Skip to main content

​
Overview

Trace accepts file attachments in the chat panel. Drag and drop (or click to upload) any supported file, and the AI incorporates its content into the conversation. This means you can:
  • Attach a datasheet and ask “Implement the reference circuit from page 12”
  • Drop a photo of a hand-drawn schematic and say “Build this”
  • Upload a BOM spreadsheet and ask “Check availability of all parts”
  • Share a requirements doc and say “Design to these specs”
Attachments become part of the conversation context — the AI references them throughout the session, not just in the message where you uploaded them.

​
Supported File Types

TypeExtensionsWhat the AI extracts
PDF.pdfText content, tables, pinouts, recommended circuits, electrical specs, application notes. Datasheets are parsed server-side for structured extraction.
Images.png, .jpg, .jpeg, .gif, .webpVisual analysis via multimodal vision. Schematic screenshots, board photos, hand-drawn circuits, oscilloscope captures, reference design images.
Text.txt, .md, .csvRequirements documents, specifications, BOMs in text/CSV format.
PDF parsing uses LlamaParse on the server for structured extraction (tables, pinouts, circuits). Simple text PDFs are read directly. Image analysis uses Claude’s multimodal vision capabilities.

​
How to Attach Files

​
Drag and Drop

Drag a file from your file manager directly into the chat input area. A drop zone highlights when you hover over it. Release to attach.

​
Click to Upload

Click the attachment icon (paperclip) in the chat input bar to open a file picker.

​
Multiple Files

You can attach multiple files to a single message. Each file adds context that the AI can reference. Common patterns:
  • Datasheet + requirements doc → “Implement this chip per these constraints”
  • Multiple images → “Compare these two approaches”
  • BOM CSV + datasheet → “Are all these parts available? Suggest alternatives for any that are EOL”

​
Datasheet Workflows

Datasheets are the most common attachment type. The AI extracts structured information automatically:

​
Implement a Reference Circuit

[attached: TPS563200_datasheet.pdf]
Implement the typical application circuit from page 18.
Use the recommended component values for 3.3V 3A output.
I'm using JLCPCB assembly, so prefer LCSC-stocked parts.
The AI extracts the reference schematic, identifies component values, cross-references with live distributor stock, and places the circuit in your design.

​
Extract a Pinout

[attached: STM32F411CE_datasheet.pdf]
Which pins support hardware SPI? I need SPI1 with
both MISO and MOSI on the same port.

[attached: LM2596_datasheet.pdf]
What inductor value and output capacitor does this
regulator need for 12V→5V at 3A? What about the
input cap — should it be bulk or ceramic?

​
Application Notes

[attached: AN4488_STM32_USB.pdf]
Walk me through the USB FS connection for the
STM32F103. Do I need series resistors on D+/D-?
What about the pull-up on D+?

​
Image Workflows

Trace’s multimodal vision analyzes images directly — no OCR preprocessing needed.

​
Hand-Drawn Schematics

[attached: whiteboard_sketch.jpg]
Build this circuit in my schematic. The triangles
are op-amps (use LM358), the zigzags are resistors.
The labels show the values I want.

​
Board Photos

[attached: competitor_board.png]
What components can you identify on this board?
I want to replicate the power supply section.

​
Reference Design Screenshots

[attached: eval_board_schematic.png]
This is the eval board schematic for the AD9833 DDS.
Implement the same circuit but replace the 5V supply
with my existing 3.3V rail.

​
Oscilloscope Captures

[attached: ringing_capture.png]
I'm seeing this ringing on the clock line. What's
causing it and what should I add to fix it?

​
BOM and Specification Documents

​
Availability Check

[attached: project_bom.csv]
Check if all these parts are in stock at LCSC for
JLCPCB assembly. Flag anything that's out of stock
or has a lead time over 2 weeks.

​
Requirements-Driven Design

[attached: power_requirements.md]
Design a power supply that meets all these specs.
Use Plan mode — I want to review before you build.

​
How Context Persists

Attached files become part of the conversation context for the entire session:
  • First message: Upload a datasheet
  • Later message: “What’s the max input voltage for that chip?” — the AI still has the datasheet in context
  • Even later: “Add the recommended decoupling from page 7” — still accessible
You don’t need to re-upload the same file for follow-up questions. The content remains available until the conversation’s context window is full, at which point older context is automatically compacted (summarized to preserve key information while freeing space).

​
Context Window Indicator

The chat panel shows a context usage indicator that reflects how much of the AI’s context window is occupied. Large attachments (especially multi-page PDFs) consume significant context space. If you’re working with many attachments:
  • The indicator shows percentage used
  • When context is nearly full, older messages are automatically compacted
  • Key facts from attachments are preserved even after compaction
  • You can start a new tab for a fresh context if needed

​
Limitations

  • File size: Large PDFs (100+ pages) may be partially processed. The AI focuses on the most relevant sections based on your question.
  • Binary formats: Raw PCB/schematic binaries (.kicad_pcb, .kicad_sch) are not attachment inputs — they’re read directly by the AI’s file tools from your project directory.
  • Proprietary formats: Word docs (.docx), PowerPoint, and Excel files are not directly supported. Export to PDF or CSV first.
  • Video/Audio: Not supported as attachments.
  • Desktop only: File attachments work in the desktop app. The web dashboard supports PDF and image uploads for Ask mode questions.

​
Tips

  1. Name your files descriptively. STM32F411_datasheet.pdf helps the AI more than download.pdf.
  2. Point to specific pages. “See page 12” is faster than the AI scanning a 200-page datasheet.
  3. Combine attachments with specific prompts. Don’t just drop a file — tell the AI what you want from it.
  4. Use images for what text can’t express. A photo of ringing on a scope is worth a thousand words of description.
  5. Start fresh for new topics. If your context is full from a previous datasheet-heavy conversation, open a new tab rather than fighting compaction.

​
Next Steps

  • Using Prompts — writing effective prompts (including attachment-based prompts)
  • Agent Mode — making edits based on attached context
  • Plan Mode — complex tasks that start with attached requirements
  • Intelligent Tools — the tools the AI uses to process your attachments