EmbedVera / AI CAN log analyzer & debugger

Debug CAN
issues with AI.

Analyze CAN logs with DBC files and engineering context, then turn them into traceable evidence and the next debugging step.

Local-first. Evidence-driven. Engineer-led.

EmbedVeraReviewed Case 001

Conversation

Customer issue

Turn indicator flashes at the wrong frequency.

CAN logsDBC files
EmbedVera

Request and response timing was checked against the observed dominant pattern.

12 shortened request pulses were found. The final software root cause remains unknown.

Presentation excerpt from a reviewed investigation.

Investigation Summary

Investigation direction · inference

Start with the upstream request.

The abnormal timing is already present in the upstream request.

Final software root causeUNKNOWN
≈ 400 msDominant ON duration
12Shortened request pulses
10 similar · 1 no response · 1 mismatchObserved response relationships

Next investigation

Inspect request-generation logic and upstream inputs.

Evidence Timeline

Reviewed pulse excerpt
Selected event 42.666821 s
Request width
199.649 ms
Response width
200.000 ms
Delay
78.640 ms
Request
Response

CAN channel 1 · one source log. Only reviewed ON intervals shown; adjacent states omitted. Correlation does not prove internal causality.

Reviewed source and limitations

Case 001 frozen regression record, event 1. Widths and delays are reviewed measurements; drawn endpoints use start + width. No expected timing requirement is supplied.

Request: BCM_LeftTurnLampReq · Response: HCML_LeftTurnLampSt

From issue to evidence

A clear path to the next step.

  1. 01

    Issue

    Bring the customer symptom, CAN logs, DBC files and engineering context.

  2. 02

    Investigation

    Focus on relevant signals, timing and supported conditions.

  3. 03

    Evidence

    Read source-backed findings, measured transitions and what remains unknown.

  4. 04

    Next step

    Move the investigation boundary with a concrete engineering question.

Reviewed investigation / Case 001

The short pulse is already
in the request.

Customer issue: Turn indicator flashes at the wrong frequency.

What the evidence supports

The dominant ON duration is approximately 400 ms. The reviewed log contains 12 shortened request pulses, with 10 similar responses, 1 no response observed and 1 response mismatch.

The abnormal timing is already present in the upstream request. This supports moving the investigation upstream; it does not prove internal software causality.

Next investigation

Inspect request-generation logic and upstream inputs.

Final software root causeUNKNOWN

No expected-behavior source is supplied. Functional abnormality remains not_assessed. A missing observed response does not prove an ECU failure.

Built around engineering judgment

Keep your evidence.
Keep your judgment.

Read the data boundaries

Local-first by design

Raw engineering artifacts stay local by default. Live AI is an explicit choice and receives selected relevant context.

Traceable to the source

Reports retain evidence references and project-relative source locations, confidence, unknowns and limitations.

Conservative about causality

Correlation can support a direction. It cannot, by itself, prove a software defect or an internal execution path.

Field notes

A better starting point.

Questions, answered

Frequently asked questions

What does EmbedVera do?

EmbedVera is an AI-assisted CAN issue investigation tool for embedded and automotive engineers. It connects a customer issue, ASC logs, DBC files and available function specifications to traceable observations, candidate explanations and a next investigation step.

Which inputs does the current Beta support?

The current engine works with tested ASC text logs, DBC signal databases, DOCX text function specifications and explicit engineer context. Other log formats, source-code analysis, Simulink and Stateflow analysis are not currently supported.

Does EmbedVera automatically find the root cause?

No. It helps narrow an investigation and preserve supporting evidence. Timing correlation and a missing observed response do not prove internal software causality. Final engineering judgment stays with the engineer.

Are raw engineering files uploaded?

Raw engineering artifacts are local-first, and large logs are not sent wholesale to a model. If live AI is explicitly enabled, selected relevant context may be sent to the configured model provider. This website does not accept engineering file uploads.

Early access Beta

Bring a question.
Leave with a direction.

Register with your email to get free Beta access and download the Windows app. Sign in after installation to start investigating.

Join the Beta