AI Discoverability Dashboard

Static GitHub Pages dashboard that shows an AI/product discoverability scorecard for:

• Supabase
• Neon
• Pinecone
• Databricks
• Redis
• ClickHouse
• MongoDB
• Couchbase

The report is generated by a scheduled/manual GitHub Action using OpenAI. The report is encrypted using a password stored in GitHub Secrets. The dashboard decrypts locally in the browser.

Assessment Criteria

Each company is scored from 0 to 10 on each criterion. Higher scores mean the signal is easier for automated agents and LLM-based systems to discover, parse, and use. Weights reflect importance in the overall score.

• llms_txt (weight 3) This checks for a published llms.txt entrypoint and whether it is discoverable. A strong implementation includes clear product scope, canonical doc roots, and stable URLs.
• mcp_server (weight 3) This checks for an official MCP server (or equivalent agent interface) that exposes product capabilities to agents. A strong implementation has installation instructions, stable tools, and useful auth guidance.
• robots_txt_ai_optimization (weight 3) This checks whether robots.txt and related crawling signals are configured to be LLM and agent friendly. A strong implementation avoids blocking docs, includes sitemap references, and is consistent across doc hosts.
• llms_full_txt (weight 2) This checks for a published llms-full.txt (or equivalent) that is discoverable and aggregates important docs into a single, agent-friendly surface. A strong implementation keeps it current and scoped to developer-relevant content.
• markdown_native_docs (weight 2) This checks for documentation availability in Markdown or other easy-to-load formats. A strong implementation has stable, crawlable markdown sources or export endpoints.
• structured_faq_jsonld (weight 2) This checks for structured FAQ content using JSON-LD (or equivalent schema) on relevant pages. A strong implementation covers common developer questions and stays consistent with docs.
• html_parse_efficiency (weight 1) This checks whether key documentation pages are easy for automated parsers to extract. A strong implementation avoids heavy client-side rendering for core content and keeps DOM structure predictable.
• live_agent_environment (weight 1) This checks for sandbox, playground, or runnable environments that agents can use to validate workflows. A strong implementation offers predictable setup steps and programmatic access.
• training_data_surface (weight 1) This checks for broad public surfaces that reliably describe the product. A strong implementation includes clear, stable docs, changelogs, and reference content that stays online.

Methodology

The report is generated by a scheduled/manual GitHub Action.

For most criteria, the report is generated by prompting a model to evaluate each company against the criteria above. The model returns per-criterion analysis, evidence, and advice.

llms.txt / llms-full.txt experiments

For llms_txt and llms_full_txt, the pipeline runs two experiments and includes their results in the report:

• Control experiment (link-based discovery) The workflow fetches the company Website landing page and Docs landing page, extracts any links that look like llms.txt / llms-full.txt, and then fetches the discovered candidate URLs to record HTTP status and basic file stats.

• AI experiment (web search discovery) The workflow asks a model (via the OpenAI Responses API) to discover and assess llms.txt / llms-full.txt using the web_search_preview tool.

The scoring prompt is required to use the experiment results as evidence (specific artifact URLs, or an explicit not-found statement).

Scoring and Weighting

Each criterion is scored from 0 to 10. The overall total is a weighted average across criteria. Weights are defined in scripts/evaluate.mjs and are included in the report payload. The UI displays totals out of 10.

GitHub Secrets

Create these repo secrets:

• OPENAI_API_KEY (required)
• REPORT_PASSWORD (required) - used to encrypt the report in CI and decrypt in the browser
• OPENAI_MODEL (optional) - defaults to gpt-4o-mini
• OPENAI_RESPONSES_MODEL (optional) - model used for the AI web-search experiment; defaults to gpt-4o

GitHub Pages

Configure GitHub Pages to publish from:

• Branch: main
• Folder: /docs

Then your dashboard is available at your GitHub Pages URL.

Refreshing the Dashboard

• Automatic: runs on a schedule via .github/workflows/refresh-report.yml
• Manual: go to Actions -> Refresh AI Discoverability Report -> Run workflow

After the workflow completes, click Reload latest in the UI.

One-Time Setup Step

In docs/app.js, replace:

• https://github.com/<ORG_OR_USER>/<REPO>/actions/workflows/refresh-report.yml

with your actual repo workflow URL.