Datasheets are the contract between your firmware, your circuit, and the physical part. Embedr can convert PDF datasheets into project-local markdown so the agent can use them as context.
What gets created
When you import a PDF, Embedr processes it and saves output in your project:
datasheets/
BME280/
BME280.md
images/
pinout.png
timing.png
The exact names depend on the source PDF and the auto-generated component name setting.
Import a PDF
Use the datasheet drop zone in the File Explorer area.
You can:
- drag a PDF onto the drop zone,
- click the drop zone and choose a PDF.
Processing requires sign-in and internet access because PDF parsing uses cloud services.
What the agent can use
After import, the agent can reference:
- pin names,
- supply voltage,
- electrical limits,
- communication interface,
- register descriptions,
- package notes,
- timing diagrams,
- extracted images,
- project-specific notes you add to the markdown.
Ask questions that force the answer back to the datasheet:
Based on the imported BME280 datasheet, make a wiring table for an ESP32 using I2C and note the voltage assumptions.
Safety rule
Do not let generated text be the only source for electrical limits. Before powering hardware, verify:
- voltage range,
- absolute maximum ratings,
- current draw,
- pin direction,
- pull-up or pull-down requirements,
- level shifting needs,
- connector orientation.
The agent can help you find and summarize these details, but you own the final electrical check.
Good datasheet habits
Use official manufacturer datasheets when possible. Avoid random summary PDFs for design limits.
Prefer one component per PDF. A catalog PDF with dozens of parts can be parsed, but the result may be harder for the agent to use.
Rename the generated folder if the auto-name is unclear. Human-readable names help search and review.
Indexing datasheets
If codebase indexing is enabled, imported markdown can be indexed with the rest of the project. That lets semantic search find component information.
Examples:
- "maximum I2C speed",
- "pinout for INT pin",
- "operating voltage",
- "wake-up timing".
When extraction is poor
PDFs are not all built the same. Scanned pages, complex tables, vector diagrams, and multi-column layouts can produce uneven extraction.
If the output is poor:
- try a smaller or newer PDF,
- manually add a short
notes.md, - copy the critical table into markdown,
- attach the original screenshot or image to the chat if the model supports images.
Project documentation
Datasheet-derived markdown is not only for the agent. It is useful project documentation.
Add a short section at the top after import:
## Project notes
- Used with ESP32 at 3.3 V.
- I2C address configured as 0x76.
- SDA: GPIO 21
- SCL: GPIO 22
That turns a generic datasheet into project knowledge.