Nucleus installs the GPU, the models and the agents on your own network. Nothing leaves it: not files, not prompts, not keys.
No servers yet? Start free on Nucleus AI
Nucleus Work, the CLI and Nucleus AI share the same agent core, the same model catalog and the same skills, so a workspace behaves the same on a desk, in a terminal, and in the browser.
For business users
A desktop AI agent that keeps everything on the machine.
Everything stays local
Conversations, files and settings live in one database file on the PC. No account, no login, no server to run.
Pick the model per workspace
Use the model server in your office, or hosted models on Nucleus AI Cloud. Switch from the header; each workspace remembers its choice.
Connects to the tools you allow
Add MCP servers from a config file and the agent discovers their tools. Nothing is connected until you add it.
Builds small apps on the desk
Ask for a dashboard or an internal tool and Nucleus Work writes and previews it on the same machine.
Read 3 PDFs (41 pages) from the shared drive.
Two renew on 30 Nov; one auto-renews unless notice is given by 15 Oct.
Summary saved to Tenders/summary.md
Model for this workspace
For developers
A terminal coding agent. Work solo, or run a four-agent team over one repository.
One command: ne
Install with one line on macOS, Linux or Windows. Run it in any repository; that directory becomes the workspace.
Solo or a team of four
Single mode for quick tasks, or a coordinated product manager, engineer, QA tester and DevOps engineer for bigger ones.
Any model, switched in place
Nucleus AI Cloud, your own vLLM or Ollama servers, or a direct provider key. Reasoning effort is a keystroke.
Skills, MCP and memory
Ships with product skills, discovers tools from your MCP servers, and keeps sessions searchable per workspace.
$ cd ~/repos/billing-api && neworkspace billing-api · qwen3.8-27b · local vllm> add a retry policy to the invoice webhook client● product manager 3 steps, tests required● engineer src/webhook/client.rs (+38 −6)● qa cargo test … 42 passed✓ done · 2 files changed · 3m 40s
For teams without servers
The hosted platform. Agents operate a real computer and browser in private sandboxes we run.
Nothing to install
Sign in from a browser and give an agent a goal. Sandboxes, models and updates are ours to run, not yours.
Computer-use and browser agents
The agent takes over a real desktop and a real browser: it screenshots, clicks, types, fills forms and extracts data, then checks its own work.
Acts across your tools
Connect Gmail, Drive, Calendar, Slack, Outlook, OneDrive, GitHub and your own MCP servers. Agents act only through what you authorize.
Apps without a front-end team
Canvas turns a prompt into a hosted full-stack app with its own database. The App Builder makes no-code multi-agent apps with row-level permissions.
Goal
Find 3 event venues in Singapore and draft an email with the shortlist.
The same desktops and the same CLI, pointed at whichever models your policy allows. Start offline and add Nucleus AI Cloud later, or the other way round.
No internet required
Models run on a GPU in your office. Desktops and developer machines talk to it over the LAN. Nothing leaves the network.
Local first, Nucleus AI Cloud when allowed
Keep sensitive work on the local model and let approved people reach hosted models on Nucleus AI Cloud, with quotas.
Hosted by us
No servers to run. Agents operate a real computer and browser in private sandboxes and act across your connected tools.
A GPU in a cupboard is not an AI deployment. We size the card for your headcount, serve an open-weight model with vLLM, benchmark it against your own prompts, and put Nucleus Work and the CLI on every seat. You get one project, one invoice and one number to call.
| GPU | Memory | What fits | Team | Basis |
|---|---|---|---|---|
| NVIDIA RTX 4090 | 24 GB | Models up to about 14B parameters at full speed; larger ones quantized | A small team doing documents, email and chat | Guidance |
| NVIDIA RTX 5090 | 32 GB | Qwen 3.8 27B in NVFP4 on vLLM, long sessions with a CPU cache tier | 8 developers per card for interactive coding agents; 16 workable | Measured |
| NVIDIA DGX Spark | 128 GB unified | 70B-class models and long context on one desk-side box | Larger teams, or bigger models for fewer people | Guidance |
Measured with Nucleus agents on one RTX 5090: 8 developers at about 57 tokens per second each with all sessions resident; 16 developers all finish their tasks at roughly 1.5 times the time.
Nucleus benchmark, 2026-09-01/02, Qwen 3.8 27B NVFP4 on vLLM 0.28
We size per team from your prompts, context lengths and headcount. Existing GPUs and multi-GPU servers are welcome.
Nucleus AI Cloud gives your team hosted frontier models through the Nucleus gateway, with workspace keys and limits your admin sets. Nucleus Work and the CLI use them out of the box.
Bring your own subscription
An admin connects the model subscription your company already pays for, once. Ownership and billing stay with you.
Quotas per member
Daily input, output and cache token budgets, plus a concurrency limit, for every person.
Keys, not credentials
Members get isolated workspace keys and never see the provider login. Revoke a key when someone leaves.
One key, every tool
The same key works in Nucleus Work, the NE CLI, and any Anthropic or OpenAI SDK your developers already use.
Members and daily token quotas
Member key (a.lim)
nk_ws_acme_••••••••••••••••
Works in Nucleus Work, the CLI, and the Anthropic or OpenAI SDK.
The Nucleus Enterprise CLI runs in any repository. Point it at the office model server or at Nucleus AI Cloud, then work solo or with the four-agent team.
macOS and Linux
curl -fsSL https://llm.nucleusenterprise.ai/install.sh | shWindows PowerShell
irm https://llm.nucleusenterprise.ai/install.ps1 | iexThen, in any repository
$ cd your-repo $ ne
Set a model key in ~/.nucleus/.env: your own DeepSeek, Gemini or Nucleus AI Cloud key, a direct provider key, or the address of the vLLM or Ollama server in your office. Rows without a key show as setup required in the picker until you add one.
No servers to run and nothing to install. Nucleus AI operates a real computer and browser in private sandboxes and acts across your connected tools. Everything below is live today.
Hand off a goal and the agent takes over a real desktop: it screenshots the screen, moves the mouse, types, scrolls, runs commands, and checks its own work.
screenshot · click · type · scroll · locate · shell
A real Playwright browser navigates any site, logs in, fills forms, and extracts structured data. No API required.
navigate · click · fill · snapshot · extract
Describe an app or dashboard and Nucleus writes and runs a full-stack FastAPI + React app, with its own database and a shareable preview URL.
Python + React · per-app database · live preview
Build multi-agent apps with a built-in database and row-level permissions, entirely no-code, just by describing what you need.
collections · agents · access control
Real actions across Gmail send, Drive file picker, Calendar, Slack, Outlook, OneDrive and GitHub. Your agents act across the tools you authorize.
7 integrations · 30+ actions
Built-in tools work out of the box, and you can plug in your own remote MCP servers. Their tools are auto-discovered and ready for agents to use.
builtin + bring-your-own · auto-discovered
Most deployments take weeks, not quarters. You keep the hardware, the models and the data; we do the sizing, the installation and the tuning.
A short workshop on the workflows you want automated, the data involved, your network policy and how many people will use it.
We propose the GPU, the model and the deployment shape for your headcount. Hardware you already own is fine.
We set up the inference server, load the models, benchmark with your prompts, and put Nucleus Work and the CLI on every seat.
Team onboarding, model updates, software upgrades and a support line for as long as you want one.
Wherever there is repetitive, multi-step work on the files and tools inside your company, a Nucleus agent can take it end-to-end: on your hardware with Nucleus Work and the CLI, or hosted on Nucleus AI.
Review, summarize and extract from PDFs and scans. On-premise, nothing is uploaded anywhere.
Reconcile spreadsheets, draft reports and prepare month-end packs from files on the shared drive or in Drive and OneDrive.
A coding agent team on your repositories, against a GPU in the building or a model through the gateway.
Draft answers from your own product notes and ticket history, with a person approving before send.
Small dashboards and trackers built from a prompt, on the desk or hosted with a shareable preview.
Source-backed briefs from the live web with a real browser, or from your own files when you are offline.
An on-premise deployment is a quoted project. The developer CLI is free to install with your own model key. Nucleus AI starts free and meters usage in credits.
One quote for sizing, hardware (or your own), installation, tuning and rollout, with an optional support plan after go-live.
The CLI installs free and runs on the model key you bring, or a gateway key. Nucleus Work is in early access for on-premise customers.
No card to begin. Usage is metered in credits with a live meter, so you always see the cost as work runs.
Quotes are per project. Nucleus AI plan details are shown inside the app.