datasheet-cli
A command-line tool that extracts structured data from PDF datasheets using LLMs. Point it at a datasheet, get back JSON you can use in your tooling.
# Extract pinout data from a microcontroller datasheet
# Get PCB footprint dimensions
# Search Mouser for a part and download its datasheet
# Search JLCPCB parts library (no API key needed)
Why?
Every electronics engineer has done this: you're designing a PCB, you need the pinout for a new chip, and you spend 20 minutes scrolling through a 500-page PDF to find Table 12 buried on page 47. Then you manually transcribe it into your symbol editor, probably making a typo that you won't catch until after the boards are fabbed.
Datasheets are the lifeblood of hardware design, but they're trapped in PDFs that were designed for humans to read, not machines to parse. This tool uses Gemini's vision capabilities to extract the data you actually need:
- Pinout & configuration - Every pin with electrical type, alternate functions, and groupings
- Footprint geometry - Package dimensions, pad sizes, and land pattern recommendations
- Electrical specs - Absolute max ratings, operating conditions, thermal data
- Power requirements - Voltage rails, sequencing rules, decoupling capacitors
- High-speed constraints - Impedance targets, length matching, termination requirements
- And more - DRC rules, boot configuration, reference design BOM
The output is structured JSON that you can pipe into your CAD tools, symbol generators, design rule checkers, or documentation pipelines.
Installation
From source (requires Rust 1.85+)
Binary releases
Coming soon.
Quick Start
-
Get a Google AI Studio API key (free tier works)
-
Set your API key:
# or -
Extract data:
That's it. The tool handles uploading the PDF to Gemini, manages caching (PDFs are cached for 48 hours to avoid re-uploads), and returns structured JSON.
Extraction Tasks
pinout - Pin Configuration
Extracts complete pin tables for schematic symbol generation.
footprint - Package Dimensions
Extracts mechanical data for PCB footprint generation.
characteristics - Electrical Specifications
Extracts parametric data for simulation and design verification.
power - Power Supply Requirements
Extracts power sequencing and decoupling requirements.
high-speed - Routing Constraints
Extracts impedance, length matching, and termination requirements for USB, Ethernet, DDR, etc.
custom - Your Own Prompts
Use your own extraction prompt and JSON schema:
All Tasks
| Task | Description |
|---|---|
pinout |
Pin configuration for schematic symbols |
footprint |
Package dimensions for PCB footprints |
characteristics |
Electrical and thermal specifications |
power |
Power rails, sequencing, decoupling |
high-speed |
High-speed interface routing constraints |
drc-rules |
PCB design rule constraints |
boot-config |
Boot mode and configuration pins |
layout-constraints |
Component placement rules |
reference-design |
Reference schematic BOM |
feature-matrix |
Part variant comparison |
custom |
User-defined extraction |
Distributor Integration
Mouser
Search parts and download datasheets directly:
# Search by keyword
# Get detailed part info
# Download datasheet
Requires: MOUSER_API_KEY (Get one here)
DigiKey
# Search parts
# Get part details
# Download datasheet
Requires: DIGIKEY_CLIENT_ID and DIGIKEY_CLIENT_SECRET (Register here)
JLCPCB / LCSC
Search the JLCPCB SMT parts library. No API key required.
# Search for parts
# Get detailed part info (includes JLCPCB assembly category, pricing, attributes)
Part details include the JLCPCB assembly category (basic, preferred, or extended) which determines feeder loading fees during assembly.
SnapEDA
Search parts and retrieve exact CAD data (symbols, footprints, pin-to-pad mappings) from SnapEDA. No API key required.
# Search for parts
# Get complete symbol + footprint + pin-to-pad mapping
# Get only footprint data
# Get only symbol / pinout data
The <part> argument is flexible — it auto-detects the input format:
- Numeric ID — treated as a SnapEDA
unipart_iddirectly (e.g.,datasheet snapeda part 123456) - Part number — searches SnapEDA and uses the first match (e.g.,
ADS1115IDGST) - SnapEDA URL — extracts the part name from the URL and searches (e.g.,
"https://www.snapeda.com/parts/TP4056-42-ESOP8/toppower/view-part/")
Example workflow:
# Search for a part
# Get footprint data by part number
# Get full data by SnapEDA URL
# Save extraction to file
SnapEDA data is derived from Eagle XML CAD models, so coordinates and dimensions are exact rather than LLM-interpreted. All coordinates include units (e.g., "-2.475mm"). The part subcommand returns a combined object with pinout, footprint, and pin_to_pad_map sections; the symbol and footprint subcommands return only their respective section.
This makes SnapEDA a useful complement to PDF extraction: use SnapEDA for precise pad geometry and pin-to-pad mappings, and PDF extraction for electrical characteristics, timing specs, and other parametric data not present in CAD models.
Pipeline Examples
Generate KiCad symbols
|
Batch process a directory
for; do
base=
done
Compare power requirements across parts
for; do
done
Options
datasheet extract <TASK> <PDF> [OPTIONS]
Options:
--model <MODEL> Gemini model (default: gemini-3-pro-preview)
--out <FILE> Output file (default: stdout)
-f, --formatted Pretty-print JSON
--prompt <TEXT|FILE> Custom prompt (for 'custom' task)
--schema <JSON|FILE> Custom JSON schema (for 'custom' task)
--no-cache Disable PDF caching (re-upload each time)
--api-key <KEY> API key (default: $GOOGLE_API_KEY or $GEMINI_API_KEY)
Caching
PDFs are uploaded to Gemini's File API and cached locally for 48 hours. This means:
- First extraction of a new PDF: uploads the file (~1-10 seconds depending on size)
- Subsequent extractions of the same PDF: uses cached reference (instant)
Cache location: ~/.cache/datasheet-cli/ (Linux) or platform equivalent.
To force re-upload: --no-cache
Accuracy
The prompts are designed with anti-hallucination measures:
- Every extraction includes verification data (part number, revision) that you can check
- Missing data is explicitly marked as
nullor"not specified"rather than guessed - Prompts instruct the model to return errors if key sections aren't found
That said, this is an LLM - always verify critical dimensions before sending boards to fab.
Model Selection
The default model is gemini-3-pro-preview. You can override with --model:
For large datasheets (500+ pages), the newer Gemini models with expanded context windows work best.
Cost
Gemini has a generous free tier. For typical usage:
- ~1000 datasheet extractions/month on free tier
- ~$0.001-0.01 per extraction on paid tier
The exact cost depends on PDF size and model used.
Limitations
- Only works with Gemini (no OpenAI/Anthropic support currently)
- PDFs must be readable (not scanned images without OCR)
- Some older/unusual datasheet formats may not extract cleanly
- Maximum PDF size depends on Gemini's limits (~100MB)
Contributing
Contributions welcome. Areas that could use help:
- Additional extraction prompts for specific use cases
- Integration with other CAD tools (Altium, Eagle, OrCAD)
- Support for other LLM providers
- Better error handling and validation
License
GPL-3.0-only. See LICENSE for details.
See Also
- kicad-footprint-generator - Generate KiCad footprints from parameters
- KicadModTree - Python library for KiCad footprint scripts