GEOInflow is a self-hosted analytics tool for AI crawler traffic

An SEO team that wants to see which pages ChatGPT and Perplexity read on their site has mostly paid tools. GEOInflow is a new open option they can run on their own server.

The repository was published by 猎看科技 and 北京日报社智源GEO. It picked up 114 stars in its first day on GitHub. The licence is MIT.

What it tracks

The built-in catalogue identifies crawlers from OpenAI, Anthropic, Perplexity, Google and Microsoft. It also covers ByteDance, Baidu, Kimi and Tencent Hunyuan. Traditional search bots are included.

The project's description says it reads the crawler's identity. It records the request volume and the HTTP response the site sent back. A page the AI bot could not fetch shows up as an access problem for the team to investigate.

The site data runs beside the crawler data. The tool stores page views, events and conversions. It labels sessions that arrived through an AI answer as AI referrals. The project's description says this ties an AI mention to a sale.

What it needs to run

The deployment script in the repository uses Docker. The services it starts are a Node.js app on port 3002, a PostgreSQL database and a Redis cache. The default sign-in is admin / liekan. The project tells the operator to change it.

The repository names its current build as version 1.2.0. The project text is in English and Simplified Chinese. A Cloudflare integration is listed for sites behind that network.

What a user can do with it

The content coverage feature imports a sitemap. It reports which pages the AI bots have fetched. A page the bots never read is a page an AI answer will never cite.

The access investigation feature breaks the fetch down by latency, cache status and HTTP error code. A slow page or a 404 becomes visible to the team.

For a smaller company, running this on a single server is a path to a view of AI search traffic. That view usually sits inside a vendor dashboard. For a company whose data must stay on its own servers, a self-hosted option is the only path.

The caveats sit in the project itself. It is one week old at version 1.2. Both publishers describe their work as a joint release. The project is small enough that a single operator can read every file. Measuring AI crawler traffic also changes as the vendors rename their bots and alter headers. The catalogue of recognised bots is the part that goes stale fastest.

Source

yaojingang/GEOInflow on GitHub.