Every Bluework feature, replayed live.

Ask for next week’s roster. Review why every name was picked.

Aegis drafts the week with a deterministic engine — the AI never picks people silently. Every candidate is scored on cost, fatigue, reliability, task performance, hazards and stated preferences, and every pick carries its reasons. You review, you apply — and applying creates the assignments without ever publishing.

• Deterministic scoring — same roster, same inputs, same draft, every time
• A one-line reason trail on every pick, with swap alternatives ranked
• Apply creates draft assignments only — publishing stays with you

The award says no — so Bluework says no

Pick an award rule pack and every roster change is checked against it: daily caps, rest floors, minimum engagements, qualification gating. Set each rule to advise — or to block the change outright, citing the exact clause.

• Rule packs for Retail, Hospitality and Building & Construction awards
• Off / Advisory (warn) / Extreme (block) — per module
• Every block cites its clause — “the actual clause, not a vibe”

Shifts that fill themselves

Workers accept or decline their shifts, claim open ones, and trade through a swap marketplace — every claim re-checked for qualifications and overlaps, every swap approved by a manager before it moves.

• Open shifts are qualification-gated — unqualified claims bounce
• Swap pick-ups wait for manager approval, then reassign atomically
• Accept / decline states visible to the manager in real time

When someone says no, the next best person gets the offer

A declined shift no longer lands back on your desk. Bluework ranks everyone who can work it — qualified, available, no clash, lowest fatigue, the client’s preferred staff first — and offers it to the top five by push and SMS. The first to accept gets the shift; the other offers close on their own.

• Unqualified, unavailable, overlapping or award-blocked workers are ruled out first
• Client-preferred staff are offered first; client-excluded staff never are
• Unanswered offers expire in 12 hours; recurring series roll eight weeks ahead

Leave balances that accrue themselves

Annual and personal leave accrue pro-rata from approved timesheet hours — the NES floor, computed, not guessed. Workers request in two taps; managers approve with the balance right in front of them.

• NES-accurate accrual from ordinary hours worked (4 weeks/yr, pro-rata)
• Casual variant: no accrual, 25% loading applies instead
• Approved leave blocks the roster and the fatigue scorer sees it

Every minute priced against the award

Geofenced clock-ins land against the roster and flag themselves — over hours, short breaks, off-site punches. Approve the week and the award engine splits every shift across penalty rates, with super, on the spot.

• Variance chips per punch — “+0.4h vs roster”, “break 15m short”, “off-site”
• Penalty rates applied minute-by-minute, single highest rate per hour
• Approving locks the week and prices it — one CSV to payroll

Forgot to clock out? Fix it with a paper trail.

Workers don’t beg the manager to “just change it” — they send a change request. The manager sees the exact old → new diff with the reason, approves or rejects, and the original times stay in the audit history forever.

• Every edit is a request with a reason — no silent rewrites
• Approving a change re-opens the entry for re-approval, never auto-locks
• Original punch times preserved in the entry’s audit trail

New pay guide? Upload the PDF.

When Fair Work moves the rates, drop the pay guide in. Bluework parses it, scores its own confidence, and shows you an old → new diff. Nothing applies without your review — and one click rolls it all back.

• Confidence-scored extraction — low-confidence rows are never pre-ticked
• You approve the diff; the live pay engine updates instantly
• Full rollback restores every prior rate, field for field
• Rates and award packs can differ per shop — same shift, right rate, every state

Detection you could defend in a tribunal

With the worker’s consent, on-shift audio is transcribed and screened against a protected lexicon — 77 phrases drawn from Australian court and tribunal decisions on workplace harm. Every hit keeps its context, its speaker, and its case citation, and a human approves before anything becomes a hazard.

• Consent-gated three times over: opt-in, clocked-in, PII redacted before analysis
• Each phrase cites the decision it came from — admins can’t water the list down
• AI verifies in context (“banter” gets rejected); a human makes the call

“Low job control” is a hazard. Bluework treats it like one.

Low job control is a named psychosocial hazard in the SWA Code of Practice. Bluework catches the situational signals workers actually say, aggregates repeated exposure across a crew, and points the fix at the work design — not the people who spoke.

• Situational signals, not accusations — “situations, not people”
• Repeated exposure aggregates: one remark is noise, a pattern is a hazard
• Recommended controls come from the Code of Practice, ready to assign

No consent, no monitoring. Provably.

Monitoring starts with a signed agreement, not a buried setting. Every worker’s consent lives in a register with its evidence — and where state law requires notice (NSW/ACT), Bluework serves the 14-day statutory period automatically before anything records.

• Consent recorded with evidence; workers without it are simply never recorded
• NSW/ACT: 14-day workplace-surveillance notice counts down, then auto-activates
• APP 11 erasure honoured — the consent evidence itself is retained

Detection is half the duty. Here’s the other half.

Every confirmed hazard opens a guided response plan — the seven steps a WHS regulator expects, in order, with owners and dates. Sign-off seals the whole response into the tamper-evident record.

• Consultation step is mandatory — a WHS Act ss.47–49 duty, tracked
• Controls follow the hierarchy, cited to the Code of Practice
• Sign-off writes the final hash — the response becomes evidence

77 phrases. Every one survived a courtroom.

The detection baseline isn’t a vibes list — each protected phrase appeared in the evidence of an Australian court or tribunal decision, and admins can’t edit or delete them. Your own additions sit alongside, clearly marked.

• Protected rows are locked — the defensible baseline can’t be watered down
• Every phrase carries its case citation, visible in the admin
• Add custom categories — your own phrase sets go live for the next scan

Hold to talk. Release to listen.

A browser-based radio: pick a channel, hold the big button, and your voice streams live to everyone on it — with earphone-button push-to-talk for gloved hands. Channels scope to everyone, a shop, or a hand-picked crew.

• Half-duplex live audio — one tap to join, instant to transmit
• Live speaker attribution: “🔊 Alice is talking…” on every listener’s screen
• Wired or Bluetooth earphone: short-press starts and stops transmit

That laminated binder page? It’s a Document now.

Drag in a scanned PDF or a photo of the old procedure, hit the magic wand, and Bluework OCRs it into a real, editable Document — reviewed in a preview before anyone relies on it.

• PDF or image in, structured Document out
• Extracted text lands in an auto-created section, ready to edit
• Preview the OCR before it joins the library

Quoted the wrong warranty? The Document catches it.

The micromanagement Bluework does so managers don’t have to: what’s said on the floor is data-matched against your own company documents. Tell a customer their repair carries a 12-month warranty when your standard says 6, and the mismatch — with the Document clause beside it — lands with the supervisor as a retraining flag, not a disciplinary one.

• Your Documents are the source of truth — every customer-facing claim checked against them
• Side-by-side evidence: what was said vs what the standard says, with confidence
• Routes to the supervisor for coaching — a knowledge-gap loop, and if the Document itself is stale, fix the Document

Off by default and switched on per worker — same consent gate as psychosocial detection.

Work items that keep the score

Folder-grouped task boards with priorities, due dates and overdue alarms — spawned from voice notes, form submissions or by hand. Compliance Mode even blocks assigning a task to someone who isn’t qualified for it.

• Live To Do / In Progress / Done tiles that recount on every action
• OVERDUE badges and a one-tap overdue filter
• Start-to-done time on every card — live while it runs
• Qualification-gated assignment — “requires Supervisor” means it

Forms that go somewhere

Build an incident report or checklist in minutes — nine field types, required flags, live preview. Then route every submission where it belongs: into a Document, a task with an assignee, a chat room, or the admin report inbox.

• Fields save as you add them — no “save form” step
• Workers fill on mobile; validation names exactly what’s missing
• One submission can fan out to Document + Task + Chat + Report

Pass the course, hold the shift

Build courses by hand, or generate them straight from your Documents, section by section — the procedure is the curriculum. A quiz pass plus approved assignments completes the course, and completion auto-grants the mapped roster qualification. Until then, the shift training gate can flag the roster — or block it outright.

• Courses built by hand or generated from your own Documents, section by section
• Quiz pass + approved assignments = completion — the mapped qualification grants itself
• Shift training gate: an unfinished required course flags or blocks rostering — your choice

Incidents and complaints with the clock running

Anyone can log an incident or complaint in three short steps, even offline. Pick a reportable category and the register sets the NDIS Commission deadline on the spot: 24 hours for serious injury, abuse or neglect, death or unlawful contact, and five business days for unauthorised restrictive practice. Managers triage, assign, tick off actions and mark it reported, and every change lands on the timeline.

• Commission deadlines set on intake, with overdue reports flagged on the register
• Assign an owner, add actions with due dates, keep a status trail from New to Closed
• Complaints tracked to acknowledgement; the whole register exports to CSV

Families see who is coming — and you bill the visit

Each family member or representative gets their own sign-in: this week’s visits and who is coming, the notes you choose to share, and a button to ask for a change. Back in the office, delivered shifts become an NDIS invoice in one click, priced from the participant’s rates and drawn against their funding budget, with your margin in view.

• Per-family access: schedule, shared notes, documents, medication and requests each switch on or off
• Unbilled shifts become invoice lines with support item numbers; budget alerts fire at 80%
• NDIA bulk-claim CSV for the myplace portal, plus Xero, MYOB and QuickBooks exports

Your AI agents join the team

Connect Grok Bot, Hermes Agent or any MCP-capable host, and every agent it runs joins as a team member. Message one directly or @mention it in a crew room, give it tasks, and its reports go to the roles they concern, not to everyone.

• Access per agent: owners only, owners and role holders, or everyone. A casual’s @mention of an admin agent is ignored
• Bluework is an MCP server: bluework_report, bluework_message, bluework_inbox, bluework_reply, bluework_tasks, bluework_roster and more
• Reports are addressed by area, so each role sees only the ones meant for it

When Grok Bot goes quiet, Hermes takes over

Set Hermes as Grok Bot’s backup. If Grok Bot is silent for six hours while Hermes is active, the panel flags the failover and Hermes reports as the same agents, each report tagged “via Hermes”. When Grok Bot checks in again, its status returns to Active by itself.

• Platform status at a glance: Active, Quiet, Standby or Failover
• Backup reports keep the agent’s name, tagged “via Hermes”
• Mark a platform down and its chat mentions and tasks go to the backup’s webhook

Let an AI agent review the detections

Micromanagement flags and psychosocial hazard detections normally wait for a person to review them. Switch the card’s gear to AI agent mode, pick an agent, and it works through the queue: coach, dismiss or mark the document outdated, approve or reject a hazard. Each decision is saved under the agent’s name with its note, the same way a reviewer’s is.

• Coach books a knowledge-gap catch-up task: training, not a conduct matter
• Every decision is in the audit trail under the agent’s name. Switch back to human review at any time
• Build your own cards from agent reports: latest report, unseen count, a number from the report, platform status

From sign-up to first roster in one checklist

New owners get a Setup checklist that checks itself against your real data: name, industry, locations, roles, team, first roster. Pick your industry and the matching modules switch on — care and NDIS included — and the list grows to match. Every worker gets a short Getting started card of their own.

• Steps tick themselves from live data, then stay on as a health check
• Industry presets switch whole module sets on; fine-tune any single module
• Workers get profile, workplace agreement and notifications in one card

Plug Bluework into the rest of your stack

Hand your BI tool a key that reads only what you tick, get a signed webhook the moment a shift is published or an offer is taken, and connect Xero, QuickBooks, MYOB, HubSpot or Employment Hero in a click. Staff sign in with the Microsoft, Google or Okta account they already use.

• Scoped keys are shown once, stored hashed and revocable any time
• Every webhook is HMAC-SHA256 signed and delivered with up to five attempts, backing off from 1 minute to 2 hours
• Connector tokens are encrypted at rest; SSO sign-in can count as two-factor