TestBench · hardware-in-the-loop

Put your hardware on a bench your agents can drive.

TestBench is a set of hardware interfaces that wrap the device under test — programmable I/O and power, bus injection, measurement. Tests run automatically on every firmware build, and a Levirge agent can run the exact same plan and read the verdict.

Automated & agent-driven Runs in CI on every build 1-Wire · CAN · RS485 · LIN · TTL
testbench · plan: ota-smoke · run #214 · agent-driven · claude
Device under test
RIOT-GW-CAN esp32 · rev C
Interfaces
Power & brownout
GPIO drive/read
CAN inject/capture
Meter · V/I/timing
Test run 4 / 5 passed
✓ Pass820 ms

Power on — boots and announces under 800 ms

✓ PassCAN

Injected 500 frames — all echoed to mqtt/can/rx

✓ Pass3.0 V

Brownout at 3.0 V handled — clean reboot, no corruption

✗ FailOTA

Rollback on a bad image didn't trigger — device hung at boot

⋯ Runningstep 5

Reflashing stable, re-running OTA rollback…

On the bench

Everything you'd poke a board with, automated

TestBench gives the device under test power, signals, traffic and measurement through one fixture — driven by a script, your CI, or a Levirge agent working a test plan.

Programmable I/O & power

Drive and read GPIO, switch and ramp the supply, and inject faults and brownouts — reproduce the nasty field conditions on demand.

Bus injection & capture

Push 1-Wire, CAN, RS485, LIN or TTL traffic into the device and capture what comes back — the same buses our gateways speak.

Measurement

Voltage, current and timing on every pin, logged per run — so a "pass" is a number with a trace behind it, not a guess.

Agent-runnable

The bench is exposed as an interface a Levirge agent can drive — it runs the plan, reads the verdict, and opens an issue in Levirge Teams when something fails.

How it works

Device under test, the bench, and whoever's running it

Device under test

An ESP32 gateway or board, wired into the fixture.

power · I/O · bus

TestBench

Programmable I/O and power, bus injection and capture, measurement — one fixture around the board.

run plan ↔ verdict

Whoever's running it

A Levirge agent, CI on every build, or an engineer — all driving the same plan.

A failed test opens an issue in Levirge Teams, with the trace attached — so an agent or a human picks it straight up.

Get started

Test your hardware like you test your code

Tell us the board and the buses it speaks. We'll spec a bench that wraps it, wire it into your CI, and let your agents run the plan on real silicon.