Skip to content

Operator X02 v1.5.8 — The AI Reads Your Schematic and Datasheet

Choose a tag to compare

View all tags
@csheng21 csheng21 released this 21 Jun 00:14
v1.5.8
f7859b1

Drop a board schematic, datasheet, or spec into Operator X02 and the AI reads the actual design — components, nets, pinouts — then helps you write firmware that matches it. No more cross-referencing PDFs by hand or hoping the AI guessed your hardware right. For embedded and edge-AI work, this is the difference between code that matches your board and code that guesses.

x02_schematic_to_code

📐 Featured: drop a schematic, get real engineering help

Here's a real example from this release. Dropping a shield PCB schematic (a Pololu Wixel wireless module on an Arduino shield), the AI read the actual board and correctly identified:
x02_schematic_datasheet_context

  • The Wixel (U1) 2.4 GHz radio module and its role
  • The Arduino headers (digital 0–13, analog A0–A5, power) that mate with an Uno/Mega
  • A voltage divider (R2, R8) dropping 5V IOREF down to ~3.3V for the Wixel's VALT pin
  • The MOSFET (Q1 + R5) as an N-channel switch for external loads
  • Indicator LED + resistor, and the pushbuttons (SW1, SW3)

That's the difference between "I attached a file" and "the AI read my board and can now help me write firmware that matches it."

Make your code match the schematic

In embedded work, the most expensive bugs come from code that drifts from the hardware — a pin assigned wrong, a peripheral configured for the wrong port, a signal read on the wrong ADC channel. With the schematic in context, the AI keeps your code aligned to the actual board:

  • Generate code from the wiring — ask for pin #defines, a HAL init, or a driver skeleton and it uses the real pin numbers, ports, and peripherals shown in the schematic, not generic placeholders.
  • Check existing code against the board — paste or open your source alongside the schematic and have the AI verify that every pin, bus, and peripheral in code matches how it's actually wired. It flags mismatches (e.g., "GPIO12 is the LED in code, but the schematic shows GPIO12 driving the MOSFET").
  • Keep firmware and hardware in sync — when the board changes, drop the new schematic and ask what in the code needs to change to match.

The schematic and your source live in the same conversation, so the AI reasons across both — hardware design and code design as one. Pair the schematic with an Excel pin-map or CAN matrix and the AI has the full picture: the board, its signal definitions, and your implementation.

Why this matters for embedded work:

  • Firmware against real pinouts — driver and init code references the actual pins and peripherals on the board, not generic assumptions.
  • Design review — sanity-check a voltage divider, flag a missing pull-up, or explain what a sub-circuit does.
  • Onboarding a board — drop the schematic plus the MCU datasheet and ask questions instead of cross-referencing PDFs by hand.
Screenshot 2026-06-21 072722

📄 Real document understanding (PDF / Word / Excel)

The schematic case is one instance of the bigger change: attaching a PDF, Word (.docx), or Excel (.xlsx) file now extracts and reads its actual text, not just the filename.

  • Spec PDF → drop it and ask the AI to implement the requirements.
  • Word design doc → have it scaffold against your written design.
  • Excel data (CAN matrices, BOM tables, pin maps, signal lists) → read as structured data, sheet by sheet.
  • Follow-up questions → ask about the same document again without re-attaching; it stays in context.

Extraction uses pdf.js (PDF), mammoth (Word), and SheetJS (Excel), all lazy-loaded so they only consume memory when you actually attach a document — startup stays fast.

✨ Also in this release

No more invented image descriptions. Attach an image to a text-only model and the IDE now tells you it can't see images and points you to a vision provider (Claude, GPT-4o, Gemini) — instead of confidently describing something it never saw.

Two ways to bring files in:

  • Drop on the chat → attach as AI context
  • Drop on the FILES panel → copy the file into your project on disk (appears in the tree), with an overwrite prompt for same-name files
x02_drag_drop_zones

Attach bypath. A new toolbar button lets you type a file path (relative or absolute) to attach it — handy for deep folder structures.

A cleaner attachment UI. Many files collapse to a one-line summary (📎 7 files · 257.9 KB) that expands into a VS Code-style list with icons, full names, paths, and sizes. Plus floating tooltips on every toolbar button.

🛠️ Full changelog

Added

  • PDF / DOCX / XLSX text extraction in the attach pipeline — the AI reads document content, including schematics exported to PDF, so it can match your code to the actual board (pins, ports, peripherals)
  • Attached files persist as conversation context for follow-up questions (sent once, no token bloat)
  • Vision-capability guard — warns instead of hallucinating when an image hits a text-only model
  • "Add file by path" toolbar button with a styled input dialog
  • Drop on the FILES panel to copy a file into the project on disk (with overwrite prompt)
  • Collapsible attach bar with total-size summary + VS Code "Open Editors" expanded list
  • Floating tooltips for all toolbar buttons (escapes overflow clipping)

Changed

  • Modern Office formats (.docx / .xlsx / .pptx) blocked from opening as raw binary in the editor

Fixed

  • Word/Excel files no longer render as garbled zip/XML in Monaco
  • Double-tooltip bug (legacy CSS ::after bubbles suppressed in favor of the floating tooltip)
  • FILES-panel copy now uses the app's permitted file-system commands (the Tauri fs plugin was not allowlisted)

📦 New dependencies

npm install mammoth xlsx pdfjs-dist

Coding is Art. Feel it. Enjoy it.

Assets 4
Loading