7 Facts About OneGlanse, the Open-Source GEO Tracker

OneGlanse is a free, open-source GEO tracker under the MIT license that monitors how your brand appears in ChatGPT, Gemini, Perplexity, Claude and Google AI Overview. It runs only on your own machine or server, captures answers from the real chat interfaces through browser automation instead of model APIs, and uses your own OpenAI or Anthropic key to score them. "Free" means no licence fee; you still pay for the analysis API calls, hosting and, on a server, a residential proxy.

All facts below come from the project's GitHub repository and README and its public website, as read on 28 September 2026. We have not run OneGlanse ourselves, so we describe what the project documents and list what you should verify before relying on it. (An earlier version of this article described OneGlanse as a location-tracking tool with paid tiers; that was wrong and has been replaced.)

1. It is open source and self-hosted only

OneGlanse is published under the MIT license. The project says there is no subscription and no usage limit, and it offers no hosted cloud version. Responses, analytics and login sessions are stored in a PostgreSQL and ClickHouse instance that you run locally via Docker or on your own VPS.

The project explains the missing cloud version with its measurement approach: it depends on logged-in browser sessions tied to your own provider accounts, and it does not want to operate shared sessions for customers. For you this means full control over the data, and full responsibility for running the stack.

2. It tracks five AI surfaces through their real interfaces

Supported providers are ChatGPT, Gemini, Perplexity, Claude and Google AI Overview. OneGlanse does not call the model APIs to generate answers. It opens the consumer interfaces in a browser, logged in with your accounts, and captures what is rendered: the full answer, inline citations, source cards and the order of recommendations.

The project's reasoning is that interface answers can differ from raw API completions, for example in which citations are shown. That distinction matters for any AI visibility measurement. Google, for instance, documents its AI features in Search as a Search product with its own source links, which a plain model API call does not reproduce.

3. Scoring runs on your own OpenAI or Anthropic key

After a prompt run, the captured answers are sent to OpenAI or Anthropic, using a key you provide. That model extracts the metrics shown in the dashboard: GEO score, visibility, sentiment, rank position, recommendation type, competitor mentions and cited sources. The README states the call goes directly from your infrastructure to the provider.

Two consequences: every analysed answer creates API costs on your account, and the scores are LLM judgements, not deterministic counts. The analysis prompt is published in the repository, so you can read exactly what the scoring model is told.

4. The scoring model is documented

The README describes the headline metric in detail:

• GEO Score (0 to 100): an equal-weighted average of visibility, rank, sentiment and recommendation (25% each).
• Visibility (0 to 100): coverage, placement, structural prominence, frequency and contextual framing of the brand in the answer.
• Sentiment (0 to 100): 50 is neutral. A brand that is not mentioned scores 50, so absence counts as neutral, not negative.
• Recommendation type: from top pick and strong alternative to mentioned only, discouraged or not mentioned.
• Rank position: the absolute position of the brand in the reading order of the whole answer.

The analysis model is instructed to quote the passage that justifies each score. Published formulas are a plus; check whether any commercial alternative you consider documents its composite score as clearly. Keep in mind that a neutral sentiment of 50 for "not mentioned" can lift averages, so read visibility and mention counts alongside the score.

5. Setup needs developer skills

The documented requirements are Node.js 20 or newer, pnpm 10 or newer and Docker. A single command starts the app locally and brings up PostgreSQL, ClickHouse and Redis. The web app is built with Next.js, the browser worker uses Playwright with Camoufox (a Firefox-based browser designed to reduce automation fingerprints), and jobs run through a Redis queue.

The README states that WSL on Windows is not supported for the browser automation; you need native macOS, Linux or Windows. After signing up in the local app you connect your AI provider accounts and add prompts.

6. On a VPS it requires a residential proxy

According to the project, AI chat products often challenge or block traffic from datacenter IP ranges. For self-hosted runs on a VPS, OneGlanse therefore requires a residential proxy, and the docs describe the setup for one proxy provider. Local runs from a home or office connection usually work without one.

This is the biggest hidden cost of "free": a residential proxy is a paid service, and scheduled runs from a server depend on it.

7. It sends minimal telemetry

OneGlanse collects anonymous usage telemetry via PostHog: a one-way SHA-256 hash of the internal user ID, the event type (sign-up or active) and a timestamp. According to the README, no email addresses, IP addresses, prompts, answers or scores are sent. If you run it inside a company, include this in your data protection review.

Before you run it: a checklist

• Terms of use: read each AI provider's terms on automated access to its consumer interface and decide whether your organisation accepts the risk. OneGlanse uses your own accounts, so the accounts are yours to lose.
• Accounts: use dedicated accounts per provider, not personal ones, and decide which plan level you measure with and keep it constant across runs.
• Costs: estimate API costs for analysis (prompts × providers × runs), plus hosting and, on a VPS, the proxy.
• Market: answers depend on location and language. Check which country your browser traffic or proxy exits from and match it to your buyers.
• Repeated runs: AI answers vary between runs. Schedule each prompt several times before you treat a change as real.
• Spot checks: read ten captured answers by hand and compare them with the scores the analysis model assigned.
• Maintenance: interface automation can break when a provider changes its UI. Check the repository's recent commits and open issues before you depend on it.

When OneGlanse fits, and when it does not

OneGlanse suits teams with a developer who can run Docker, maintain browser sessions and read TypeScript, and who want raw data in their own database with a transparent scoring prompt. It also suits anyone who wants to study how interface answers differ from API answers.

It fits less well if nobody on the team can operate a self-hosted stack, if you need a vendor contract and support, or if your organisation cannot accept automated use of consumer AI accounts. In those cases, compare hosted monitoring tools using the criteria in our GEO tool comparison 2026 and our guide to choosing a GEO monitoring dashboard.

Monitoring shows what AI systems say today. It does not tell you why your pages are or are not cited. Research on generative engine optimisation, such as Aggarwal et al. (2023), looks at how content changes affect visibility in generated answers. To check whether a page is technically and structurally ready to be read and cited, run the free AI visibility checker.

Frequently asked questions

What is OneGlanse?

OneGlanse is an open-source, self-hosted GEO tracker that records how your brand appears in ChatGPT, Gemini, Perplexity, Claude and Google AI Overview answers and scores those answers with an LLM of your choice.

Is OneGlanse really free?

The software is free under the MIT license, with no subscription or usage limit according to the project. You pay for the OpenAI or Anthropic API calls used for scoring, your hosting and, on a VPS, a residential proxy.

Does OneGlanse have a cloud version?

No. The project says it deliberately offers no hosted version because measurement depends on your own logged-in provider accounts and browser sessions.

Which AI platforms does OneGlanse monitor?

ChatGPT, Gemini, Perplexity, Claude and Google AI Overview, captured through their web interfaces rather than model APIs.

How is the OneGlanse GEO score calculated?

It is the equal-weighted average of four components: visibility, rank, sentiment and recommendation, each 25%. An LLM derives each component from the captured answer text using a prompt published in the repository.

Does OneGlanse collect my data?

Prompts, answers and scores stay in your own databases. The project sends anonymous telemetry to PostHog consisting of a hashed user ID, an event type and a timestamp.