NiubiGEO Open-Source Tool Lets Developers Audit How AI Models Describe Brands

·2 views

NiubiGEO has released an Apache-2.0 workbench that tests how AI models describe a product, which competitors they surface, and what sources they cite. Users configure one or more OpenRouter models and optional web search per model, with each model running independently so a single failure does not invalidate other results. The tool distinguishes between prompted brand recognition and unprompted mentions, and separately classifies a mention, a positive description, and an explicit recommendation. Results include original model answers, provider citations, and URLs, though the project cautions that a citation does not explain why a model recommended a product and that publishing content does not guarantee an AI mention. Self-hosting requires Node.js 22 or later and an OpenRouter API key, while model and search API usage incurs charges; the project also offers paid human-led AI testing and GEO optimization services separately.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

MAGES: The .NET Scripting Engine Born from a Game and a Thiel Grant

Developer Florian Rappl details the origin story of MAGES, a lightweight expression parser and interpreter for .NET, in his ongoing open-source project series. MAGES evolved from an earlier project called YAMP, a reflection-based math parser that powered SineRider, a 2013 Unity web game where players graph equations to navigate a sled through checkpoints. When YAMP's performance proved insufficient for a more ambitious real-time, function-driven game world envisioned by SineRider's creator, a Peter Thiel grant helped fund the development of a faster successor. The result was MAGES, which retained YAMP's MATLAB-like syntax but replaced its reflection-driven internals with a more performant architecture suited for real-time use. MAGES is available as an embeddable scripting engine for .NET applications, requiring minimal setup and no heavy runtime dependencies.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Ubuntu's Node 22.22 silently disables TypeScript stripping, breaking tests

Ubuntu's packaged Node.js 22.22.1 has TypeScript type-stripping disabled because the package is built without Amaro, the dependency that enables the feature introduced in Node 22.18.0. A developer discovered this when four TypeScript-importing test files passed in CI — which uses the official nodejs.org binary — but failed on their Ubuntu server with the same version number. Running `node -p 'process.features.typescript'` reveals the discrepancy: the official build returns 'strip', while Ubuntu's returns false. The author resolved the conflict by installing the official Node.js tarball into /usr/local/bin, giving it PATH priority over the Ubuntu package without removing it. The key takeaway is that `node --version` alone cannot confirm whether type stripping is available; only `process.features.typescript` reliably reflects the build's actual capabilities.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

HyperLogLog: How a 12KB Algorithm Can Count 100 Billion Unique Items

Counting billions of unique visitors with a hash set is impractical because memory usage grows linearly, potentially demanding terabytes of RAM. A probabilistic approach called HyperLogLog, rooted in Flajolet and Martin's 1985 paper and refined in a 2007 follow-up, solves this by trading exact accuracy for extreme memory efficiency. The algorithm works by hashing each incoming ID and tracking the longest streak of leading zero bits seen, using that maximum as a statistical estimator of unique item count. A single such counter is highly volatile on its own, but HyperLogLog combines many counters and averages their results to dramatically reduce error. The result is a data structure that can estimate cardinality across hundreds of billions of items while fitting comfortably within a few kilobytes of memory.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Seven Cloudflare Settings That Silently Blocked Paying API Agents, Developers Warn

A developer team at ForgeMesh discovered that several default Cloudflare security settings had been silently returning 403 errors to legitimate AI agents attempting to access their paid API endpoints, instead of the expected 402 payment-required responses. The issue was uncovered when an AI agent named Coppice flagged that Python's standard library and Perl's HTTP client were being blocked, while curl and Node requests passed through normally. Cloudflare's Browser Integrity Check, enabled by default, rejects requests from User-Agents like Python-urllib and libwww-perl, and because the block happens at the edge, no trace appears in origin server logs. The team found 13 paid hostnames across three zones had been affected this way since the zones were created, meaning potential customers never received pricing information and simply never converted. Beyond Browser Integrity Check, the article identifies six other Cloudflare features — including Bot Fight Mode, challenge-based WAF rules, and the updated AI Crawl Control — that can similarly obstruct machine clients on API endpoints.

0 comments Read more at DEV Community