# vortx.ai · run > A rung of https://vortx.ai/llms.txt: the six devices on the home page, each a published sample run end to end; every step is a request you can make. Verbs first: each line reads `verb noun ref key=value`. ## run (the home page's workflows: each step is a request you can make yourself; rows 0-based) satellite twlpco5kin6qlz5eplt2pjm7n4 Sentinel-2B scene, 351.0 MB GET note → base32(blake3(note)[0:16])==cid → rebuild root → GET https://emem.dev/v1/tree/?row=68 (level 4 tile 0,0) → Range read 601,342 B from sentinel-cogs → blake3==row → inflate, undo predictor 2 → mean RGB 156,137,113 at 98% valid == the row's stats telescope wkxa7tcmw2orf7ujjf5yi66dhe Webb, Cosmic Cliffs, 143.7 MB row 1 (strip 0 rows 0…4), 80.0 KB from esawebb.org robot qcoylkllqzfqnsinn4af5i2mi4 ALOHA arms, 502.5 MB MP4 row 3 (frames 5000…5001 · 100.0 s), 15.3 KB from huggingface.co (302 to its CDN) drone ejvovl6sz7d3sugfcrwwie4rma OpenAerialMap survey row 77 (level 6 tile 0,0), 13.2 KB; decode = the header row's JPEGTables + the tile camera t7ebh6s6imxrwdmizebnw52nwe 12 TfL cameras via geo.qa row aldgate: GET clip → sha256 and blake3 == the row → GET https://emem.dev/v1/perception/verify/clip/ → ed25519 over the canonical payload JSON (sorted keys, no spaces), key yQOKkNB+c+zl9kxxanwcNTZLgtHBsLvQH+q8pECrm/Q= → NOAA sun from the signed lat, lng, captured_at within 0.1° of the note machine emem:trace:mxyer5c2oxn4ud3xqbtnaxcxhdv4kkxiu67q32inbwxgpa7r3s6a POST /v1/trace_resolve → POST /v1/trace_verify (host.counters.v1 → admit) → chain, root and the device's ed25519, rechecked send Range a single `Range: bytes=a-b` and no other header: esawebb.org, esahubble.org, oin-hotosm-temp and mux.dev refuse CORS preflights fallback range_hash POST https://emem.dev/v1/range_hash {"url","offset","length"} when a source refuses browsers; signed, domain emem.range_hash.v1 count moved a token of 26 to 85 bytes instead of the file; each note costs ~1.2k to ~6.1k tokens to read, where the files cost ~91M to ~319M as raw bytes