Component checks for AI-assisted circuit design

Tested with the same AI:
more accurate than attaching the PDF directly.

Your AI picks the parts. PartsPlan hands it the evidence for each value, with the page in the manufacturer's original.

Start free Connect to Claude ↗Connect to ChatGPT →

Free $0 · 300 tool calls a month · no card needed · Personal $19/month plus applicable sales tax / VAT · Companies & organizations: by agreement — See pricing

PartsPlan is a hosted developer API (MCP server). Subscriptions pay for API usage. We are not a marketplace and don't sell parts or documents.

We don't hand your AI a PDF. We hand it evidence, already organized.

ChatGPT conversation. The answer shows the dropout-voltage row (1 A, TYP 307 mV, MAX 450 mV) from the electrical characteristics table on page 6 of the TPS7A4701 datasheet, with a button that opens page 6 of the original. — A real ChatGPT session using PartsPlan · recorded October 2026

Datasheet original we checked it against · Texas Instruments · TPS7A4700/TPS7A4701 (SBVS204G) · page 6Dropout voltage V(DO)

⋯ Rows in between omitted ⋯

Header row and the two dropout-voltage rows (0.5 A 216 mV; 1 A 307 mV, max 450 mV) of the electrical characteristics table on page 6 of the TPS7A4701 datasheet

Row the answer cited · highlighted by us

Rows in between and the MIN column are omitted. Enlarge to see all of page 6.

Open the manufacturer's original ↗

Q: Where is the TPS7A4701 dropout voltage at 1 A? → A: page 6 table, TYP 307 mV · MAX 450 mV

Small tests with the same AI

30 values that exist only in graphs →

Conduction voltage, switching losses and diode voltage of a power IGBT (IKW40N120CS6) — printed only in graphs, not in tables

30 / 30

PartsPlan connected · all 6 runs

25 / 30

PDF attached directly · 1 run

Same AI · one part (16 pages) · within 3% or 0.05 V counts as correct · September 2026

12 hard questions, run again →

4 deep-table, 5 graph and 3 package-drawing questions — each with a similar wrong value right next to the answer

12 / 12

PartsPlan connected · all 4 runs

10–11 / 12

PDF attached directly · 3 runs: 10, 11, 10

Same AI · tables, graphs and drawings · September 2026

The same work, faster →

Drawing a KiCad footprint and routing a 2-layer board — same AI, same task

10.1 min

Footprints · PartsPlan connected

24.8 min

Without PartsPlan

10.6 min

PCB routing · PartsPlan connected

21.3 min

Without PartsPlan

3-run averages · compared with no PartsPlan (web, shell and KiCad) · routing quality score: both 20 out of 20 (final run) · October 2026

See the 12 questions and answers

Table · STM32F103C8T6 — max Run-mode current with code running from RAM (36 MHz, peripherals on) at 105 °C: 25.5 mA (p. 41) · trap: 28.7 mA in the Flash table just above
Table · STM32F103C8T6 — ΘJA of the package its ordering code implies (T = LQFP48): 55 °C/W (p. 101) · trap: 32 °C/W for the other 48-pin package, UFQFPN48
Table · STM32F103C8T6 — max Stop-mode current (regulator in low-power mode) at 105 °C: 340 µA (p. 44) · trap: 370 µA in the next row (regulator in Run mode)
Table · STM32F103C8T6 — max external source impedance RAIN at fADC 14 MHz with 13.5 sampling cycles: 11.4 kΩ (p. 75) · trap: the general 50 kΩ on the previous page
Graph · LM358DR — typical output source current limit at 70 °C: 29.4 mA (p. 23) · trap: 38.6 mA from the LM358B curve in the same document
Graph · AP2112K-3.3 — dropout at 400 mA and 85 °C: 200 mV (p. 15) · trap: 171 mV on the neighbouring 25 °C curve
Graph · BSS138 — RDS(on) at VGS 3.3 V, ID 110 mA and 125 °C: 1.95 Ω (p. 4) · trap: 1.19 Ω on the 25 °C curve
Graph · BSS138 — body-diode VSD at IS 10 mA and 25 °C: 0.646 V (p. 4) · trap: 0.47 V on the 125 °C curve
Graph · BSS138 — drain current at VDS 10 V, VGS 1.75 V and 125 °C: 0.117 A (p. 4) · trap: 0.076 A on the 25 °C curve
Drawing · STM32F103C8T6 — max LQFP48 lead width b: 0.27 mm (p. 99) · trap: 0.23 mm for b1 in the next row
Drawing · SS34 — max lead width in the top view: 3.27 mm (p. 4) · trap: the 3.39 mm land-pattern pad right beside it
Drawing · AP2112K-3.3 — EN pin and pin 4: EN is pin 3; pin 4 is NC on the fixed-output version (p. 2) · trap: EN = pin 5 in the SOIC-8 column

Not perfect: in the 12-question test with a small AI model, PartsPlan gave 3 wrong values and the attached PDF 5. Conditions and full results →

Whose document it is, first
We check the company and part number printed on the document, so it is the maker's document for the part you asked about.
A page number for every value
Values from tables, graphs and drawings come with the document and page, plus the original page image to check against.
Hard-to-read documents too
Values inside graphs are read out as numbers, and scanned documents and Chinese-language datasheets have already been read.
Organized in advance, so faster
Prepared documents are found and read from the first question. In the footprint and routing tests, the three-run average time was half or less.

14M+Datasheet pages

7.2M+Part numbers

Connects to the AI you already use

Pick your AI below
Sign in when the connect window opens (first time: free sign-up)
Claude, ChatGPT, Cursor and VS Code open the PartsPlan sign-in window when you connect; first-timers can sign up for free right there (no card). Codex and Gemini CLI get a key from the setup guide.
Ask about parts as you normally would
What is the TPS7A4701 dropout voltage at 1 A? Give me the datasheet page.
More examples
Check every line of this BOM for the manufacturer's datasheet and tell me which pages to open first.
What does the TPS563200 current-vs-ambient curve give at 85 °C?
Put AO3400A, 2N7002 and BSS138 in one table — VDS, ID, VGS(th), RDS(on), Qg — with the page and condition for each cell.

What the AI built with PartsPlan

We gave the same AI the same tasks. These are what the side with PartsPlan connected actually produced.

KiCad 3D top view of the AI-designed 4-channel isolated PT100 board — four sensor terminals, ADC, isolator and transformer — Designed by AI end to end
4-channel isolated PT100 measurement board — from requirements to fabrication files by AI
AI agents produced the schematic, placement, routing, BOM, checks and documents; no human edited the schematic or the copper
4 layers, 90 × 65 mm · 88 parts · ADS124S08 24-bit ADC
ERC 0 · DRC 0 violations · 0 unconnected (an independent re-run agrees)
Every datasheet value cited by document and page — evidence from PartsPlan
No human edited the schematic or the copper · no human EE review
Unbuilt and unmeasured design (calculated values) · October 2026
Download the design files(5.0 MB · ZIP)Read the EEVblog post ↗

Top silkscreen and copper of the same board — reference designators, pin-1 marks, isolation strip note — Designed by AI end to end
4-channel isolated PT100 measurement board — from requirements to fabrication files by AI
AI agents produced the schematic, placement, routing, BOM, checks and documents; no human edited the schematic or the copper
4 layers, 90 × 65 mm · 88 parts · ADS124S08 24-bit ADC
ERC 0 · DRC 0 violations · 0 unconnected (an independent re-run agrees)
Every datasheet value cited by document and page — evidence from PartsPlan
No human edited the schematic or the copper · no human EE review
Unbuilt and unmeasured design (calculated values) · October 2026
Download the design files(5.0 MB · ZIP)Read the EEVblog post ↗

The BQ25895 charger IC section of the AI-made schematic — net labels such as VBUS, SDA, SW and BAT on all 25 pins. — Circuit design
Charger IC schematic
Part of the schematic the AI made in the test · every pin connected by a net label
Major errors
0
PartsPlan connected
3
Without PartsPlan
Same AI · same task · 90-minute cap · one run per side · the other side also used web, shell and KiCad · October 2026
Download KiCad files(124 KB · ZIP)
Test output example — review before building
See the use cases →

The STM32 development board with all 49 parts placed by the AI, drawn in the KiCad PCB editor — pads, part outlines and silkscreen — PCB placement
49-part STM32 board placement
The board the AI placed in the test (KiCad 2D and 3D) · run 2 of 3
Average score (of 100, three-run average)
92.6
PartsPlan connected
81.7
Without PartsPlan
Same AI · three runs per side, ahead in all three · the other side also used a shell and KiCad · October 2026
See the use cases →

KiCad 3D view of the STM32 development board with all 49 parts placed by the AI — pin headers, microcontroller, crystal and power parts — PCB placement
49-part STM32 board placement
The board the AI placed in the test (KiCad 2D and 3D) · run 2 of 3
Average score (of 100, three-run average)
92.6
PartsPlan connected
81.7
Without PartsPlan
Same AI · three runs per side, ahead in all three · the other side also used a shell and KiCad · October 2026
See the use cases →

The two-layer board routed by the AI, drawn in the KiCad PCB editor — red front and blue back copper traces, ground pour and reference designators — PCB routing
Buck and LDO two-layer board
The board the AI routed in the test (KiCad 2D and 3D) · run 2 of 3 (20/20)
Average time (three-run average)
10.6 min
PartsPlan connected
21.3 min
Without PartsPlan
Same quality score: both scored 20 out of 20 (final run)
same AI · the other side also used a shell and KiCad · October 2026
Download KiCad files(31 KB · ZIP)
Test output example — review before building
See the use cases →

Top of the two-layer board routed by the AI, rendered in KiCad 3D view — connectors J1 and J2, regulators U1 and U2, capacitors, resistors and traces — PCB routing
Buck and LDO two-layer board
The board the AI routed in the test (KiCad 2D and 3D) · run 2 of 3 (20/20)
Average time (three-run average)
10.6 min
PartsPlan connected
21.3 min
Without PartsPlan
Same quality score: both scored 20 out of 20 (final run)
same AI · the other side also used a shell and KiCad · October 2026
Download KiCad files(31 KB · ZIP)
Test output example — review before building
See the use cases →

The BME280 footprint the AI made, shown in the KiCad footprint editor — eight pads, silkscreen outline and courtyard — Footprints
BME280 footprint
The footprint the AI made in the test, in the KiCad footprint editor · run 1 of 3
Average time (three-run average)
10.1 min
PartsPlan connected
24.8 min
Without PartsPlan
Six parts · same AI · three runs per side · the other side also used a shell and KiCad · October 2026
Download KiCad files(12 KB · ZIP)
Test output example — review before building
See the use cases →

Circuit simulation — 12 V→5 V buck load stepTime from schematic to simulation to PCB (first comparison, one run): 64 min vs 85 min

ngspice load-step response of a 12 V to 5 V buck (XL1509-5.0) at 10.8, 12 and 13.2 V input — output voltage and inductor current against ±250 mV limit lines — Circuit simulation
12 V→5 V buck load step
An ngspice model the AI built from manufacturer datasheet values · values missing from the documents were stated as assumptions · shown: the re-run
Time from schematic to simulation to PCB (first comparison, one run)
64 min
PartsPlan connected
85 min
Without PartsPlan
Lesser-known Chinese and Taiwanese parts · all seven documents from the manufacturers · design quality on par · same AI · one run each · October 2026

With the EDA tools you use

In the tests the AI produced schematics, boards and footprints in KiCad format, and those files open as they are in KiCad 10 (the images above are those files and placements drawn by KiCad).
Package drawings and land patterns come exactly as printed, so whichever tool you draw in — EasyEDA, Altium or others — the evidence is the same original page.

Draw in EasyEDA, build with JLCPCBThe BOM exported from EasyEDA, as is — the maker's document found for 14 of 14 real lines (October 2026, one run)

Pass on the BOM exactly as EasyEDA exports it — design, datasheet checks and assembly ordering stay on the same part numbers.

Design in EasyEDADraw the circuit and export the BOM. The manufacturer column (for example MDD(辰达半导体)) comes out together with C numbers.
Check with AI + PartsPlanPass the BOM on as it is, and each line gets that maker's document — 14 of 14 lines of an actually exported BOM (October 2026, one run).
Order assembly at JLCPCBThe C numbers in an EasyEDA BOM are the same as JLCPCB assembly part numbers, so the checked part numbers go into the order as they are.

Hand the AI the BOM exactly as EasyEDA exports it, and each line, passives included, is checked against that manufacturer's original datasheet — in a comparison test with the same AI (Claude) and a 14-line BOM in EasyEDA's format, lines confirmed from the manufacturer's original: web only 8, web + PartsPlan 14 (October 2026, one run).
Each line comes back with the manufacturer's document and the pages to open first (pinout, package, electrical characteristics), and the AI opens those pages to review them.
In a test with a BOM actually exported from EasyEDA Pro and passed on unedited — 14 lines, maker names as the export writes them (for example 'MDD(辰达半导体)', with the Chinese name alongside) and C numbers included — that maker's document was found for all 14 lines (October 2026, one run).