The whole table is a dataset. Take it.

Everything on the table — every element, group, seat coordinate, leading tool, price band, maturity phase, and editorial take — ships as open, versioned JSON under CC BY 4.0. Use it in your product, your research, your agent. The only ask is the license's: cite elems.ai.

The surfaces

elements.json — the canonical dataset, current edition, regenerated with every release.
editions/2026-q3.json — dated snapshots, one per edition, never rewritten. Diff them to see every succession.
llms-full.txt — the full taxonomy as plain markdown, one fetch, token-cheap.
llms.txt — the machine-readable site summary.
Markdown twins — every element page has a .md twin: e/ca.md, e/wf.md, and so on for all 58.

Element dossiers — one file per category

Every element page has a machine-readable dossier at /e/<symbol>.json: the ranked top five with vendor-verified pricing and dated evidence, the comparison matrix, the decision rules, and the complete field including the dead and renamed. Fifty-eight files — e/ca.json, e/wf.json, and so on for every symbol on the table. This is the part nobody else publishes: a category's whole population, graveyard included, with the evidence attached.

Reading the price bands

Every element carries a price band, and the dataset spells out what each one means: Free is $0, $ is under $30 a month, $$ is $30–150, $$$ is $150 and up — per seat, at typical solo or small-team plans, as of the current edition. The bands ship as price_band_usd_per_month so nothing has to be inferred from a glyph.

For agents: the MCP server

The taxonomy ships as a read-only MCP server — mcp.elems.ai — so Claude, ChatGPT, Cursor, and anything else that speaks the protocol can query elements, compare them, and generate a recommended stack from five answers, without scraping a single page. Five tools: list_elements, get_element, compare_elements, search_taxonomy, recommend_stack.

Embedding the table

The table travels. Drop it into a post, a cohort wiki, or an internal doc — attribution is built in:

<iframe src="https://elems.ai/#table"
  title="elems — the periodic table of the AI-led startup"
  width="100%" height="720" loading="lazy"
  style="border:1px solid #ddd"></iframe>

Citing elems

In prose: "elems.ai, the periodic table of the AI-led startup, edition v2026.Q3." In data: keep the edition and updated fields attached to anything you derive — undated stack advice ages like milk.

What changes between editions

Elements are permanent; the tools that hold them aren't. Each quarterly edition records successions (a tool losing its seat), discoveries (new elements earning one), and phase changes (maturity shifts) — all in the changelog, all diffable from the archived editions.

Questions, corrections, or a use worth showing off: [email protected]