TaskRox Help & User Manual

AI Studio is the dedicated AI workspace in TaskRox. It is separate from normal module pages and supports both fast Q&A and reusable outputs.

Tabs: - Chat — free-form project-aware conversation - Chat Assist — guided prompt categories (including Help & how-to) that start a well-formed question for you - Reports — structured generated outputs - Saved — retained drafts and summaries

In Chat, ask questions in either portfolio scope or a selected project scope.

When scoped to a project, TaskRox AI builds a live context snapshot before response generation (for example task, RFI, NCR, HSE, budget, and milestone signals).

Streaming responses support: - Stop while generating - Regenerate the latest assistant response without duplicating the user prompt - Partial-response preservation if a stream is interrupted

For safety, a conversation's project scope is immutable after creation (you cannot swap project scope mid-thread).

TaskRox AI also answers product how-to questions — for example *"How do I open a document in Google Sheets?"* or *"How do I set up a roster rotation?"*.

When it detects a how-to question, it grounds the answer in this Help manual and links the relevant help page so you can read the full documentation. Where a feature is controlled by a workspace setting (for example the Google Sheets connector), the answer reflects whether that feature is currently enabled for your workspace.

TaskRox AI deliberately does not provide general tech support for unrelated third-party tools — it answers the TaskRox-relevant part of a question and points you to the right help page.

Assistant responses include provenance metadata for decision-grade prompts: - scope indicator (Project or All projects) - Data as of timestamp - queried module/domain summary - confidence label (High / Medium / Low) - assumption list when inference was required

Where supported, responses also show contextual quick actions (for example open a linked module/register). Quick actions only appear if your role has permission for that module.

Use Reports to generate structured outputs and Saved to retain reusable drafts/summaries.

Daily message limits apply to non-super users. The UI shows remaining allowance and reset time. Reset policy is UTC, with local display in the client.

If upstream AI service issues occur, user-safe error messages are shown (for example timeout, temporary provider unavailability, or retryable rate-limit errors). AI usage health is visible to super users in Administration → AI.

Compliance Navigator is a dedicated AI Studio lane that answers "what applies to this job?" — the permits, approvals, utility checks, safety controls, evidence, and TaskRox actions a piece of work needs, grounded in reviewed compliance sources with citations. It is compliance support and workflow assistance, not legal advice. The lane is enabled per workspace by an admin; when it is not enabled it does not appear.

Use the context builder (project, jurisdiction, work type, focus areas) so answers are scoped. If a project has no usable location, the lane says jurisdiction-specific requirements can't be determined and links you to complete the project location — it never guesses.

Every answer cites its sources with their publisher, dates, and review status (reviewed / unreviewed-verify / stale / superseded). Unreviewed sources are flagged and never give definitive guidance.

Action packs: turn an answer into a reviewable pack, then export selected items into real TaskRox records — tasks, HSE actions, document placeholders, QA hold points, training requirements, daily-report prompts, and correspondence drafts. Each item is gated by the *target* module's own permission, and only items grounded in a cited source can be exported; the review screen shows what you can and can't create.

Incident triage: for a near miss or incident, the lane leads with immediate-safety steps, a fact-gathering checklist, and a notification decision path — it never concludes whether an event is notifiable. "Report as HSE incident" creates a real incident marked under review (notifiability is not pre-judged) plus an investigation checklist for the accountable person to assess.