Skip to content

Native AI

Bromel.

The native chatbot, the agentic workforce, and the insight engine — allowed to help, not allowed to improvise.

Every vendor says they have AI. The question a regulated buyer actually asks is what it is prevented from doing — and whether that prevention is a prompt or a mechanism.

Build with Bromel

From a conversation to a workflow draft.

Describe a supported approval workflow to Bromel, review its preview and save an inactive draft for authorized activation

Authorized users review and activate drafts. Supported lifecycle types add controlled release and signature requirements; unified chat-to-launch and broader regulated lifecycle coverage remain in development.

Ask Bromel panel answering which contracts expire this year, flagging a renewal opt-out deadline six days away, over the morning briefing and task list
Asked live against demo data: the answer names the contract, its value, and the renewal opt-out deadline six days out — then offers to file the task. Behind it, the morning briefing and a task cross-fired from a sourcing award.demo data

01/The guardrail

The no-list is code, not a request

Writes go through an action registry: propose, confirm, execute. And the registry carries a machine-readable list of what may never be built as an AI action at all.

Keep AI assistance within enforced boundaries: no journal entries, regulated Part 11 record changes, pipeline-stage changes or external communicationssrc/lib/bromel/action-registry.ts — the no-list throws at module load. If someone tries to register a forbidden action, the product does not start.
Asked live: post $40,000 straight into the general ledger. Bromel refuses and says why — it cannot post journal entries or GL transactions. The no-list is enforced when the code loads, not asked of the model.Recorded live, demo data

02/Three shapes

Chat, insights, agents

Ask questions of your own operational data in plain languageper-module toolpacks, gated on licensing — an org without a module never sees its tools
Get calculated business insights explained in plain languagesrc/lib/signals — the model narrates, it never computes
Have named agents work overnight — controller, FP&A analyst, supply planner, production scheduler, quality manager, treasury analyst14 registered agent slots
See exactly what AI cost you, by org and by daydaily per-org caps checked before the model call

The insight engine

Describe the view you need in a sentence. Bromel assembles the charts and the metrics from your real data — the reporting layer without the BI backlog.

Recorded live: describe a CFO dashboard in one sentence and Analytics Studio builds it — a revenue trend, revenue by customer, and DSO, charted from your real data. No BI team, no ticket. Save it or share it with the org.Recorded live, demo data

03/The roster

Named agents, stated bounds

Each runs on its own cadence with per-org failure isolation and once-a-day notification dedupe. Where an agent has a stated bound, it is printed. And the roster is not fixed — new agents ship continuously, and arrive as part of the product, not an upgrade project.

8 shown
  • Production schedule sweepNightly

    Sweeps the next two weeks every night — work-centre overloads, maintenance overlaps, work orders at late risk, run orders wasting changeover hours — and proposes resequences of planned, unlocked blocks.

    Recommendations apply to planned, unlocked schedule blocks. The agent does not change released work, batch records or regulated electronic records.

  • MRP exception triageNightly

    Runs live MRP every night, then triages the exceptions — shortages ranked by stockout risk with the demand that causes them — and proposes the purchasing moves.

  • Rep account-health briefingWeekly

    Watches the book: stale quotes and past-due closes nightly, cadence breaks and gone-quiet customers every Monday — then briefs each rep and proposes the follow-ups.

  • AR collections sweepNightly

    Watches receivables every night — aging drift, the largest past-due balances, payment behaviour shifting — and proposes collection follow-ups and dunning runs.

  • Quality system sweepNightly

    Sweeps the quality system every night and proposes chase tasks to the responsible owners — the deviation that is quietly aging out gets chased before the audit finds it.

  • Weekly cash triageWeekly

    Runs the cash checks weekly and proposes the triage — what must be funded, what can stretch, and which flags need a human decision.

  • PM owner briefingNightly

    Reviews every active project each night — schedule gaps, overdue and blocked work, statuses that don't match reality — and briefs the owners.

  • Morning synthesisNightly

    Reads every other agent's overnight report and writes the one morning note that matters: the state of the business, the three items worth your attention, and which agent found each.


An agent, at work

The chief-of-staff agent reads every other agent’s report overnight and writes the one note that matters — so the briefing is already on the page when you log in, not a chat you have to open.

Recorded live on demo data: the morning note rewrites itself, synthesizing Finance, Quality, Supply, and Projects into one paragraph — the single most pressing item, and where to start. Written overnight, waiting when you log in.Recorded live, demo data

The AI is per-module, so start where you start.