UDMI / Docs / Tools / workbench
UDMI Workbench: Visual Workspace Guide
UDMI Workbench is a unified web application for UDMI device development, compliance testing, and automated AI diagnostics. It replaces complex CLI commands with an interactive dashboard.
Quick Start: Launching Workbench
The bin/workbench launcher manages the Python backend API server and opens Workbench in your browser:
# Launch Workbench v2 (Modern, Default)
bin/workbench
# Or explicitly launch Modern v2:
bin/workbench --v2
# Launch Classic Workbench v1:
bin/workbench --v1
# Specify custom port (default: 8080)
bin/workbench --port=8080
# Stop running Workbench server
bin/workbench stop
- Site Model Path: Path to the active site model directory (e.g.,
sites/udmi_site_model). Enter manually or click the folder icon to launch the interactive directory browser.
2. The Sequencer Screen
Used to run and monitor end-to-end UDMI device compliance tests.
Main Header Inputs:
- Device Under Test (DUT): Dropdown selecting the target device from the site model (
devices/ folder).
- Target Project: Text field specifying the MQTT broker or cloud project endpoint (default:
//mqtt/localhost).
- Sequencer Version: Displays the active Sequencer version (
HEAD).
Advanced Settings (Gear Icon Popover):
- Log Level: Verbosity selection (
INFO default, DEBUG -v, TRACE -vv).
- Min Test Stage: Test stage filter (
PREVIEW default, ALPHA -a, ALPHA_ONLY -x).
- Serial Number: Optional text override for device serial number (
-s).
Test Suite Checklist & Historical Runs:
- Search Box: Filter test cases by name.
- Checkboxes: Select specific test cases to execute (includes Select All / Deselect All toggles).
- Historical Status Display: Automatically loads and displays previous execution results (
Passed ✔, Failed ✘, Skipped ⊘) for each test case from disk artifacts (out/devices/).
- Artifact Inspection: Click any test item to open the artifact viewer modal and inspect historical
sequence.md summaries and sequence.log output.
- Direct AI Diagnosis: Click the inline Diagnose with Mantis AI brain icon on any failed test to switch directly to the Mantis screen for immediate triage of past run results.
3. The Mantis Screen
Used to inspect MQTT telemetry transactions and run AI-powered triage root cause analysis (RCA).
Main Header Inputs:
- Device: Dropdown selecting the target device.
- Debug Scenario: Dropdown selecting test execution runs. Automatically scans historical test artifacts from disk, displaying historical run verdicts (
Passed ✔, Failed ✘, Skipped ⊘), target project badges, and execution timestamps.
- Historical Run Triage: Historical test runs can be selected and triaged immediately using AI Diagnostics without triggering a fresh test execution.
Timeline Trace Tab:
- Interactive Node Selection: Click any node on the sequence flow graph to view raw
config, state, or events MQTT payloads in the JSON inspector with one-click clipboard copying.
- Diagnostic Playbook: Select triage depth:
Device Compliance Auditor (Standard OEM): Hardware/protocol compliance validation.Codebase Debugger (Advanced SWE): Source code debugging.
- Reference Successful Run (Optional): Text path or folder browser selection pointing to a baseline passing test run (e.g.,
out_prev). Enables differential diagnosis against a known working state.
- AI Auth Provider: Choose authentication mode:
Gemini API Key: Enter your Gemini API key in the password field.Google Vertex AI (ADC): Enter GCP Project ID and GCP Location (default: global).
- Fetch UDMIS Logs: Checkbox toggling automatic fetching of cloud-hosted UDMIS container logs when debugging remote GKE/cloud targets.
- Cloud Logging GCP Project ID: Text field for the target cloud project ID (auto-populated from historical test metadata).
- Run AI Triage Button: Launches the backend diagnostic agent on the selected scenario. Disabled for tests with a
Passed status to preserve API quota.