Context and starters

Integrity AI is a read-only workspace assistant. It can summarize and connect the workspace objects you can access, but it cannot create, edit, approve, archive, delete, or otherwise change product data.

Every retrieval uses your current Integrity access. Private Teams, restricted Modules and Folders, unavailable objects, and deleted references remain behind the same permission and display-safety rules as the product. A later access change can also lock an older answer when you reopen its conversation.

Supported workspace context

Integrity AI has bounded readers for:

• workspace overview, Inbox, Reviews, Meeting Minutes, Favorites, Trash, and recent Activity;
• Initiatives, Features, Workspace Views, Teams, Triage, Issues, Cycles, Team Projects, and Team Views;
• Modules, Folders, Artefacts, and Baselines.

Coverage is checked from an independent product-intent catalogue rather than from the tools that happen to exist. The current release gate covers 120 required read-only intent cells across all 19 workspace areas. Thirteen additional cells are deliberately unavailable because the product has no safe or meaningful read contract for them; they are not treated as silently covered.

It can answer exact caller-accessible counts where a narrow count reader exists, including Modules and Initiatives. For Module, Initiative, Feature, and Baseline collections it can also return a bounded browse result with the exact matching total reported separately from the number of displayed rows. Meeting Minutes support the same bounded collection behavior plus title search and scope, status, and Team filters. Selecting one Meeting can return display-safe Notes, attendees, linked context, transcript and correction context, AI summaries, action relationships, and lifecycle state when your current access allows it.

Module browsing is capped at 24 returned Modules. When more Modules match, Integrity AI states that the result is partial, reports how many were omitted, and asks you to narrow the request. It does not automatically page through the workspace. Artefact retrieval remains bounded to exact lookup or keyword and conceptual search. Integrity AI does not expose generic database queries or send the whole workspace to a model.

Prompt starters

Try natural questions such as:

• “How many Modules can I access?”
• “List the names of those Modules.”
• “Find Modules beginning with Safety.”
• “Tell me about the Initiatives in this workspace.”
• “Show active Features related to verification.”
• “Find Baselines with release in the name.”
• “Summarize the Tests Module.”
• “Find Artefacts about emergency shutdown.”
• “What is blocking this Feature?”
• “Show my open Issues and summarize their priorities.”
• “List the Meeting Minutes I can access.”
• “Show the second Meeting’s Notes, attendees, linked context, and actions.”
• “Find archived Meeting Minutes about release readiness.”
• “What changed recently in the workspace?”

Use a follow-up when you want to narrow the result:

• “Only show active items.”
• “Search for thermal.”
• “Open the second result.”
• “Tell me about the second Module.”
• “Summarize its related Issues.”

How to read an answer

Exact counts are computed through access-aware readers and do not require listing every row. Browse and search answers are bounded, use stable ordering, and cite only returned objects that can safely be referenced. Zero results are kept distinct from unavailable, denied, or incomplete coverage.

Follow-ups perform a fresh access-aware read. If your access or an object's lifecycle changes between turns, the later answer reflects the current state instead of reusing stale names or content.

Semantic Artefact search can report partial coverage when some accessible Artefacts do not have a current compatible embedding. In that case, use the suggested keyword search rather than treating the semantic result as complete.

Limitations

• Integrity AI is read-only and does not perform remediation or approvals.
• It does not reveal hidden object names, routes, descriptions, membership, or restricted content.
• It does not provide an unbounded export or a generic cross-domain query.
• Some operations and transient/provider-owned data are deliberately unavailable; Integrity AI offers the nearest safe read or navigation path instead of inventing an answer.
• A cited historical answer can become unavailable after access or lifecycle changes.
• Answers should be treated as a navigation and synthesis aid; the cited Integrity objects remain the durable source of truth.