Albert-Weasker/niubigeo

What it does

NiubiGEO is a self-hosted research workbench for GEO (generative engine optimization). You create a project for a domain, pick one or more models, and record what each model says about the product: its description, the competitors it names, related keywords, and the sources it returns. Every result keeps the original answer, the execution conditions and any failures, so a visibility observation can be traced back to the response that produced it.

The v0.3 release adds a separate Keyword Monitoring module: define or batch-import prompts, classify them (Discovery, Alternative, Comparison, Brand, Use case), run them on an hourly, daily or weekly schedule, and compare stored runs in a Diff View that shows added or removed recommendations, position changes and source links. It is aimed at product, marketing and developer teams who want raw evidence rather than an opaque score.

Quick start

To try the workflow without installing anything, read the 20 recorded cases. To test your own product you need Node.js 22.13+ and your own OpenRouter API key.

git clone --branch v0.3.3 --depth 1 https://github.com/Albert-Weasker/niubigeo.git
cd niubigeo
npm ci
cp .env.example .env

Set OPENROUTER_API_KEY in .env, then run:

npm run server

Open http://localhost:8787, create a project with your domain, select models and set web search per model, save the configuration and start a test. After results come back, open each answer to inspect descriptions, competitors, keywords and citations. A Docker guide covers container deployment, and scheduled monitoring requires the monitoring worker to be running.

Constraints

• Model, search and hosting costs are separate from the open-source license; tests consume your provider credits.
• Only Provider API responses are observed, not consumer chat interfaces. Offline and web-search-enabled runs must be interpreted separately, and traditional search-engine rank tracking is not included.
• Some features depend on provider capabilities: structured analysis needs JSON Schema support, and custom search support is labeled unverified.
• The current v0.3 scope has no separate competitor detection dashboard; independent history is kept when the model selection changes.
• README links an optional advisor card that opens an external video service; it can be disabled with NIUBIGEO_VIDEO_ADVISOR_ENABLED=false.
• Licensed under Apache-2.0.

HotRepo's take

On 2026-11-08 the repository recorded 1,307 stars gained in a single day on GitHub's TypeScript daily list, placing 4th there, with 10,138 stars and 815 forks overall. That is a strong single-snapshot signal for a tool published in September 2026 and already at v0.3.3. The distinguishing design choice is placing original model answers and citations ahead of an aggregate visibility number, with self-hosting and bring-your-own-key as the default; whether the keyword-monitoring beta holds up under real schedules is the part worth watching across future releases.