New Multiplayer workbench — your team and AI in one thread, with a full audit trail.
niveau

The AI-native workbench for regulated enterprises.

Run local and cloud models side by side — on your own hardware or your approved cloud — with audit trails and data controls built in. On every device your teams use.

Download niveau See the workbench
  • macOS
  • Windows
  • Linux
  • iPhone
  • Android
Three team members and niveau working together in one thread on a CAPA inside a quality management system, with documents, review status and an audit trail alongside.

acme-medical › qms-iso-13485

CAPA-0142 root cause
3 here
AK

Anna Kowalski QA lead 09:12

@niveau draft a root-cause summary for CAPA-0142 from DEV-2291 and the last three supplier audits.

niveau llama3.3:70b · on-prem

4 tool calls · DEV-2291, SA-17, SA-21, SA-24

Draft based on DEV-2291 and supplier audits SA-17, SA-21 and SA-24:

  • Root cause — sterilizer cycle was not revalidated after the supplier changed its load configuration.
  • Impact — three lots on hold, no product released to market.
  • Action — revise SOP-014 §4.2 to require revalidation on any load change.

Mapped to ISO 13485 §7.5.6 and §8.5.2 · 4 citations

JW

Jonas Weber Regulatory 09:15

Matches what I saw. I’ll check it against our 21 CFR 820 mapping before we sign.

Marta is typing…

Reply to the team, @mention people or documents, run /playbooks
Model llama3.3 · on-prem

The name

niveau

/ni.ˈvo/ · noun, from French

A level. A standard. The bar you refuse to drop below.

AI can produce the work. People decide whether it is good enough.

Models are fast, tireless and occasionally wrong. In a regulated enterprise, someone still has to bring taste, exercise judgement and sign their name under the result. niveau is built around that person — the AI does the lifting, your team holds the level.

  1. Taste Knowing what good looks like before the model does.
  2. Judgement Choosing which answer to trust, and when to ask again.
  3. Accountability A named human approves every change that ships.
  4. Niveau The standard your organisation is held to — every time.

Your AI shouldn’t be locked to one vendor, one device, or one data center.

niveau puts every model on the same level. Two ways to run them — and you can use both in the same conversation:

niveau Local

Private models, on your own hardware.

Run open-weight models through Ollama, llama.cpp or MLX. Nothing leaves the machine, it works fully offline, and sensitive data never crosses your network boundary.

  • Ollama
  • llama.cpp
  • MLX
  • LM Studio
Explore local models

niveau Cloud

Frontier models, with your own keys.

Connect the providers your organisation has approved. Route each prompt to the right model, track cost per team, and switch providers without rewriting a single playbook.

  • Anthropic
  • OpenAI
  • Google
  • Mistral
  • Azure OpenAI
Explore cloud providers

Every platform

Start on your desktop. Finish on your phone.

One workbench on five platforms. Conversations, playbooks and model policies follow your people — and phones can borrow the compute of on-prem GPUs.

For your people

Native apps on every device.

  • macOS

    Native app tuned for Apple silicon. Runs MLX models on the GPU and Neural Engine.

  • Windows

    CUDA and DirectML acceleration, deployable via Intune and Group Policy.

  • Linux

    AppImage, .deb and Flatpak. Headless mode for servers and air-gapped sites.

  • iPhone

    Small on-device models, or reach the models running in your data center.

  • Android

    Same playbooks and history, synced end-to-end encrypted.

For your IT

Enterprise sign-in, switched on when you need it.

Connect the identity provider you already run. niveau follows your directory: who gets in, with which role, and who loses access the day they leave.

  • Okta
  • Microsoft Entra ID
  • Google Workspace
  • Any SAML or OIDC provider
  • Single sign-on SAML and OpenID Connect, with the login your people already use.
  • Directory sync SCIM provisioning. Joiners get access, leavers lose it, automatically.
  • Enforced MFA Require a second factor for every account in your domain.
  • Self-serve setup Your IT connects the provider in an admin portal. No support ticket.

Connections

Works with the tools your team already uses.

Mail, calendars, files and chat. niveau reads and drafts where the work already happens, with the same review and sign-off as everything else.

  • Gmail
  • Google Drive
  • Google Calendar
  • Google Meet
  • Outlook
  • Teams
  • SharePoint
  • OneDrive
  • Excel
  • Slack
  • Notion
  • Dropbox

Playbooks

Purchasing, Controlling and Bookkeeping. Playbooks with agents that just work.

Out of the box, niveau ships with playbooks for the finance back office. Agents do the repetitive work end to end; a named person signs off on everything that matters.

Playbook · Purchasing

A self-learning purchasing flywheel. From the first order to the negotiating table.

Tens of thousands of products, dozens of suppliers, each with its own terms. niveau checks every purchase against what was agreed, product by product and supplier by supplier, and shows a traffic light for each. Before a negotiation it prepares your buyer: every condition, every shortfall and every number in one brief. Your buyers walk in with ammunition, not gut feeling.

The purchasing flywheel

  1. Every purchase
  2. Checked per product & supplier
  3. Negotiation brief
  4. Better terms

over €60,000

saved, combined across our suppliers, by negotiating each one on its own numbers.

Rebate check · per product Wholesaler North 12,480 products · 6 agreed conditions
ProductAgreedReceived
on terms Screws M8 4711-2083.0%3.0%
missed Hand tool set 2204-1152.5%1.8%
slipping Cable ties 3390-0424.0%3.7%
on terms Safety gloves 5120-3172.0%2.1%
missed Cutting discs 6045-0903.5%2.4%
  • on terms
  • slipping
  • missed

Playbook · Bookkeeping

Posting entries, invoicing and dunning, automated in house.

The agent proactively hunts down missing invoices, reads them, proposes the posting entry and assigns the accounts. It drafts outgoing invoices, matches incoming payments and marks invoices as paid, and sets the dunning level on anything overdue. Sensitive documents stay on a local model; your accountant reviews the exceptions and signs off.

−€36,000

in external advisory costs, every year.

  1. Find missing invoices · proactive step 1
  2. Propose posting entry & assign accounts step 2
  3. Create outgoing invoices step 3
  4. Match payments · mark as paid step 4
  5. Set dunning levels step 5
  6. Accountant signs off human sign-off

“A playbook is a recipe: which models, in which order, with which rules. Write it once and it runs the same on a laptop, a phone, or a server in your own data center.”

From the niveau docs

Anthropic, OpenAI, Ollama, Google Gemini, Mistral, llama.cpp, MLX, OpenRouter, Groq, LM Studio, DeepSeek, xAI

Why niveau

Control. Compliance. Choice.

Private by default

Local models never phone home. Cloud keys live in the OS keychain, and sync is end-to-end encrypted. You decide exactly which data may leave the device.

Audit-ready from day one

Every prompt, model and response is recorded in a tamper-evident log. Export it for GDPR, DORA or ISO 27001 reviews without extra tooling.

Every model on the same level

Ask several models at once and read the answers side by side. Swap the model behind a playbook step and the approved workflow keeps running.

Ready to level up?

niveau runs on every platform your organisation uses. Bring your own models, bring your own keys, keep your own data.