Software, built for your team

Your safety analysis tool. Built into your stack.

sev10 builds the AI engine behind our analysis service into your company. It connects to your ECAD, PLM and requirements tools, runs on your own AI gateway and speaks your templates. The AI drafts and checks. Your engineers decide.

The example build: a DFMEA editor showing failure chains for the brake ECU, with an AI-proposed change to one row shown as a tracked diff.
Your AIRuns on the model endpoint your company already approved
Your dataDeployed in your environment. We have no standing access
Your callEvery AI change is a proposal an engineer accepts or rejects

See the example build.

A working prototype on our brake-by-wire case study: the same engine that drafted that 783-chain package in about 10 engineer-hours, and the same findings our reviewers raised on it. Sound on.

An example of what we build, not an off-the-shelf product. Each build is scoped to your tools, templates and process. Quillon Motors is a fictional client.

Measured on the example engagement

  • 110 of 1,572datasheet facts the engine extracted were wrong and corrected in review. About 7 %. That's why nothing it writes is final until an engineer accepts it.
  • 3,764candidate failure causes generated from checklists and pad adjacency. Every one ended as a row, a pointer to a row, or a written reason for no effect.
  • ~10 hof engineer time for 783 failure chains and six computed FMEDA goals, review included.

What your engineers get.

Two editors, a connection to every system that feeds them, and exports in the formats your process already uses. Shown here on the example build.

Altium, Teamcenter, SOLIDWORKS, Jama Connect and a datasheet library wired into the sev10 engine, which runs on the client's AI gateway.

01 · Connect

Reads the design where it lives.

Schematic, BOM and layout from your ECAD and PLM, safety goals from your requirements tool, datasheets and safety manuals from your library. No re-keying and no stale copies.

The DFMEA editor: the structure tree, failure chains with severity, occurrence, detection and action priority, and an AI-proposed correction to row EPB-420 shown as a tracked change with its reasoning, the new ratings and an Accept button.

02 · DFMEA editor

Say what's wrong. Get a sourced fix.

Select the cells and write what you disagree with, the way you would to a colleague. The AI answers with a tracked change, its reasoning and the datasheet pages behind it. You accept, edit or reject it; action priority recomputes as you do.

  • AIAG-VDA structure, your rating tables
  • Every cell shows AI draft, reviewed or approved
  • Assumed values marked, with the question that closes them
The FMEDA editor: the single-point fault metric for safety goal SG-03 rising from 98.65 % to 99.08 % and crossing its ASIL D target.

03 · FMEDA editor

Ask what it takes to hit the target.

SPFM, LFM and PMHF for every safety goal, live. Ask what closes a gap and preview the answer on the metrics before anything changes. Every diagnostic-coverage claim links to its implementation evidence, with the ISO 26262-5 Annex D category as the reference point.

Revision impact: four design changes in a new board revision traced to 59 affected DFMEA rows, 51 re-rated automatically and 8 queued for review.

04 · Revisions

New board revision? See what moved.

When a layout change lands, every row it touches is flagged and ranked. Routine re-ratings come as a batch you accept or reject; anything that needs judgment goes to a review queue. The impact rules are written down, so you can audit what it didn't flag.

The exported package: HARA, Design FMEA workbook, FMEDA workbook and safety analysis report.

05 · Export

The documents your process expects.

Excel workbooks with live formulas, a report in your template, the assumptions register and open questions, and trace links pushed back to your requirements tool. The demo uses sev10's template; yours is loaded during the build.

Your data stays under your control.

Safety data is some of the most sensitive engineering data a company has. So the tool runs where your data already is, on the AI your IT team already approved, and it talks to nothing you didn't configure.

Your AI

Every model call goes to the endpoint you configure: a deployment in your own cloud tenant, or a model on your own hardware. We don't bring a model you have to approve. Model and prompt versions are pinned and recorded with every output.

Your data

Deployed on-prem or in your cloud tenant, behind your single sign-on. Design data, prompts and outputs stay on your systems. Support access is off unless you grant it, per session.

Your process

The AI proposes; an engineer accepts, edits or rejects. Every prompt, source and decision is logged with who, when and why. Your safety manager still approves the analysis, and your assessor still assesses it.

Connectors are built for your tools

  • ECADAltium, Cadence, Siemens Xpedition
  • PLMTeamcenter, Windchill, 3DEXPERIENCE
  • MCADSOLIDWORKS, CATIA, Creo
  • RequirementsJama Connect, Polarion, codebeamer

Examples, not a certified integrations list. Each connector is built and tested against your instance's documented APIs, with read-only service accounts unless you enable write-back.

A typical build

  • Runs asContainer images or a VM in your environment
  • Sign-inYour SSO (SAML or OIDC), roles from your groups
  • AuditAppend-only log, exportable to your SIEM
  • UpdatesVersioned releases your team installs in its own change window

Start with one ECU.

  1. 01

    Pilot

    We run the analysis on one of your subsystems with our own engine, at a fixed price. You judge the output on your own design, and you keep the analysis whether or not you continue.

  2. 02

    Connect

    We connect it to your tools and load your templates, rating tables and ID schemes. Output follows your formats, with every AI-drafted cell still marked as AI-drafted.

  3. 03

    Deploy

    Our engineers come on site for a week or two: we install it in your environment on your AI gateway, train your team and run it with them on a live program.

  4. 04

    Hand over

    Your team runs it. We maintain the connectors and keep the rules current under a yearly subscription.

The hard questions.

Is this a product we can buy today?

No. The demo is an example build on our case-study board. We build each tool for the team that uses it: your tools, your templates, your review process. The pilot is how you find out whether it's worth it.

We already have APIS / medini / Windchill. Why this?

Keep them. Those tools store the analysis; your engineers still write every row. This drafts the rows from your design data with sources, flags what a revision changed, and exports to the tool you already run.

Will the AI invent failure modes or ratings?

It drafts, and drafts can be wrong: on our example, review corrected about 7 % of the datasheet facts it extracted. That's why nothing it writes is final until an engineer accepts it, every rating carries its reasoning and its sources, and every value without a source is marked ASSUMED with the question that would close it. The demo shows a reviewer catching exactly that kind of error.

Does any of our data leave our network?

No. The tool runs in your environment and calls only the AI endpoint you point it at. If your policy allows no cloud at all, it can run on a model hosted on your own hardware.

Can we use it on a program that needs ISO 26262?

ISO 26262-8 clause 11 asks you to classify each tool use case by its impact and the chance an error is detected. Engineer review catches wrong proposals; it can't catch rows the tool never drafted. So for each use case we give you a classification proposal, the known limitations and our validation results, and we pin and record model and prompt versions so a run can be reproduced. Your safety manager decides.

What does it cost?

Three parts. A pilot on one ECU, priced to fit your program. Then a one-time build (the connectors, your templates and rules) plus a week or two of our engineers on site to deploy it and train your team. Then a yearly subscription for maintenance, updates and support. The pilot is the cheap way to find out if the rest is worth it.

Who owns what?

Your design data and every analysis the tool produces are yours, in standard Excel, PDF and ReqIF that open without us. The engine is licensed to you, and its source code for your version is held in escrow: if we ever stop supporting it, the source is released to you so your safety program never loses its tool.

Let's build yours.

Tell us which tools you run, what analyses your team does and where it hurts. We'll come back with a pilot scope and a price.

Built by Frank Walsh, who designed and validated production vehicle controllers at Tesla and built the protection-design tools his team used. You work with him directly. More about sev10

Scope a pilot

walshf@oregonstate.edu