Redline LabsOur work

Redline MachineOS

Fix it yourself.
Keep the record.

MachineOS builds a digital twin of your vehicle from the photos, video and measurements you take, and puts a mechanic on your iPhone that has read it. Every job starts from what’s known and ends on the record.

Status
In development
Platform
iPhone · iOS 17.4+
AI
Claude, with your own key
Data
Kept on your iPhone

The digital twin

The digital twin

A twin you can turn in your hand.

Home is your vehicle in 3D. Drag to orbit, pinch to zoom, and tap any of 11 regions — a wheel, the cabin, the engine bay — to see what’s been captured there and what’s wrong with it.

Problems glow where they live.

The Issues layer tints each region from amber to red by severity, with a count pinned to every side that faces you. Switch to Captured to see what the baseline still hasn’t covered.

One line on what matters.

Tap a region and the focus bar names its worst issue, with a one-line tip on what it probably means. Open it for its intake steps, its components, and the next shot to take.

The Ford Bronco model from MachineOS, turning as you scroll. The front-left wheel and hood are tinted red for open issues.
◆ 2
Front-left wheel (driver side)Front-left tire sidewall cracking

Scroll to turn · drag to spin

The structure

The record, in plain terms.

MachineOS keeps what you captured apart from what it read, and both apart from what you’ve confirmed. Each layer points back to the one beneath it.

Evidence

What you captured.

Photos, video with sound, measurements and optional LiDAR scans — taken step by step through the 60-step baseline, or whenever something comes up. They’re stored on your iPhone, in a folder you can open in the Files app.

  • Photo · video · measurement · LiDAR
  • Stored on your iPhone
  • Captures work offline

Observations

What it read.

Claude reads each capture and returns what it can see, each marked by how it’s known — observed, inferred or unknown — with a confidence and a link back to the image. Unknowns come with the exact shot that would settle them.

  • Observed · inferred · unknown
  • Linked to the capture
  • New parts wait for you

The twin

What you’ve confirmed.

Confirmed facts land on the components they belong to, grouped by system and mapped to 11 regions on the 3D model, with issues, measurements and a timeline attached to each.

  • Components by system
  • Issues ranked by severity
  • A timeline to the minute

Memory

What it has learned.

After chats and finished jobs, MachineOS proposes what it learned — a part replaced, the tools you own, where you buy parts. Nothing is kept until you accept it, and what you accept goes into a hash-chained log that can be exported for Polis.

  • Proposed, then accepted
  • Hash-chained log
  • Exports for Polis

In your hand

From the first photo to the finished job.

Capture

Sixty steps, each with a place to stand.

The baseline walks you around the truck in 60 guided captures, from the VIN plate to the frame rails. Every step says where to stand and what to fill the frame with, and marks you on a top-down map of the truck. Can’t reach one? Skip it with a reason and it waits in Requested Captures.

Photo · video with sound · measurements · optional LiDAR

Read straight off the sidewall.

Each capture is analyzed by Claude, and what comes back is sorted by how it was known: identified from the image, possible but unverified, or unknown — with the exact shot that would settle it. New parts stay out of your twin until you confirm them.

Identified · Possible · Unknown
Capture Review: Falken Wildpeak A/T, 35x12.50R15LT and an aftermarket alloy wheel, each marked Observed

Fix

Problems, ranked and sourced.

Issues are sorted by severity and keep their evidence: why each was flagged, how to verify it, and whether the AI or you raised it. Report a new one on video — show it, say what’s wrong, and the form fills itself in from what you said.

Swipe right to resolve · left to dismiss

A mechanic who has read the file.

Every issue has its own Mechanic. It knows the truck’s recorded facts, the issue, and the tools you own. It leads with the likely cause, tells you what to check first, flags what’s dangerous, and asks the questions that narrow it down. Try answering it.

Streams on your iPhone · typical specs, labelled verify

The whole job, step by step.

When the fix is clear, the Mechanic writes the job: prep, access, repair, reassembly, test, cleanup — with gotchas, typical torque, and when to stop and get help. It keeps the plan current as you talk. A clip breaks; a step appears.

Plans update from the conversation
A job plan to replace the driver door lock actuator, 6 of 14 steps done

Parts, with links that exist.

Parts and tools are checked against what you already own, and where you last saw them. For the rest, Claude searches fourteen retailers. Only links that came back from the search are kept, and every price says verify.

RockAuto · AutoZone · Amazon · NAPA · and ten more
Parts and tools for the job: a Dorman actuator marked bought, with prices from RockAuto, AutoZone and Amazon

Keep

Photograph a drawer. Get an inventory.

Point the camera at a drawer and MachineOS boxes every tool it sees. Keep each one, mark what isn’t a tool, add what it missed. One inventory serves every vehicle, organized down to the wall, the cabinet and the drawer.

Recognized against the tools you already have
Review photo: fifteen tools detected in a drawer, numbered with boxes

The paperwork, read and remembered.

Scan registration, insurance and receipts. Text is recognized on the iPhone first, and title, registration and insurance stay on the device until you tap Read it. Fuel receipts become rolling MPG; expiry dates become reminders.

Names and addresses are never stored

It remembers what you agree to.

After chats and finished jobs, MachineOS proposes what it learned — the actuator you replaced, that you buy from RockAuto, the torque wrench you own. Nothing is kept until you accept it, and what you accept goes into a hash-chained log.

Exportable in the Polis Memory format
Memory proposals to review, each with Accept and Reject
Capture Review: Falken Wildpeak A/T, 35x12.50R15LT and an aftermarket alloy wheel, each marked Observed
A job plan to replace the driver door lock actuator, 6 of 14 steps done
Parts and tools for the job: a Dorman actuator marked bought, with prices from RockAuto, AutoZone and Amazon
Review photo: fifteen tools detected in a drawer, numbered with boxes
Memory proposals to review, each with Accept and Reject

The Mechanic

Answers it can stand behind.

Every issue has its own Mechanic. It reads the truck’s recorded facts, the issue and your tools, and follows the same rules every time.

Likely cause first

Every answer leads with the likely cause, then what to check first.

Safety, called out

Anything that can hurt you gets its own Safety note.

Questions that narrow it

Up to three tappable questions, plus the photo or video that would settle it.

Typical, verify

Torque and sizes only as typical ranges, marked verify.

A plan that keeps up

The job plan changes in small edits as you talk — a step added, a part swapped — and is never rebuilt from scratch.

Real links only

Parts from fourteen retailers and how-to videos are kept only if they came back from the search.

Your tools first

Checked against the tools you own, and where you last saw them.

It knows your truck.

Every turn carries the vehicle’s recorded facts, the issue, other open issues, your tool inventory and what MachineOS remembers.

It works from evidence.

Issues stay possibilities until they’re verified, and your corrections outrank anything the AI inferred.

You bring the key.

Answers come from Claude using your own Anthropic API key, kept in the iOS Keychain. There is no MachineOS server in between.

one Mechanic turnillustrative
  1. context1990 Bronco · door facts · 12 tools · 3 memories
  2. replylikely cause · check first · safety
  3. ask2 questions · 1 video capture
  4. planrev 1 · 14 steps · ~2.5 h · moderate
  5. parts14 retailers searched · 3 links kept
  6. memory1 fact proposed · waiting for you

Evidence first

The AI suggests. You decide what’s true.

Every fact in the twin says how it’s known, with a link back to the capture it came from. What the AI inferred never quietly turns into what you confirmed.

  • OBSERVED · 99%Read directly from the image.
  • ~ INFERRED · 64%Supported by the evidence, not yet verified.
  • ? UNKNOWNNeeds more evidence — and says which shot.
  • ✓ VERIFIED · YOUConfirmed by you. It outranks everything else.
Wheel size~15x10~ INFERRED · 64%Estimated from tire width and lip depth · IMG_0388

The correction sheet from Capture Review. Tap to try it.

The philosophy

What we believe about your machine.

Evidence before assertion.

If it isn’t legible in the image, it isn’t stated as fact. It’s marked unknown, with the shot that would settle it.

“Never state a part number, brand, size, date or specification that is not legible or unambiguous in the images.”

Possibilities, not diagnoses.

An issue says what was seen, how serious it might be, and how to verify it — never that it’s certain.

“Issues are possibilities, never diagnoses.”

You have the last word.

Confirm, reject or correct anything, and it stays that way. The AI never re-asserts a value you overruled.

“Human corrections are authoritative.”

Numbers you can check.

Specs come as typical ranges marked verify. A torque value is a starting point, not an order.

“Torque and sizes only as “typical … , verify”.”

Talk like a mechanic.

Plain sentences, the way someone standing next to you in the garage would put it.

“…as an engineer standing beside the owner would say them.”

Plans grow; they don’t restart.

The job plan changes in small edits as you talk, so progress and notes are never lost to a rewrite.

“Small edits only; never rebuild it.”

Quoted lines are from the instructions MachineOS gives Claude.

In numbers

60guided captures in the baseline, VIN plate to frame rails
11tappable regions on the Bronco’s 3D twin
14retailers searched for each part, links kept only if real
0accounts, and no MachineOS server

Your machine. Your record.

No account, no server.

There is no MachineOS cloud. AI features use your own Anthropic API key, kept in the iOS Keychain, and requests go from your iPhone to Anthropic.

Kept on your iPhone.

The twin, every photo and video, Mechanic chats and the tool inventory are stored on the device — browsable in the Files app, exportable as JSON.

Held until you say.

VIN, door-jamb and registration captures, and title, registration and insurance documents, stay on the iPhone until you choose to send them.

Works in the garage.

No signal under the truck? Captures save offline and are analyzed when you’re back in range.

In development, for iPhone.

MachineOS is being built for iOS 17.4 and later. The 3D twin and the guided baseline are made for the Ford Bronco first; any vehicle gets issues, documents, tools, memory and the Mechanic. LiDAR scanning needs an iPhone with LiDAR.

All Redline offerings

Screens are from the app running in the iOS Simulator with sample data; some are redrawn to animate. Renders use the Bronco model that ships in the app.