Sources and method

How it grows

Every AI, data, cloud, developer, software and security company in the fru.dev registry is in scope, and so are the well-known companies above whatever their category. Each week a slice of them is looked at for an official academy: the training links on the company's own home page and its academy, learn, university and training subdomains. An academy on a platform Training can read (Skilljar, Thought Industries, Docebo, Northpass) is read at once and its company joins the vendors; others are kept as sources until a reader exists.

Official sources only

Every course comes from the vendor’s own catalog: AWS Skill Builder, Google Skills, Microsoft Learn, Oracle MyLearn, Snowflake University, Trailhead, black-duck, IBM SkillsBuild, Okta Learning, GitLab University, ServiceNow University, MongoDB University, OpenAI Academy, Zscaler training, Databricks Academy, DeepLearning.AI, NVIDIA Deep Learning Institute, JFrog Academy, Fin Academy, Asana Academy, dbt Learn, Elastic Training, Meta Blueprint, LLM University, Cloudflare training, GitHub Skills, Confluent Training, Anthropic Academy, Cursor Learn, Hugging Face Learn, Atlassian Community Learning, Tenable University. No resellers, no review sites, no course marketplaces. Each row keeps two links: the official course page and the source the facts were read from (often the same page; for Microsoft, the public Catalog API), plus a short quote of what the source says, on the course page under About.

The full catalogs

Besides the courses checked by hand, Training reads every item each academy lists, from the source its own catalog page uses: Snowflake University’s course page and the Snowflake developer guides (the quickstarts, listed from Snowflake’s site map because robots.txt keeps crawlers off the page’s own data endpoint), the Databricks training catalog’s data file, the Google Skills catalog, the AWS Skill Builder search, the Microsoft Learn Catalog API (kept to the Artificial intelligence and Data management subjects), DeepLearning.AI’s course list, Hugging Face Learn, the dbt Learn and MongoDB University catalogs, Confluent Developer’s courses, Anthropic Academy and OpenAI Academy. All are public, free and read without a key, as TrainingBot, at a polite pace, after checking each host’s robots.txt. NVIDIA’s catalog API asks crawlers to stay away, so NVIDIA’s list comes from the Internet Archive’s copies of it and is as fresh as the latest copy. Read the same way: Salesforce Trailhead from its content site map (each new item’s title and length from its own page’s schema.org data), IBM SkillsBuild’s catalog, Oracle MyLearn’s catalog search (with the anonymous guest session the public site gives every visitor), GitHub Skills, Elastic training, Atlassian Community Learning, ServiceNow University’s site map, Cohere’s LLM University, Cursor Learn and Meta Blueprint. Where an academy covers far more than data and AI (Salesforce, IBM, Oracle, ServiceNow, Atlassian, Meta), only the items the area rules file under data and AI are kept. Academies on Docebo (when the academy publishes its public catalog), Northpass (public schools), Skilljar and Thought Industries are read by one reader per platform, so a new academy on one of them needs no code. A host that answers 401, 403 or 429 is not asked again that run, and nothing behind a sign-in is read.

Counts are English items listed today; translations are counted apart. Databricks lists most courses up to three times (free, paid with labs, instructor-led); the self-paced offerings count as one course and the instructor-led one as a live class. Each item is filed into one or more areas (generative AI, agents, data engineering, streaming and so on) by keyword rules over its title and the vendor’s own tags or subjects. That is a transparent heuristic, not a hand review: an item no rule matches is filed under Other. On Compare, “courses and classes only” leaves out quickstart guides, videos and short Microsoft modules.

History from the archive

When an item first appeared is the vendor’s own date where it gives one (DeepLearning.AI release dates, dbt and MongoDB publish dates), else the first Internet Archive copy of its page, else Training’s own first read; the earliest of these wins. The archive is searched by every URL pattern a vendor has used, old platforms included (Qwiklabs, docs.microsoft.com/learn, explore.skillbuilder.aws, the first Snowflake University, courses.getdbt.com, university.mongodb.com), sampled year by year. An archived item that is no longer listed counts as retired in the last year the archive saw its page. Archive coverage is thin in the earliest years, so they undercount; Microsoft and AWS items no longer listed are kept only when their title reads as data or AI.

Fields

• Format: self-paced (on demand) or instructor-led (a live class, in person or virtual).
• Length: as the vendor states it. Some vendors state a weekly pace instead of a total (Hugging Face); some state none (Anthropic Academy). For the length strip a class day counts as 8 hours.
• Price: Free only when the vendor says the course is free. A course that is free to take but charges for a certificate is marked free, with the certificate price in the price note. Microsoft instructor-led courses are sold by Microsoft Learning Partners at their own prices, so none is shown.
• Level: the vendor’s own word, mapped to beginner, intermediate or advanced; blank when the vendor does not say.
• Prepares for: only when the vendor ties the course to a certification (Microsoft course pages list “Related certifications”; others say so in the course or learning path). Certification details, exams and prices are on the fru.dev certifications directory.

The weekly check

Every Wednesday at 10:20 UTC a job re-reads the Microsoft Learn Catalog API and the tracked Microsoft course pages, requests every course’s official page, and reads a third of the AWS Skill Builder catalog (its search returns 100 results a query, so the large slices rotate over three weeks). At 10:50 UTC a second job re-reads every other academy’s full catalog; the three slowest (Trailhead, Oracle MyLearn and Meta Blueprint) take turns, one a week. New items, retired items and changed titles, formats, levels, lengths or prices are appended to the catalog log; an item counts as retired once a healthy read has missed it for three weeks (five for AWS and the academies that take turns). New Microsoft AI and data instructor-led courses are added as Unverified until checked by hand. A course whose page answers “not found” on two checks in a row is marked retired. The job calls no AI model and no search API.

History is kept

Nothing is overwritten silently. Each change to a price, free status, length, level, format, certification or status is added to the change log with the date and the source, and retired courses stay listed, marked retired. See Changes, or /api/changes.

Across fru.dev

Vendor names and slugs follow the shared registry at companies.fru.dev, and /api/companies lists every vendor with its dated items so the registry can build a cross-site timeline. Certifications link to the fru.dev certifications directory.

Disclaimer

Courses from each vendor's own training catalog, re-read every Wednesday; new Microsoft courses the scan finds are marked unverified until checked by hand. The list grows as vendors publish new courses, and new vendors join as their official academies are found. Prices and schedules change, so confirm on the course page before you enroll. Logos via logo.dev; trademarks belong to their owners. Anonymous usage stats via Google Analytics. Use Suggest a correction on any page.