by Innominds
RELEASE 2026.11.18.2 · 18 NOVEMBER 2026

The TruVelocity UI: all six steps, on your real agents and runs

tv ui opens the TEVV wizard in a browser over your project: from an agent's context and test plan to starting runs, the release decision and watching it after release. Everything comes from the engine and your project files; there is no separate database.

6 of 6wizard steps live, plus Home
1039automated tests pass
Localtoken-protected, this machine only
0new services to run: one command
What's new

The wizard, step by step

Each step reads the same files and runs the same engine functions as the command line, so the UI and tv commands always agree.

Home

Every agent in tv-ui.yaml: verdict, defects, pass-rate trend and risk by pillar from its latest full release runs, recent runs, and what needs attention. Click an agent to open its workspace at the step it is on.

1Context

The agent's purpose, users, systems, risks and intents.

  • Import the Context workbook (.xlsx) or its YAML, or answer in the page
  • Saved per agent in the project; Essential questions first

2Catalog

The agent's real test plan, read from its catalog.

  • Each of the 71 test types: in plan, runnable, not applicable or not built yet
  • Suites with questions, turns and estimated time
  • Add or remove library packs with a preview of what they add; the catalog file is checked before it is written
  • Tick types in or out, with a reason for gaps; confirm the plan
  • A test plan document to keep or print

3Configure

How TruVelocity reaches the agent.

  • Its environment's settings with current values; secrets only as set or not set
  • The preset a run starts from, and what the latest release run used (model, versions, deployments)
  • Check connection now: every address is called; the result sets the step

4Conduct

Start real runs from the browser.

  • Choose suites, wordings, repeats, workers and mode within the preset's limits, with a time estimate
  • Follow progress every 2 seconds, see the output, cancel, open the report
  • Needs tv ui --allow-runs; runs only presets from tv-console.yaml

5Conclude

The release decision, with evidence.

  • The run against the release gate, rule by rule, and its defects
  • A baseline: failures of an accepted earlier run count as known
  • Go, Go with conditions or No-go, recorded with who and when; a Go below the bar needs a justification
  • An evidence pack (zip) for auditors

6Continuous

Keep watching after release.

  • Trend of release runs and what changed (model, prompts, tools)
  • Drift: regressions, fixes and behaviour agreement between the last two comparable runs
  • Alert rules checked against the latest run
  • A nightly GitHub Actions or Azure Pipelines file per agent, or Windows and cron lines
Get started

Install and open the UI

The UI ships in the same package. The ui extra adds the workbook reader for Context import.

Upgrade or install

pip install -U "truvelocity-tevv[ui]"

Open the UI in your project folder (where tv-ui.yaml and tv-console.yaml are)

tv ui --allow-runs --open

Or try the clickable prototype with sample agents (no project needed)

tv ui --demo --open

From the repository, scripts\dev_deploy.bat (or .sh) creates the virtual environment, installs everything and opens the UI; scripts\docker_deploy runs it in a container.

Safety

Built like the console

Local and token-protected

Listens on 127.0.0.1; a one-time URL token becomes an HttpOnly, SameSite=Strict cookie, and every change carries a form token.

No arbitrary commands

Runs start only from presets in tv-console.yaml, only with --allow-runs, only on this machine. The page chooses options within the preset's limits.

Secrets stay out of the page

Tokens and passwords are reported as set or not set. With --allow-runs, the project's .env.session tokens reach runs; your environment and .env win.

Files checked before writing

Catalog edits load and validate with the engine first, and are refused if the file changed meanwhile. Pipelines are never written over existing ones.

Fixes

Also in this release

AreaChange
ConsoleA started run's progress (episodes done, passed, failed) counted 0 on Windows; output line endings are now normalised.
BuildThe type check of the test suite failed in CI on a helper imported under two module names; fixed.
Deploy scriptsdev_deploy starts the UI with --allow-runs.
Not yet

Coming in later releases

Sign-off by several approvers · editing environment settings and calibrating evaluators in the UI · per-metric targets and Gate thresholds · writing new questions and custom test types · comparing runs and shadow runs from Conduct · sending alerts by email or chat.
Quality

How this release was checked

Lint (ruff, format, mypy strict over 184 files, the core import rule) and the full suite: 1039 passed, 6 skipped. Each step was also exercised end to end in a browser at desktop and phone width, including a real run started, followed and cancelled from Conduct.