Read the source. Run it offline. Keep it forever.
Hire your first employee tonight. Own it forever.
It answers the phone, replies to email, works every lead, and remembers each customer.
Point it at a mailbox, a phone line, or a folder. It runs on your machine, against the AI provider you choose, and the rules about what it may touch are written in code you can read.
Free to download.
macOS, Windows and Linux. Bring your own model keys, or run local models with Ollama.
Where it stands today
414
employees, plugins and skills on the marketplace
48
releases shipped since March 2026
938
commits you can read, all Apache-2.0
Nebo went public in March 2026. These are the real numbers, not a launch graphic.
Running in three commands
- 1
Install it
$ brew install --cask neboloop/tap/neboOn macOS. Windows and Linux installers are on the download page. All downloads
- 2
Hire your first employee
Paste a code from the marketplace — it arrives with its workflows, tools and skills already wired.
- 3
Point it at your work
Connect a mailbox, a phone line, or a folder. It works the job from there.
Nineteen step-by-step tutorials, from your first hire to a workforce that hands work between itself.
What it does on day one
Answers the phone
A real number rings, an employee picks up, takes the details and books the callback. No hold music and no missed lead at 9pm.
Works the inbox
It reads the thread, replies in your voice, and keeps threading correctly on the fourth reply. Gmail and IMAP plug straight in.
Chases every lead
New enquiry comes in, it answers within the minute, follows up on the ones that go quiet, and hands the hot ones to you.
Remembers the customer
Names, deals, prior conversations. Shared across your whole workforce, or sealed per employee and per matter when the work demands it.
Runs declared workflows
Steps written down in advance, an approval gate before anything irreversible, and every run recorded step by step.
Uses your own tools
Plugins are native binaries, skills are markdown, MCP servers plug in as they are. Write one in an afternoon.
Talks to any model
OpenAI, Anthropic, Google, DeepSeek, or Ollama on your own GPU. Swap provider without rewriting an employee.
Hands work to each other
The receptionist books it, the coordinator drafts it, the closer follows up. They pass the job along without you relaying it.
The three questions you are about to ask
- Where does my data go?
- Conversations, files and memory sit in SQLite on your disk. Only the prompt goes to the model provider you configured, and with Ollama not even that leaves the machine. We do not train on your business.
- What happens if I stop paying you?
- The runtime is Apache-2.0 and it is already on your hardware. Cancel NeboAI, put your own provider key in, and every employee you built keeps working. Nothing phones home to check.
- What does it cost at scale?
- The runtime is free on every machine you own, with no per-seat charge. Bring your own key and you pay the model provider directly at their price. NeboAI is for the parts we host: the marketplace, cloud employees, phone numbers and shared memory.
Open core, stated plainly
What's open — this is the whole runtime
Employees, workflows, skills, plugins, MCP, local memory and the approval gate. Apache-2.0. Point it at OpenAI, Anthropic, Google, DeepSeek — or Ollama, and it never leaves the machine. Fork it, ship it, run it forever.
Read the sourceWhat we sell — NeboAI
The marketplace of ready-to-hire employees, cloud employees that never sleep, a real phone line, company memory shared across your workforce, and the Right Fit Gateway, which fits each request to the model that should handle it. That is the business, and it is how the runtime stays free.
See the managed platform →It plugs into what you already run
Models
OpenAI · Anthropic · Google · DeepSeek · Ollama
Tools
Gmail · IMAP · Calendar · Drive · Slack · any MCP server
Platforms
macOS · Windows · Linux
One command. Then judge it yourself.
Free to download.