It chases invoices, preps your calls and cancels trials, on real websites.
It runs on your machine with the model you choose, and asks before it acts for you.
macOS 14+ (Apple silicon or Intel) · Windows 11 preview · Free and open source · ChatGPT Plus or Pro, or an AI key
It gets on with the work across Gmail, Calendar, Slack, Notion, and the web,
and asks before it speaks for you.
Every morning, Nod reads your threads, your calendar, and the client's latest news,
then writes a one-page brief for each call.
It drafts the reply you owe from your own pricing sheet,
and sends it only after you say yes.
It checks a page on your schedule and tells you only when something changes.
Jobs that eat an afternoon of tabs, logins, and forms.
Nod has its own computer and browser, so it finishes them.
Say once that Anna, not Bob, handles invoices. It remembers.
| Every note, visible. See, edit, or forget each one, with the message it came from. |
Drafts that sound like you. Learns your style from sent emails, with no names or numbers. |
| Tell it once. Say something changed, and it uses the new fact from then on. |
It learns how to do a task. Reuses the steps, and fixes them when a site changes. |
It learns about you only from your own messages, never from the pages or emails it reads.
How memory works
- Runs on your computer or server. Files, memory, and saved logins stay there. Your model provider gets what each task needs.
- Uses the model you choose. OpenAI, Anthropic, Gemini, OpenRouter, Bedrock, Vertex AI, Azure, 20+ more, or your own server. Or sign in with ChatGPT Plus or Pro.
- Follows your rules. The server checks each app action, and each website step that can change something, against your request and rules. Anything they don't cover waits for your yes.
- Keeps secrets out of the model. Vault enters passwords, codes, and cards privately on the website.
- Shows what it did. Every action and sign-in is in Usage, with costs and a monthly limit on API models.
How trust works · Where your data goes · Security model
We give Nod real errands in test apps and websites, then check what actually changed.
98%tasks finished226 errands, 3 runs each |
0unsafe actionsin those 678 runs |
93.2%memory recallLongMemEval-S |
0 of 216false memoriesWhisperBench attacks |
How we measured this · Run the evals yourself
GPT-6.1 Sol, October 2026, in simulated accounts and websites. Test results, not a guarantee.
One Node process and a few Docker containers, all on the machine that runs Nod.
flowchart TB
you(["You<br/>app · Telegram · WhatsApp"]):::pink
server["Nod server<br/>Pi runtime · trust check<br/>Vault · memory · schedules"]:::pink
you --> server
server --> workspace["Linux workspace<br/>sandbox container"]
server --> browser["Chromium<br/>browser container"]
server --> db[("PostgreSQL")]
server -- "your key" --> model["Your model<br/>provider"]
server --> apps["Your apps<br/>Gmail · Slack · MCP"]
classDef pink fill:#ff91ac,stroke:#191919,color:#191919- Policy lives in the server, not the prompt. The trust check, Vault, payments, and spending limits run in the server.
- Untrusted by default. App, web, and browser results reach the model marked as data, not instructions.
- Keys stay out of the sandbox. Provider keys never enter the workspace or the browser.
- Built with TypeScript on Node.js 24, Fastify, React, PostgreSQL with pg-boss, the Pi agent runtime, agent-browser, Composio, MCP, and Electron.
Architecture · Agent runtime · Security model
- Run while your machine is asleep. Hosted, so the morning brief still arrives.
- Call businesses on the phone. For the errands that still need a voice.
- One Nod for your team. Today one owner per Nod. Next, a few people sharing one.
- Make it your own. Change how Nod works in the open code, and send it back.
What does it cost?
Nod is free and open source. You bring your own model provider, which bills you directly. With an API model selected, Nod shows what each task cost and stops at your monthly limit.
Do I need n8n?
No. Nod is its own app, made by the team behind n8n. It is not the n8n AI Assistant.
Where does my data go?
Your files, memory, and saved logins stay on the machine running Nod. Your model provider gets what each task needs, and app connections go through a connection service that n8n hosts by default. Nod also shares basic usage data with n8n, on by default for new users. Turn off Share usage data in Settings → Privacy & data to stop new sends. Your data and privacy lists exactly what is sent.
Does it work when my laptop is closed?
Not yet. Schedules run while the computer that runs Nod is on. Run it on a Mac mini or your own server to keep it always on. A hosted version is next on the list.
Can it make mistakes?
Yes. Its safety check is a model's judgement, and Nod is a research preview. That is why it asks before anything you didn't ask for, and why everything it did is on record.
Pick a capable model. In 240 test decisions, GPT-6.1 Sol and Claude Opus 5.5 got every one right. Claude Haiku 5.5 let 8 unsafe actions through.
Install the app, connect a model and your apps, then ask for something
in the app, on Telegram, or on WhatsApp. Setup takes about ten minutes.
Install guide · Connect a model · Connect your apps · Host your own server
Run from source
On macOS or Linux, install Node.js 24.21 or later, pnpm, and Docker. Then run:
git clone https://github.com/n8n-io/nod.git
cd nod
corepack enable
pnpm startSee Install from source for the Linux Vault key and access link, and docs/ for the docs of this revision.
Nod is open source, from the team behind n8n.
Read CONTRIBUTING to send your first change.
Report a bug · Share an idea · Discord · Forum · Security
Copyright 2026 n8n GmbH · Apache License 2.0 · Notice
Bundled third-party assets keep their own terms: Geist, provider logos,
and the desktop runtime components listed in the app's licenses/COMPONENTS.md.