v1.5.8 → v1.5.9.1b · live

AI that produces engineering evidence, not just edits.

Operator X02 reads your board, reviews its own changes, and hands you a report an engineer can sign off on — engineering scores, findings with clause references, root-cause analysis and a release recommendation. Nothing is applied to your source without your confirmation.

⬇ Download for Windows — Free See it on real hardware

✓ Review before apply ✓ ISO 26262 & MISRA aware ✓ JSON export ✓ 100% local

AI Engineering Assessment Report example output

Engineering score

published formula, counted findings

Findings

1 critical · 2 major · 4 minor

ISO 26262 checklist

9/14

covered, with code evidence

Confidence

HIGH

qualitative — never a fake %

Findings — each traceable, each with a verification check

X02-A-001CRITICALReturn value of the safety-relevant call is discarded

X02-A-002MAJORUnbounded loop on a hardware-dependent flag

X02-A-003MAJORImplicit conversion narrows a signed value

X02-A-004MINORMagic number where a named constant exists

AI engineering decision

Release not recommended

One critical finding remains open in a safety-relevant path. Projected score after the two required actions: 92. Each action carries a verification check, so re-running the assessment closes the finding rather than restating it.

Illustrative report. Every number shown is computed by the IDE from counted findings — never invented by the model.

The rule everything else follows

AI never edits your source behind your back.

Every AI-assisted change in Operator X02 travels the same path. The diff is unified and Git-style, with real line numbers, added and removed counts, and a warning when a change is larger than you probably expected.

Operator X02 AI edit review flow: a proposal is shown as a diff, reviewed, confirmed, then applied. — The real thing — proposal, review, confirmation, apply. v1.5.9.1b

AI proposal

The model suggests a change. Nothing is written yet.

→

Engineering review

Unified diff, real line numbers, added/removed stats.

→

Your confirmation

You accept or reject. Large changes are flagged first.

→

Apply

Only then does anything touch the file on disk.

AI-generated source changes are never applied automatically. Popups are isolated, confirmation is explicit, and the API layer validates defensively.

Engineering AI Reports

Four AI actions. Four engineering artifacts.

The AI actions in the editor no longer return a wall of prose. Each one produces a structured result you can read, act on, and hand to someone else.

The Operator X02 editor context menu showing the AI engineering actions. — Where they live — the AI actions in the editor context menu

A review dialog showing an ISO 26262 functional-safety analysis beside the unified diff. — ISO 26262 analysis alongside the diff

A review dialog showing a MISRA C compliance analysis beside the unified diff. — MISRA C compliance, same review-before-apply path

🔍

Review this code

A verdict and an engineering score, backed by a findings table rather than an opinion.

Engineering score
Findings table with severity
Risk analysis
ISO 26262 references
MISRA compliance notes
One-click “Edit with AI” per finding

🧪

Generate tests

A case inventory before the code, so you can see what is and isn’t covered.

Nominal, edge and error cases
Edge-case detection
Correct test file naming
Created as a file, never injected

🔧

Suggest refactoring

Typed suggestions with before and after, each routed back through the review dialog.

Before / after comparison
Engineering rationale per suggestion
Apply only via review

📖

Generate documentation

Documentation is previewed in context first. It is never silently inserted into your source.

Preview before insert
Explicit confirmation
No automatic source modification

🎨

Professional Analysis, redesigned

The full clause-cited analysis report, restructured so you stop scrolling and start reading.

Engineering grid layout
Sticky navigation
Compliance indicators
Save as HTML

🛡️

Safety by construction

The failure mode for an AI editor is silently destroying working code. That path is closed.

Protection against accidental overwrite
Popup isolation
Explicit confirmation
Defensive API validation

AI Engineering Assessment Report

Built for a decision, not for an explanation.

Most AI review output tells you what it thinks. This one tells you whether to ship — and what specifically has to change first. It sits alongside the Professional Analysis report as a second, decision-oriented view.

The AI Engineering Assessment Report: engineering score, findings, checklist coverage and release recommendation. — The Assessment Report as it is generated in the IDE — score, findings, checklist, decision

Capability	What it gives you	Status
Engineering score	A number with a published formula, computed from counted findings	✓ Live
Root cause analysis	Findings grouped by cause, not listed as symptoms	✓ Live
ISO 26262 checklist	Coverage as N-of-M, each item backed by code evidence	✓ Live
Risk assessment	A qualitative risk matrix rather than invented percentages	✓ Live
JSON export	A machine-readable twin with stable finding IDs	✓ Live
Historical trend	Score movement across runs on the same file	✓ Live
Release recommendation	An explicit ship / don’t-ship call with the reason attached	✓ Live

Stable finding IDs are what make the JSON export useful: the same finding keeps its identity across runs, so a re-run closes an issue instead of restating it.

Schematic + datasheet context

Code that matches the board — not code that guesses it.

In embedded work the expensive bugs come from code that has drifted away from the hardware: a pin assigned wrong, a peripheral configured for the wrong port, a signal read on the wrong channel. Put the schematic in the conversation and that gap closes.

Operator X02 reading a board schematic and producing firmware that matches it. — Drop a schematic in, get firmware that matches the board. v1.5.8

📐 schematic.pdf 📄 datasheet.pdf 📊 pin-map.xlsx 📄 spec.docx

✓Generate from the wiring. Ask for pin #defines, a HAL init or a driver skeleton and it uses the real pins, ports and peripherals on the board.

✓Check code against the board. Open your source alongside the schematic and have every pin, bus and peripheral verified against how it is actually wired.

!Catch the drift. Mismatches are flagged — a pin the code drives as an LED that the schematic shows switching a MOSFET.

Read from a real shield PCB

The 2.4 GHz radio module and the role it plays on the board
The Arduino headers — digital, analog and power — that mate with an Uno or Mega
A voltage divider dropping 5 V IOREF to roughly 3.3 V for the module’s alternate supply pin
An N-channel MOSFET and its gate resistor, acting as a switch for external loads
The indicator LED, its series resistor, and the pushbuttons

That is the difference between attaching a file and the AI having actually read your board.

The AI identifying components on a Pololu Wixel shield: the radio module, Arduino headers, voltage divider, MOSFET, LED and pushbuttons. — A real Pololu Wixel shield, read component by component — not a diagram of one

Real document understanding

The schematic is one case of a bigger change.

Attaching a PDF, Word or Excel file extracts and reads the actual content — not just the filename. Extraction is lazy-loaded, so it only costs memory when you actually attach something and startup stays fast.

📃

Spec PDF

Drop in a requirements document and ask for an implementation against what it actually says.

📝

Word design doc

Scaffold against a written design instead of re-typing it into the chat.

📈

Excel data

CAN matrices, BOM tables, pin maps and signal lists, read as structured data sheet by sheet.

Two drop zones in Operator X02: dropping on the chat attaches a file as AI context, dropping on the FILES panel copies it into the project. — Two drop zones — on the chat it becomes AI context, on the FILES panel it lands in your project

Attached files stay in the conversation, so follow-up questions don’t need a re-attach — and an image sent to a text-only model now says so, instead of describing something it never saw.

AI that produces engineering evidence, not just edits.

AI never edits your source behind your back.