TraceSpeak CT · Continuous Testing Available for pilots
Root cause in minutes, not hours.
From a failed test to the suspect commit, automatically, on your own network.
What it is
AI powered continuous testing platform
TraceSpeak CT connects to your HiL benches and CI pipeline. After every test run it reads the test reports and the bus recordings, works out what your vehicle network is supposed to do, and answers three questions for every failed test:
- Is the failure real? Or is it a false failure caused by the test, the bench or the setup?
- What is the root cause, and what does it affect in the vehicle?
- Which software change most likely caused it?
Everything runs on one workstation inside your network.
Who it’s for
One answer, four teams.
Test & validation managers
Fewer hours lost to triage, and a clear picture of software maturity every morning.
HiL & test engineers
An answer with its evidence, instead of another log to open.
Software developers
The failed test, the logs and the suspect commit, delivered straight to you.
Quality & ASPICE managers
Traceability from requirement to commit, built up automatically.
Continuous integration & testing
From commit to verdict, on every build.
Commit
A developer pushes to Git. The change set is indexed: files, components, author.
Build
CI builds the software. Every build is linked to the exact commits it contains.
Test on HiL
The build is flashed to real or virtual HiL benches. Tests run in parallel.
Reports & logs
Test reports, bus traces and logs are ingested automatically, per run.
Analysis
Regressions, findings and anomalies. Suspect commits ranked, fixes verified.
Closed loop. The verdict, the logs and the root-cause commit go straight back to the developer on every build. Every result keeps its lineage:
Test case→HiL→Sample→SW build→Commits
Features
Everything between a failed test and its fix.
Ingest & map
Your data, read and connected automatically.
Every report and log, read automatically
Test reports and log files (CAN, CAN FD, LIN, Ethernet, SysVars) are ingested and decoded as soon as they land. Nothing to convert, nothing to upload.
Every verdict linked to its evidence
Each test report, down to the individual test step, is mapped to the exact section of the recording behind it.
Understand the vehicle
Built from your own network descriptions.
It knows how your vehicle should behave
From your network descriptions (DBC, ARXML, LDF, PDX), the TraceSpeak engine builds a model of your vehicle: ECUs, signals, routing, timing, safety relevance and the user functions they support.
Anomalies found without writing a rule
Error frames, bus-off, Ethernet link down, frame drops, silent channels, late or missing messages. Every recording is checked against what the network should do, including failures no test was written for.
Check & explain
Answers with evidence, not alerts.
Every test verdict double-checked
Each test result is judged against the real network definition and the recording. False failures caused by the test script, the bench or the setup are flagged automatically, so your team stops chasing them.
Root cause and vehicle impact
For each real failure: the most likely root cause, the ECUs and functions affected, a severity, a confidence score and the checks behind the answer.
The suspect commit, pinpointed
When a test that used to pass starts failing, TraceSpeak CT finds the last passing build, lists the commits in between and ranks the most suspicious one by file, component and change.
Track & report
From single failures to project maturity.
Software maturity at a glance
Dashboards show pass rates, regressions, open root causes and trends per build, per ECU and per feature, so release decisions rest on data.
Works with the tools you already run
HiL benches, Vector CANoe, IBM DOORS NG, IBM RQM, IBM RTC and Artifactory. No change to your test process.
Local AI with automotive know-how
Specialised agents, tuned with professional automotive embedded expertise, run on an open-weight model on your own GPU. Ask in plain English: “Why did last night’s run fail?”
The agents
A team of specialists, in plain language.
Five agents work on every failure today, and two more are coming to take it all the way to a fix. Agents call precise tools for every measurement. The AI reasons over the results; it does not guess numbers.
| Agent | What it does |
|---|---|
| Tester | Runs the tests, evaluates the results and understands what each verdict means. |
| Network System Architect | Reads the vehicle network definition and works out how the ECUs depend on each other. |
| Root-Cause Analyser | Checks each test case against the network recordings and the network definition, down to the impact on the vehicle. |
| Incident Manager | Finds the last passing build and ranks the commits in between to pinpoint the most suspicious one. |
| Judge | Weighs all the evidence and gives a verdict on the final root cause, with a confidence score. |
| Software Developer Coming soon | Reviews the commit that caused the failure and implements a fix. |
| Team Lead Coming soon | Leads the agent team and coordinates the automated software development process. |
Agents in motion
From failed test to fixed regression.
One failure, seven agents. Each agent adds what it finds to the case and hands it on, until the tests pass again.
Case fileTest run #1482
-
Runs the test. Receives the report and the logs of the failure and evaluates them.
- ✕ Step 4 failed: brake request not received
- Test report and CAN logs received
-
Finds the affected ECUs and builds the vehicle context for the analysis.
- Affected: ADAS → GW → ESP
- Vehicle context passed on
-
Determines the root cause, checks the test itself and names the failing ECU.
- Test case correct ✓
- Root cause: forwarding delay 38 ms (limit 5 ms)
- Failing ECU: GW
-
Finds the last passing build and lists the suspect commits.
- Last passing build: #1478
- 6 suspect commits in GW software
-
Weighs all the evidence and points to the commit that most probably caused the failure.
- Commit a3f9c1 · confidence 94%
-
Reviews the commit that caused the failure and implements a fix.
- Fix committed: 7be2d4
-
Receives the result and re-runs the tests until the regression is fixed.
- Re-run #1483: all tests pass ✓
↻ If the regression is still there, the loop runs again.
Examples · recorded runs
See what an answer looks like.
Real runs recorded on our demo setup. The identifiers come from our simulated vehicle.
Example 1 · Ask Testradamus
One question, 24 seconds, all local.
An engineer asks why a routing test failed. Testradamus answers with the cause, the impact and the evidence.
CentralGateway forwarded PDU_BrakeStatus_01 late: 260.5 ms to Powertrain CAN, where at most 5 ms is expected.
BrakeCtrlUnit sent it on time. Chassis CAN and the gateway’s other PDUs stayed alive.
Impact: 8 ECUs consume it, routed onto 3 buses. Severity high.
Reports
Test case, steps and verdicts
Tester
4 failed steps → finds the primary one
Network System Architect
Forwarded ≤ 5 ms on Powertrain CAN
Root-Cause Analyser
Route, sender, bus, gateway, impact
Judge
Weighs the evidence → gateway_fault
Example 2 · The full chain
From failure to suspect commit in about 2 minutes.
A speed test fails after a new build. Testradamus finds the root cause, then traces the bench, the sample, the build and the commits.
HiL bench: VULCAN-HIL-01
Sample: VT7CGW26091100482 (HW 7VT035901B, C02)
SW build: 414K
Most probable commit: cgw-app #1201, “on Powertrain CAN, send PDU_WheelSpeeds_01 before PDU_VehicleSpeed_01”, in src/com_gw/egress_order.c.
Tied with cgw-app #1195: both score 8.
Test failed
TCS_710209 VehicleSpeed
Root cause
Gateway silent for 6.1 s
HiL bench
VULCAN-HIL-01
Sample & SW build
414K, last pass 411K
Commits
412K–414K: 30 commits
Suspect
#1201, tied with #1195
Example 3 · Picture your lab
Tonight’s run, explained by morning.
22:00 · TONIGHT
Your benches run
Your HiL benches and your tests, exactly as today.
DURING THE NIGHT
One workstation reads
Every report, trace and build, on your network, on one GPU.
08:00 · MORNING
Your team has answers
- Root cause per failure
- False failures flagged
- Suspect commit per regression
Fits your toolchain
On top of the tools you already run.
HiL benches
Runs, bench and sample status
SupportedVector CANoe reports
Verdicts, steps, measured values
SupportedTraces & logs
CAN, CAN FD, LIN, Ethernet, SysVars
SupportedNetwork description
DBC, ARXML, LDF, PDX
SupportedIBM DOORS NG
Requirements ↔ test cases
In pilotIBM RQM
Test plans, cases and results
In pilotIBM RTC
Change sets, work items, defects
In pilotArtifactory
SW builds and what is in them
In pilotJenkins
CI pipelines, build and test jobs
In pilotGitLab
Repositories, commits, merge requests
In pilotBitbucket
Repositories, commits, pull requests
In pilotdSPACE
HiL systems and test automation
In pilotETAS
HiL systems, measurement and calibration
In pilotMissing a tool? Connectors are built during the pilot. Talk to us →
Runs on your side
One workstation, one GPU, one project.
One workstation serves one project.
Tokens per root-cause analysis, so it stays light and fast.
Cloud calls. Code and test data stay in your network.
The setup
- Open-weight 9B model, 262k context
- Runs offline, with no internet access needed
- UI, agents and data on the same machine
- Proven on one RTX 3090 GPU
Pilot support
The TraceSpeak founding team supports every pilot directly: setup, connectors and analysis reviews.
Security details →FAQ
Questions engineers ask us.
Do we need to change our test process?
No. TraceSpeak CT reads the outputs your benches and CI already produce.
Does any data leave our network?
No. The model, the agents and the data run on one workstation inside your network, fully offline.
Which ECUs does it support?
Any ECU described in your network files. It is not limited to one ECU type.
How do we know an answer is right?
Every answer comes with a confidence score and the checks it is based on, so your engineers can verify it in seconds.
What hardware is needed?
One workstation with one GPU per project. It is proven on a single RTX 3090.
How do we start?
Try the Demo Playground today, then run a pilot on your own data. Pilot programme →
Demo
Try TraceSpeak CT yourself.
Get remote access to our Demo Playground: a simulated vehicle network with realistic HiL test runs, failures included. Ask your own questions and follow each answer back to the log.