The on-prem AI agent factory.

Own your intelligence. One sealed machine runs thousands of AI agents entirely inside your walls. No public cloud. No external model calls. No token meter.

Lyzr Optimus Micro silver desktop with the Lyzr Optimus logo on its side panel

Meet Optimus Micro.

100concurrent users
on one machine

Built in

  • Lyzr Studio · Build deterministic agents and fine-tune models
  • Lyzr Open Cowork · A governed AI workspace for every employee
  • Department workbenches · Finance, HR, legal and operations on agents

Serves models from

NVIDIA NemotronOpenAIGemmaMistral

Year one

$150,000

Year two onwards

$50,000 / year

The on-prem advantage

Sovereign by design.

Models, prompts, logs, embeddings — everything lives on drives you own, inside your borders, under your law.

Ownership

AI you own.

The models, the agents, the fine-tunes and the machine they run on. Yours, outright.

Zero data ingress

No data ever leaves your walls.

Your office, comms room or air-gapped site.

Regulatory compliance

Built for regulated work.

Nothing leaves the building — so GDPR, HIPAA and data-localization reviews get short.

Software

The full Lyzr stack, on the box.

Four capabilities ship on every Optimus. Each one below is a live demo: click through it, it is the real interface.

    • Chat with open-weight LLMs running on the machine
    • Search documents across your own knowledge bases
    • Advanced reasoning that shows its working, with approvals on every step
    • Visual Superflows with guardrails and human approvals
    • Tools, knowledge and memory wired in without code
    • Deploy to every team from the same box
    • Secure document processing agents for finance, HR and legal
    • Multi-agent analysis that shows its working
    • Sensitive numbers and records never leave the building
    • Curate, dedupe and pair your own documents into datasets
    • Train open-weight candidates with LoRA, then DPO
    • Evaluate against a baseline before anything goes live

Good morning, Alex.

Answers stay within your access

01 / 04 · Chat with LLM

Scroll for Build agents

Why it changes the economics

No meter. No limits. No reason to say no.

Cloud AI charges for every word. Optimus is a machine you own, so your teams can use it like one.

$0

per token

No tokens to count.

Optimus has no meter. Every prompt, agent run and document costs the same on day one and day one thousand.

0

external calls

Air-gapped models.

American open-weight models run on the machine itself. Pull the network cable and every agent keeps working.

∞

prompts

Practically unlimited usage.

A hundred people, all day, every day. No rate limits, no quotas, no surprise invoice at the end of the month.

24/7

experimentation

Experiment without a budget.

Hand it to interns, analysts and engineers. Fine-tune, re-run, break things: tokenomics stop being a reason to say no.

Anywhere

Optimus goes where the internet doesn't.

Ocean, underground, offshore, or a room with no wires out. Build the agents where you are, ship them where the work is, and they keep answering with zero connectivity.

  • Banking and financial services.

    Compliance analysis agents run inside the bank: KYC files, credit memos and exam prep, with every prompt, model and log on drives you own.

    KYC / AMLCredit memosExam prep
    Explore banking
  • Situation rooms.

    Networked machines at the end of the table. Briefings, intelligence summaries and scenario analysis run on hardware in the room, with nothing crossing the wire.

    BriefingsIntelligence summariesScenario analysis
    Explore government
  • Container vessels.

    Train the agents ashore, ship them aboard. When an engine alarm sounds mid-Atlantic, the crew asks Optimus how to fix it — manuals, maintenance logs and schematics, answered on the bridge with no satellite link.

    Engine troubleshootingMaintenance logsCrew Q&A
  • Mines and tunnels.

    Optimus sits at the face, not the surface. Crews ask about equipment, safety procedures and geology reports hundreds of metres down, where no network reaches.

    Safety proceduresEquipment fixesGeology reports
  • Oil, gas and process manufacturing.

    On the rig or in the control room, operators read P&IDs, drilling reports and permits to work, and troubleshoot units without opening a route to the internet or waiting for a satellite window.

    Permits to workP&ID lookupTroubleshooting
    Explore manufacturing
  • Pharma and biotech.

    Lab notebooks, assay data and trial documents analysed where they are produced. Research IP never leaves the building.

    Literature reviewAssay analysisRegulatory documents
    Explore healthcare
  • Factory floors.

    Beside the assembly line, Optimus answers “how do I fix this machine?” from your manuals and maintenance history, and the line keeps moving.

    Machine fixesQuality checksMaintenance
    Explore manufacturing
  • Stadiums and venues.

    Sixty thousand phones jam the network. Optimus, beside the kiosk, keeps answering fans and staff on venue hardware: tickets, wayfinding and security.

    Fan kiosksIngress operationsSecurity
    Explore stadiums

Models

Open models. Running on the box.

Optimus serves open-weight models from American labs, led by NVIDIA Nemotron. Switch to US + Europe and the same machine runs Mistral and Apertus too. Import your own models or fine-tunes.

Optimus · 9 models
United States
Nemotron 3.5 LightningNVIDIA · Default model · fast answers256K
Nemotron 3 UltraNVIDIA · Agentic reasoning1M
Nemotron 3.5 Super VLNVIDIA · Scans, tables and charts256K
Muse Spark OpenMeta · Coding and agents · multimodal1M
gpt-oss-120bOpenAI · Reasoning and tool calls128K
gpt-oss-20bOpenAI · Lightweight reasoning and tool calls128K
Gemma 4 31BGoogle · Documents and vision256K
Gemma 4 26B A4BGoogle · Fast mixture-of-experts256K
Llama 4 MaverickMeta · General purpose1M
EuropeWith US + Europe
Mistral Large 4Mistral · Flagship · multilingual256K
Mistral Medium 3.5Mistral · Agents and tools128K
Mistral Small 4Mistral · Fast · multimodal128K
Apertus 70BSwiss AI · Fully open · 1,000+ languages64K

Every model runs on the machine itself · closed models never publish their weights, so they cannot run on-premise

FAQ

Questions enterprises ask about Optimus.

Straight answers on sovereignty, hardware, pricing and deployment — the topics that come up in every security review.

What is Lyzr Optimus?
A sealed, on-premise AI appliance that runs your enterprise's AI agents entirely within your own physical infrastructure. It provides full sovereignty over your models and data.
Are there any external model calls?
No. Optimus operates fully air-gapped if required. There is zero public-cloud dependency and absolutely zero data leaves your network.
What models come with it?
Optimus runs NVIDIA Nemotron 3.5 Lightning by default, alongside open-weight models such as Nemotron 3 Ultra, Nemotron 3.5 Super VL, OpenAI gpt-oss-120b, Google Gemma 4 and Mistral Small 4, plus search models and Whisper large-v3. You can also import your own models or fine-tunes.
Can I run GPT-5.1 or Claude Opus 5.5 on this?
No. Proprietary models like GPT-5.1 and Claude Opus 5.5 do not publish their weights, so they cannot run on the machine. Optimus runs open-weight models locally; if a team chooses to use proprietary models, that traffic goes through Lyzr's cloud, outside the box.

Components originate from non-adversarial jurisdictions and partners of the United States of America.

Manufactured in
  • 🇲🇾Malaysia
  • 🇰🇷South Korea
  • 🇹🇼Taiwan
  • 🇻🇳Vietnam

Talk to Sales

Bring the models to the data.

Pre-orders secure priority allocation in the next manufacturing batch.

  • White-glove delivery & 30-minute setup
  • Data never leaves your environment

Secure your unit

Pre-orders secure priority allocation.

Optimus Micro$150,000 year one · $50,000 per year from year two onwards.
Full name
Work email
Company (optional)
Purpose (optional)

Security

Audit-ready, out of the box.

End-to-end encryption and tamper-evident logs that satisfy examiners — not just checklists.

  • SOC 2

    Type II certified

  • HIPAA

    Compliant

  • ISO 27001

    Certified

  • GDPR

    Ready