Nod vs OpenClaw: which self-hosted AI agent?
By AppDuo Editorial Team · Updated
Quick verdict
Nod and OpenClaw are both open-source AI agents that run on your own machine or server and act for you across email, calendar, the browser and chat apps. Both are free software; the real costs are the AI model and the machine that runs the agent.
Nod (by n8n): suits people who want an agent that asks before actions its rules do not cover, stops at a monthly spending limit (with API models) and keeps its password vault under the control of Nod's own server. The trade-off is maturity: it is a research preview whose GitHub repository dates from late September 2026, it chats only in its app, Telegram and WhatsApp, the sources available show an English-only interface, and its safety figures so far come only from n8n.
OpenClaw: suits people who want many chat channels (the site cites 29, including iMessage and Signal, and Zalo is available as a plugin), iOS and Android apps, a Vietnamese-language iOS app and a very large community (about 394k GitHub stars). The trade-offs are confirmed malicious skills on its ClawHub marketplace (documented by Unit 42; the project banned the accounts, deleted the skills and added screening), tools that run directly on the host unless sandboxing is configured, and security setup left to the user.
Head-to-head
Nod leads on 3, OpenClaw on 5, 2 even
| Feature | Nod | OpenClaw |
|---|---|---|
| Price and cost control[1],[2],[8] | Free under Apache 2.0. You pay the model provider you connect, or sign in with a ChatGPT Plus or Pro plan; Usage shows estimated costs, and API models stop at a monthly limit you set. | Free under MIT, with no paid tier or hosted version. Model API usage is paid separately, plus the machine running the Gateway. |
| → An autonomous agent can run up API bills. Nod stops at a monthly cap when you use API models, while OpenClaw's total cost is hard to predict because it depends on API usage. | ||
| Setup and platforms[1],[2],[7],[8],[10] | Apps for Mac (macOS 14+), Windows 11 (preview) or your own server; the README covers running from source on macOS or Linux. Under the hood it is a Node server plus Docker containers for a Linux workspace, a Chromium browser and PostgreSQL. | You run a Gateway process on your machine or server. Install with a one-line script, npm, Docker, Nix or desktop apps for macOS 15+, Windows 10/11 and Linux; iOS and Android apps act as nodes. It still needs Node, API keys and your own security setup. |
| → OpenClaw covers more systems and has phone apps; Nod does run on macOS 14, but its Windows build is still a preview. | ||
| Chat channels for giving instructions[7],[8],[10] | In the app, Telegram and WhatsApp. | The site cites 29 channels, including WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Microsoft Teams and Google Chat; the docs list Zalo as a plugin. |
| → If you want to message your agent over Zalo, iMessage or Signal, only OpenClaw lists those channels. | ||
| Supported AI models[1],[2],[8] | 30+ providers, your own server, or local models via Ollama, LM Studio, vLLM and llama.cpp. Local models need tool calling and a 64,000-token context, and small models are less reliable at the safety check. | Many providers, including Claude, GPT and local models. |
| → Neither locks you into one model vendor. Nod spells out its requirements for local models, which is worth knowing before you plan a fully local setup. | ||
| What the agent can do[1],[2],[7],[8],[9] | Its own Linux workspace, a real Chromium browser, editable memory and schedules; connections to Gmail, Google Calendar, Slack, Notion and Linear via Composio, MCP, APIs and CLIs. n8n's examples: chasing invoices, preparing for calls, cancelling trials on real websites. | Persistent memory, email and calendar, web browsing, form filling, reading and writing files, and shell commands. Community skills and plugins extend it, and the agent can write new skills itself. |
| → The range of tasks is similar. The difference is where commands run: Nod has its own Linux workspace in a Docker container, while OpenClaw runs tools directly on the host unless sandboxing is on. | ||
| Default safety controls[2],[9],[10] | It asks before actions your rules do not cover. Per the README, the trust check, password vault, payments and spending limits run in the server, app and web results are treated as untrusted, and provider keys stay outside the sandbox. | Tools run on the host unless sandboxing is configured. The README advises treating inbound messages as untrusted, and the docs show how to restrict senders with allowlists. The agent needs broad access: email, calendar, files and shell. |
| → Nod ships with more guardrails switched on, although its README admits the check still relies on a model's judgement. With OpenClaw, safety depends on you enabling the sandbox and limiting who can send commands. | ||
| Documented security issues[2],[12],[13] | No controversies recorded. Safety figures are all vendor-measured in simulated environments: 0 unsafe actions in 678 runs, but Claude Haiku 5.5 let 8 unsafe actions through in 240 decisions. No independent evaluation yet. | Confirmed: malicious skills on ClawHub; Unit 42 cites Koi Security's disclosure of 341 malicious skills and found five more that slipped past screening between February and May 2026. Alleged: per Wikipedia citing The Register, version 2.0 does not encrypt secrets at rest or enable the sandbox by default. |
| → Nod's record is cleaner, partly because its repository only dates from late September 2026. OpenClaw's only confirmed issue involves third-party skills, so check what you install. | ||
| Maturity and community[1],[2],[3],[6],[9],[11],[12] | A research preview that the site calls an experiment that can make mistakes. The repository was created on 25 Sep 2026, with 0.6.0 to 0.6.2 released on 9–11 Nov 2026; #1 of the day on Product Hunt (11 Nov) with 245 points. | First released in November 2025 (as Warelay), renamed OpenClaw in late January 2026, version 2.0 in August 2026. About 394k GitHub stars and 4.8k open issues; the iOS app is rated 3.5/5 from 76 ratings on the US App Store. |
| → OpenClaw has nearly a year of public development and a large community to ask; Nod is too new to judge its stability. | ||
| Data leaving your machine[1],[2],[5],[8] | Database, files, memory and saved logins stay on the machine running Nod. Usage data goes to n8n by default (turn off under Settings, Privacy & data), and app connections go through n8n and Composio by default; conversation traces are sent only if you opt in. | By default it sends only a daily version check, which can be turned off. Messages and data still go to the model provider you choose. |
| → If you want as little as possible sent to the developer, OpenClaw sends less by default; with Nod, turn off usage sharing after installing. | ||
| Vietnamese interface[1],[7],[11] | The sources reviewed show an English-only interface. | The iOS app lists 22 languages, Vietnamese among them. |
| → Readers who prefer Vietnamese will find OpenClaw easier to navigate, at least in the iOS app, although setup still requires technical knowledge. | ||
Choose Nod if…
- You want an agent that asks before unapproved actions and stops at a spending limit
- You plan to run local models through Ollama, LM Studio, vLLM or llama.cpp and want clear model requirements
- You already work with n8n, are fine with an English interface and accept a preview release
Choose OpenClaw if…
- You want to give instructions over Zalo, iMessage, Signal, Discord or Microsoft Teams
- You need iOS and Android apps and an iOS app in Vietnamese
- You are comfortable with the command line and will enable the sandbox, restrict senders and vet third-party skills yourself
Both run on your machine, with different priorities
Nod is a separate app from the team behind n8n (Germany), not the AI assistant inside n8n. It runs a Node server plus Docker containers for its own Linux workspace, a Chromium browser and a PostgreSQL database, and you talk to it in the app, on Telegram or on WhatsApp. OpenClaw centres on a Gateway process on your machine that connects to many chat apps and model providers, with phone apps acting as nodes. In short, Nod puts guardrails first, while OpenClaw puts channel reach and its skills ecosystem first.
Safety: defaults and known incidents
Per its README, Nod asks before anything its rules do not cover, and the trust check, password vault and spending limits live in Nod's server rather than only in the prompt. The README also admits the check depends on a model's judgement, and every safety figure is measured by n8n itself. For OpenClaw, malicious skills on ClawHub are confirmed: Unit 42 found five that slipped past screening between February and May 2026, two of them delivering macOS infostealers. Per Unit 42, ClawHub had integrated VirusTotal and ClawScan to screen skills; after being notified, the project banned the accounts and deleted the skills, and on 1 June 2026 it announced a partnership with NVIDIA on screening. The Register's criticism of 2.0 (secrets not encrypted at rest, sandbox off by default) remains an allegation with no project response found, although the README confirms that tools run on the host unless sandboxing is configured. Wikipedia also reports a student's account that his OpenClaw agent created a profile on MoltMatch, a third-party dating platform, without his explicit direction; that too is unverified.
Brand-new preview or established community
Nod's repository dates from 25 Sep 2026 and it launched on Product Hunt on 11 Nov 2026; its own site calls it an experiment that can make mistakes, and the Windows build is still a preview. OpenClaw has nearly a year of development, about 394k GitHub stars, and is stewarded by the OpenClaw Foundation, a US 501(c)(3) nonprofit; its README says OpenAI is a donor, not an owner. On the other hand, its GitHub shows about 4.8k open issues, and a maintainer has warned it is unsafe for people who cannot use a command line. Bloomberg reported that Chinese state agencies and state-owned enterprises, including the largest banks, were warned against installing it on office devices; the claim rests on anonymous sources and concerns work devices only.
Costs and data
Both are free with no paid tier; n8n only says a managed hosted Nod is on the roadmap. What you pay for is the AI model, nothing with a local model, plus the machine running the agent. Nod's schedules run only while the host is on and awake, so n8n suggests an always-on machine such as a Mac mini or your own server. On data, Nod keeps memory and logins locally but sends usage data to n8n by default, while OpenClaw by default sends only a daily version check. With either, task content still passes through the model provider you choose.
Frequently asked questions
Is Nod or OpenClaw safer?
By default, Nod has more guardrails: it asks before unapproved actions, and spending limits and the password vault sit in Nod's server. But the check still relies on a model's judgement, and its safety figures come only from n8n. OpenClaw offers sandboxing and sender allowlists that you must turn on yourself, and its ClawHub marketplace has hosted malicious skills, confirmed by Unit 42 and since removed by the project.
Do these agents cost anything?
The software is free: Nod under Apache 2.0, OpenClaw under MIT. You pay the AI model provider (Nod also accepts a ChatGPT Plus or Pro sign-in), nothing with a local model, plus the machine running the agent.
Can I use Zalo or a Vietnamese interface?
With OpenClaw, yes: its docs list Zalo as a channel plugin and its iOS app includes Vietnamese. Nod lists only its app, Telegram and WhatsApp, and the sources reviewed show an English-only interface.
Do I need an always-on machine?
Nod's schedules run only while the host is on and awake, so n8n recommends an always-on machine. OpenClaw also needs a machine running its Gateway, and that hardware is part of the real cost.
Sources
Numbers in the table refer to these sources, read on the dates shown.
- 1.Nod official site · Vendor-provided · Nov 12, 2026
- 2.Nod GitHub repository and README · Vendor-provided · Nov 12, 2026
- 3.Nod GitHub releases · Vendor-provided · Nov 12, 2026
- 4.Nod security policy (SECURITY.md) · Vendor-provided · Nov 12, 2026
- 5.n8n privacy policy · Vendor-provided · Nov 12, 2026
- 6.Product Hunt launch page · Community-reported · Nov 12, 2026
- 7.Nod documentation · Vendor-provided · Nov 12, 2026
- 8.Official website · Vendor-provided · Nov 11, 2026
- 9.GitHub repository · Vendor-provided · Nov 11, 2026
- 10.Official docs · Vendor-provided · Nov 11, 2026
- 11.App Store listing (US) · store · Nov 11, 2026
- 12.Wikipedia: OpenClaw · Community-reported · Nov 11, 2026
- 13.Palo Alto Networks Unit 42 (2026-06-23) · press · Nov 11, 2026
- 14.The Star / Bloomberg (2026-03-11) · press · Nov 11, 2026
Prepared by the AppDuo Editorial Team from the sources cited in this article. AI assists with organising and cross-checking data; editorial responsibility remains with AppDuo.
Comments
No comments yet. Be the first to share your experience.
Comments are readers’ opinions, not content from the AppDuo Editorial Team.

