business data.

Where applicable, its records use the universal RecordId conventions so they can participate in secured relationships, activity history, favorites, dynamic fields, notifications, Inbox attention, and global search without copying the source record.

Security and data boundary

Insight generation applies the AI-read envelope before model context is assembled. Insight visibility is inherited from the source record and rechecked at read time. Evidence, recommendations, and actions cannot reveal or operate on records, fields, locations, or tools outside the current user's permissions; material execution uses action-specific permission and policy/approval.

• The authenticated service derives the SaaS app, tenant account, user, DataRole, and location scope; browser identifiers are never authorization proof.
• The page editor composes React components with validated data bindings. Those bindings call typed runtime APIs, whose application services execute reviewed stored procedures.
• List, search, detail, relation, activity, favorite, and write operations reapply their required server-side permissions.

Add it to an app

1. Choose one decision users should make from a daily finding and define its source modules, eligible states, evidence, freshness window, and materiality threshold.
2. Select the exact users/roles and AI-readable record scope; normal record access alone does not imply AI eligibility.
3. Configure the daily schedule, model/harness profile, priority/confidence/severity rules, expiry, deduplication/current-context policy, and safe failure behavior.
4. Define the user-facing insight lifecycle and any optional next action as a reviewed, permission-keyed contract with the required approval policy.
5. Test no-change days, changed source context, duplicates, expired findings, revoked source access, restricted fields/locations, model failure/retry, and proposed-action denial.