PartsPlan finds what an AI needs in the manufacturer's datasheet — pages, tables, graphs, drawings and condition sentences — and hands it over with evidence. On top of that, the AI does circuit design, PCB work and part review. Each item below carries a result we tested ourselves.
From choosing parts to a circuit that actually works, it keeps track of conditions scattered across the datasheet.
Helps the AI apply conditions buried inside each datasheet — start-up voltage, supply current, monitoring limits — citing the relevant pages.
Test resultPortable pump board design — major errors 0 vs 3, 25.6 vs 63.7 min (the side without PartsPlan also had web, shell and KiCad) →Toolscheck_bomsearch_datasheetget_page_blocks
Printed ordering and restriction sentences ("must … before", "only these commands can be issued when …") are shown with their page the first time a part is looked up. A rule to follow and the restriction that backs it, on the same signal, are paired.
Test resultIn a web environment like Claude.ai or ChatGPT (no shell), kept the ADC's START ordering rule (the web-only AI missed it) →Toolsfind_datasheetcheck_bomsearch_datasheet
Looks up every line of a schematic or BOM at once: whether the manufacturer's datasheet is held, whether it is readable, and which pages to open first (pin table, absolute maximum ratings, recommended operating conditions). The AI opens those pages and flags rating violations.
Test resultAll 6 rating violations hidden in a 30-part BOM found in 1.9 min (PartsPlan only · web search: 12.2 min) →Toolscheck_bom
For values with conditions such as "at 85 °C and 400 mA", shows the printed row together with its conditions and footnotes, so a similar value on the next row is not taken by mistake.
Test resultValues deep inside 30–330-page datasheets, with the printed page: 8 / 8 (web search alone 6 / 8) →Toolssearch_datasheetget_page_blocks
Opens the datasheet pages that placement, routing and footprints need.
Finds the pages that print placement requirements — where each supply pin's capacitor goes, the analog supply filter, crystal conditions.
Test resultPlacing 49 parts of an STM32 board — ahead in all three runs, average 92.6 vs 81.7, 21.4 vs 32.2 min (the side without PartsPlan also had web, shell and KiCad) →Toolsget_datasheet_mapsearch_datasheetget_page_blocks
Points to the page and block of placement and routing instructions — Kelvin connections, ground planes, power loops, "as close as possible" — including those printed outside the layout section, in pin descriptions and application sections.
Toolsfind_datasheetcheck_bomsearch_datasheet
Lists every package outline and recommended land pattern page in the document and lets the AI read the dimensions from the page image — from the archived original even when the manufacturer's site does not respond.
Test resultFootprints for six parts — 10.1 vs 24.8 min (2.5× faster), 59.3 vs 58.3 points (the side without PartsPlan also had web, shell and KiCad) →Toolsfind_datasheetget_page_image
Gives the page, caption and region of the application circuits the manufacturer printed. The AI reads the circuit by opening the figure.
Toolsreference_circuitsget_page_image
Not only text tables, but graphs, drawings and documents in other languages.
Returns values that exist only in graphs — temperature curves, loss curves, SOA — as numbers, only when certain. It can also produce model data for circuit simulation.
Test result30 graph-only values — 30 / 30 in all six repeats (within 3% or 0.05 V) →Toolsget_figure_data
When a similar but wrong value sits right next to the answer — another temperature's curve, the next row's condition — it hands over the source and curve values so the AI picks the requested condition.
Test result12 hard design questions — 12 / 12 in all four runs →Toolsget_figure_datasearch_datasheetget_page_image
Checks pin numbers and functions, and drawing dimensions such as lead width, from the page image.
Test resultPinouts and package drawings of 8 parts — 8 / 8 in 1.3 min →Toolsfind_datasheetget_page_image
Finds and reads Chinese-language datasheets of parts from Chinese manufacturers (for example, searching 引脚 pin descriptions).
Test result10 questions on 6 parts from Chinese manufacturers — 10 / 10 (PartsPlan only · web search 8 / 10) →Toolssearch_datasheetget_page_blocks
Gives a signed link to the archived original PDF (a hash-verified copy). An AI with a shell can search the whole document and see conditions spread across several sections at once.
Toolsfind_datasheetsearch_datasheetget_page_blocks
Helps pick the right document and the right part.
Finds datasheets by part number and separates the requested manufacturer's document from other manufacturers'. For similar part numbers it shows whether they look like ordering variants or may be different devices.
Toolsfind_datasheet
For a source part and up to three candidates you picked, points out the items that must match, the weights and the items to read, with the page in each datasheet where each item is printed. Adjust the criteria to the purpose (battery, low noise and so on) and the situation (keep the board or redesign), and the AI opens those pages to compare the originals side by side. No candidate suggestions, scores or compatibility verdicts.
Test resultOne alternate — both pin tables compared from the originals in 45 s (PartsPlan only · web search 3.1 min) →Toolscompare_pointsfind_datasheetget_page_image
Browse with questions like "all of this maker's MOSFETs" or "makers of Hall sensors" — only parts printed as that type in the manufacturer's datasheet, with the printed page as evidence.
Toolssearch_parts
To use an answer in a design, you need to see where it came from.
Every piece of evidence carries the manufacturer's document (hash), page and block. The AI answers from the opened source, not from memory.
Test resultAddresses of 5 I²C parts — evidence page 3 / 3 (web search alone 0 / 3) →Toolsget_page_blocksget_page_image
Graph values are returned as numbers rather than eyeballed, so asking the same question again gets an answer from the same evidence.
Test resultMOSFET short-circuit timer limit — the same answer, "about 1.8 ms", all three times (web search alone gave three different answers) →Toolsget_figure_data
Values that could not be confirmed are marked as unconfirmed. Asked for a value the datasheet does not have, the AI says so.
Test result8 values the datasheets do not contain, mixed into the questions — 0 invented →Toolssearch_datasheetget_page_blocks
BOM lines we do not hold can be queued for datasheet collection, and an answer that does not match the source can be reported with its document, page and block.
Toolsrequest_datasheet_collectionreport_problem
