Type plain English — "village in plains with a ruined portal within 200 blocks, no oceans or deserts within 500" — and the AI turns it into a structured multi-condition filter. A pool of WebAssembly workers then sweeps millions of seeds in your browser until they match. No DSL to learn, no schema to memorize. Try a free demo on the home page first — every demo is a real recorded search, no signup needed.
"village in plains with a ruined portal within 200 blocks and no oceans within 500 blocks." No DSL to learn, no query syntax to memorize. If you can describe it, you can search for it.
A strict JSON schema enforces the shape — each condition is a structure check, biome inclusion, biome exclusion, or distance constraint, and any condition can be anchored relative to another. The AI handles the ambiguity so you don't have to.
A pool of Web Workers — one per CPU core — runs a WebAssembly build of Minecraft's worldgen in parallel, evaluating the filter against thousands of seeds per second. Matches stream back as they're found — you don't wait in a queue, you don't wait on a server.
The map centers on the matching anchor, switches dimension if the structure lives in the Nether or End, and highlights the biomes that satisfied the filter. Save the query so you can re-run it later — for free.
Most multi-condition seed tools make you learn a query language or click through a filter UI. Type what you want — the AI does the structured-filter part for you.
Stack structures, biomes, exclusions, and distances in one search. Anchor any condition to another — "portal within 200 blocks of the village" measured from the village, not from spawn.
Once a query is translated, save it. Loading a saved query bypasses the AI step entirely — no credit deduction, just a fresh sweep with the cached filter. You only pay for each unique query once.
Every demo on the home page is a pre-recorded real search — same flow, same timing, same matching seeds. Watch the whole pipeline end-to-end before deciding whether to buy credits.
The WASM worldgen, the worker pool, and the map renderer all run in your browser. The only network call is the natural-language translate — which is also the only step that costs anything.
Minecraft Java 1.16 through 26.1 (current Drop 26), including the 1.21.5+ Pale Garden expansion. See "Version support" below for the placement-accuracy specifics per release.
Demos are free — no signup required, no card on file. When you're ready to run your own queries, sign up and buy a credit pack at /buy. Each natural-language translate costs 1 credit; the seed search that follows is always free (it runs locally in your browser, so it costs us nothing). Saved queries replay for free too — you pay once for the AI to translate a query, and after that you can re-run it forever.
Minecraft Java edition 1.16 through 26.1 (the current Drop 26 release). The underlying worldgen library only lists versions where worldgen changed, so each dropdown entry covers all patches up to the next entry — "1.20" means "1.20 through 1.20.4".
1.21.4 vs 1.21.5+ is a real placement change. Mojang expanded the Pale Garden biome at snapshot 25w02a (the 1.21.5 cycle), dropping its weirdness lower-bound from 2666 to IMIN — it now consumes territory previously assigned to dark_forest. The Mojang Wiki confirms it ("Pale Garden biome now occupies more space that was previously taken by dark forests"). We pull in the new biome tree (tables/btree215.h) so 1.21.5+ gets accurate Pale Garden placement, not 1.21.4's pre-expansion version.
1.21.5 → 26.1 are placement-identical. Cross-referenced Mojang's release notes, third-party per-snapshot logs, and the byte-identicalmulti_noise_biome_source_parameter_list JSONs across these versions. Per-biome JSONs and the overworld noise router changed in 1.21.9 / 1.21.11 / 26.1, but only for mob spawns, surface decorations, lava aquifers, and preliminary-surface terrain — none of which affect biome IDs or structure positions. So we serve them all from one slot.
Other caveats. On 1.18+, Desert Pyramids, Jungle Pyramids, and Woodland Mansions can have false-positive placements because the underlying worldgen library only approximates terrain height — same limitation every other browser-based seed tool has. For multi-piece structures like Nether Fortresses, the marker shows the region center; the real structure can extend ~150 blocks beyond — an "approximate extent" ring surrounds the anchor when you jump to a result.