What Is NiubiGEO? The Open-Source Tool That Measures How AI Answers Mention Your Brand

You give NiubiGEO your domain and keywords; it queries the APIs of the models you choose and measures how AI answers describe your brand, which competitors it's mentioned alongside and which sources the answer cites. It stores every finding together with the raw answer and repeats the measurement to show the change. It's an open-source GEO tool that runs on your own server with your own model API key: the software is free (Apache-2.0), and model and server costs are yours. The interface is in English, Chinese and Portuguese; no Turkish.

What it does: Measures brand visibility in AI answers
Where it runs: On your own server (Node.js or Docker)
No-code: Setup needs a terminal; the maker sells a paid report
Turkish interface: No (English, Chinese, Portuguese)
Version: v0.3.2 (October 23, 2026), new project
Price: Software free; model API and server costs are yours

What does NiubiGEO do? Can I use it without code?

Customers searching for products now ask AI too: "which accounting software for a small business?" Does your brand appear in that answer, is it described correctly, who are your competitors? GEO (generative engine optimization) is the name for the work done to make a brand appear in AI answers; NiubiGEO measures that visibility. It doesn't give a single score: it ties every finding to the model's raw answer and the source it cited, so you can check it yourself.

Can I use it without code?

Not the open-source version: it needs a terminal and knowledge of Docker or Node.js. If you want results without installing anything, the maker offers a paid GEO diagnostic report on niubigeo.ai (CNY 99, ≈ $15 at the TCMB rates of November 4, 2026; up to 3 questions × 3 models, raw answers, sources, competitor comparison, recommendations). The site is in Chinese and partly English; whether payment from Turkey is possible is not stated in the official documentation.

Ready-made examples

The repository includes measurements of 20 real domains, including Notion, Figma and PostHog, that you can read without installing anything or using a key.

How AI mentions your brandWhat you get: Per model: description, category, competitors, related keywords and the raw answerNeeds a separate setup Who gets recommended for a keywordWhat you get: Mentioned and recommended products, the top one, the original sentenceNeeds a separate setup Scheduled measurement and diff viewWhat you get: Added and dropped recommendations, rank and source changesNeeds a separate setup

Who it helps: marketers, SEO/GEO agencies and software teams who want to measure their product's or client's AI visibility on their own infrastructure with models they choose. Running on your own server is a plus if you don't want commercial information like keyword and competitor lists sitting in an outside dashboard.

How does NiubiGEO work?

Enter your domainEach product is a separate project

Choose modelsOpenRouter or another API; web search per model

Review the answersDescription, competitors, keywords, sources, raw answer

Repeat and compareMeasure the same scope again and see the change

What's measured is the models' API answer, not the answer a user sees in the ChatGPT or Gemini app.

Models answer independently of each other; if one fails, the others' results remain and the error is logged. You turn web search on or off separately for each model; answers with and without search should be interpreted separately.

How do you install NiubiGEO? Node.js and Docker

The published image is for Linux amd64 and arm64. First define the OPENROUTER_API_KEY variable in your shell, or connect a provider in the workbench via a platform logo.

Start the workbench (official)

› docker pull ghcr.io/albert-weasker/niubigeo:v0.3.2
docker run -d --name niubigeo \
  -p 127.0.0.1:8787:8787 \
  -e OPENROUTER_API_KEY \
  -v niubigeo-data:/app/data/product-v2 \
  ghcr.io/albert-weasker/niubigeo:v0.3.2

Then open http://localhost:8787. Ongoing monitoring needs a separate worker container from the same image; see the monitoring section below.

Requires Node.js 22.13 or later and your own OpenRouter API key.

Install (README)

› git clone --branch v0.3.2 --depth 1 https://github.com/Albert-Weasker/niubigeo.git
cd niubigeo
npm ci
cp .env.example .env

Enter the OPENROUTER_API_KEY value in the .env file, then start:

Start› npm run server

In the repository folder (official)

› docker compose pull
docker compose up --no-build -d niubistar-geo
# Optional: ongoing monitoring
docker compose --profile monitoring up --no-build -d niubigeo-worker

Compose keeps data in the ./data/product-v2 folder; the standalone Docker command uses a volume named niubigeo-data. They aren't the same storage: always use the same one for the server and the worker.

Requirement	What the docs say
Runtime	Node.js 22.13+ or Docker (Linux amd64/arm64)
Model access	An OpenRouter key; also OpenAI-compatible APIs. The example config file has key fields for OpenAI, Anthropic, Gemini, Perplexity and DeepSeek
Database	No separate database; project data is on disk (`data/product-v2`)
Upgrading	Back up the whole data folder first; don't delete the data volume. It runs the storage migration step for data coming from v0.1/v0.2

How do you run a measurement with NiubiGEO? Examples

Enter your domain: Create a project for your product; it's saved before the test starts.
Choose models: Search for and select one or more models, and set web search separately for each. The docs recommend starting with a single model.
Start the test: Save the setup and run the test.
Open the results: Review descriptions, competitors, keywords and sources; open the raw answer to verify a finding.
Confirm keywords and measure: Confirm the keywords you want to measure, run tests without the brand name, repeat or schedule them.

1. Domain test (input)"yourcompany.com"What you get: for each model, how the brand is described, which category it's placed in, which competing products it's mentioned with and the keywords associated with the brand; behind each row, the model's raw answer. The docs point out that a model recognizing the domain when asked is not the same as recommending it unprompted.

2. Keyword test (without the brand name)"bookkeeping software for small businesses"What you get: which products are mentioned and recommended in the answer, who comes first, the original sentence and, if web search is on, the cited sources. You can sort keywords into discovery, alternative, comparison, brand or use-case types. Measuring with Turkish keywords isn't covered separately in the docs; check the result against the raw answer.

The docs' warning: being mentioned, being described positively and being explicitly recommended are different things; a cited source doesn't by itself explain why the model recommended something.

Source: NiubiGEO README ("How to use it", sample measurements); October 29, 2026. Input examples are illustrative; output is described as far as the docs explain it.

Ongoing monitoring and diff view

A single answer is a snapshot. In the keyword monitoring module that came with v0.3 (BETA), you add keywords one by one or in bulk, start from ready-made templates and choose hourly, daily or weekly frequency. After two measurements, Diff View shows added and dropped recommendations, rank changes and source links. This version has no separate competitor detection panel.

Start the monitoring worker (official, Docker)

› docker run -d --name niubigeo-worker --no-healthcheck \
  -e OPENROUTER_API_KEY \
  -v niubigeo-data:/app/data/product-v2 \
  ghcr.io/albert-weasker/niubigeo:v0.3.2 \
  node dist/src/product/scheduling/schedule-worker.js 60

Every scheduled measurement that runs incurs API charges. The worker can't use a temporary key you entered in the browser; it runs with the server-side key.

Is NiubiGEO free? Costs

The software (Community Edition) is free, open source and Apache-2.0 licensed: commercial use, modification and distribution are allowed. What you pay for is separate: the model APIs and web search you choose (at each provider's own prices) and the server. The amount depends on the number of models, the number of keywords and measurement frequency; the official documentation gives no figure.

The project's known limitations document (September 8, 2026, covering the v0.2.0-rc.1 pre-release) says core measurement has no per-request token or cost limit; the situation for v0.3.2 is not stated in the official documentation. Set a spending limit in your OpenRouter or model provider account.

The maker's paid service	Price (niubigeo.ai)
GEO diagnostic report (up to 3 questions × 3 models, within 1 business day)	CNY 99 (≈ $15) / report
AI testing with real people (choose platform, country, web or app)	CNY 5 (≈ $0.75) / person / question
Product trials, content publishing, custom tasks	Depends on budget and site

Prices are in Chinese yuan; dollar amounts are approximate, at the TCMB rates of November 4, 2026. No Turkish lira price is published; payment methods and payment from Turkey aren't stated on the pricing page.

My data and security

Where the data lives

Projects, keywords, competitor lists and measurement history stay on your own server's disk. The docs note this doesn't mean "zero data sharing": your queries go to the model APIs you choose.

API keys

Keys entered in the browser are kept only in browser memory; they aren't written to settings, evidence or browser storage, and you reconnect after a page refresh. Don't put the key in source code or image settings.

Advisor card

The small advisor card in the workbench links to the maker's external video advisor; the link carries no project data or keys. NIUBIGEO_VIDEO_ADVISOR_ENABLED=false in .env hides the card.

The workbench has no built-in login or authorization: if you'll open the server from another machine, put authentication and TLS in front of it; that's why the official commands expose the port only to your own machine (127.0.0.1).

If you're using an older image, pull v0.3.2 and recreate the container.

Turkish, and what NiubiGEO doesn't measure

Turkish interfaceNo	The interface is in English, Simplified Chinese and Portuguese (Brazil).
Measuring with Turkish keywordsUnclear	The docs don't describe language support; check the result against the raw answer.
The answer in consumer appsNo	What's measured is the API answer, not the answer in the ChatGPT or Gemini app. Real-interface testing is part of the maker's paid service.
Classic search engine rankingsNo	Google rank tracking is out of scope.

The docs clearly limit how results should be read: asking the same question again doesn't guarantee the same answer; a few short measurements don't prove a long-term trend; publishing an article doesn't guarantee AI will recommend you.

Frequently asked questions

What is NiubiGEO?

An open-source GEO tool that measures how AI answers describe your brand, which competitors it's mentioned with and which sources are cited. It runs on your own server with your own model API key.

Is NiubiGEO free?

The software is free and Apache-2.0 licensed. Model API, web search and server costs are yours; the maker also offers a paid diagnostic report and a testing service with real people.

Can I use it without code?

The open-source version needs a terminal and Docker or Node.js. For results without installing anything, the maker offers a diagnostic report for CNY 99 (about $15 at the TCMB rates of November 4, 2026); the site is in Chinese and partly English.

Which models does it measure?

Models on OpenRouter, providers' own APIs (OpenAI, Anthropic, Gemini, Perplexity, DeepSeek and others) and custom OpenAI-compatible endpoints. What's measured is the API answer, not the answer shown in the ChatGPT or Gemini app.

Does it support Turkish?

The interface isn't in Turkish: it's in English, Chinese and Portuguese. The docs don't describe language support; when measuring with Turkish keywords, check the result against the raw answer.

Related guides

Sources

Not clear in the official sources: Measurement quality with Turkish keywords · the status of a per-request budget limit in v0.3.2 (the known limitations document covers the v0.2 pre-release) · paying for the paid service from Turkey and the report's language · typical API cost per measurement (depends on model and frequency; no official figure). Each is covered in the relevant section along with what you can do.

You can do it yourself; if you’d like support

The guide is free and needs no sign-up; you can follow the steps here yourself. If you’d like professional support at some point, choose the option that fits your needs. This doesn’t affect the rankings in the guide.