by Innominds
RELEASE 2026.11.19.1 · 19 NOVEMBER 2026

Ready to show: an offline demo project, a realistic release bar and a UI written for leadership

tv demo-project gives anyone with the package a working demo in seconds: three agents with a real run history, one that meets the release bar and one that doesn't. The UI now leads with plain summaries and one pass-rate vocabulary, and a new runbook shows how to present it.

4commands from pip install to a running demo
0Docker or language model needed for the demo
2 + 1Go and No-go agents in the demo story
1060automated tests pass
What's new

A demo anyone can run

Every verdict, defect and count in the demo is real engine output, from runs against scripted builds of three agents.

tv demo-project

Writes a project that tv ui opens as is: three agents, their catalogs and Context, presets, the release bar and four weeks of release runs. Same files every time; about 3 seconds.

Claims back-office agent

Meets the bar on every build: Go. The release decision and evidence pack to show.

Customer chat assistant

Build 2.4.0 asks customers to approve paying a balance they only asked about (critical): No-go, 100% to 67% of cases passed every try. Continuous shows the regressions, the version change and the alerts.

Member service desk

A billing outage cascaded and a handoff passed too many permissions on in 1.0.0; fixed by 1.2.0, Go at 1.2.1. A trend that improves.

Runs from Conduct

The presets run each agent's latest build, so a run started in the browser reproduces its verdict in under a minute (TV_DEMO_PACE=0: instant).

Guides

A UI demo runbook (what to click, say and show on each screen, a 19- and a 5-minute path, leadership questions) and a local install guide for Windows, macOS and Linux, in docs/demo.

Release decisions

A release bar an AI agent can meet

The engine's default bar (no critical defect, every case passing every try) is unreachable for an agent built on a language model, so every agent was a No-go. Projects now carry their own bar in truvelocity-tevv-gate.yaml; the repository and the demo project use:

no critical defect · at most 2 major · at least 85% of cases passing every try · evidence seams clean · parity holding
The UI

Written for the people who decide

A landing page

The design's sign-in page opens each session, with the eight agent types TruVelocity assures around it; Explore as demo user (or any sign-in) opens Home. Decisions carry the name set in TV_UI_USER.

Home at a glance

Three figures (release-ready, open critical defects, decisions waiting), the portfolio with a Verdicts / Risk by pillar switch, what needs attention and the latest five runs: one screen. Presets moved to Conduct's run form; standards to Conclude.

One vocabulary

Cases passed every try (the gate's measure) leads on every screen, with Tries passed beside it. The same run used to read 69%, 87% and 91% under "Pass rate".

Summary first

Configure opens with "At a glance"; settings and versions fold under technical details. Catalog's deep tables and Continuous's pipeline files fold too.

Defects in words

"Asked to approve an action this step should not take", with how often ("2 of 13 tries"); the raw check message underneath.

Start from a preset

Conduct's run form starts from Smoke (one wording, minutes), Pre-release gate or Reliability run. What isn't built is hidden, not greyed out.

No simulated figures

The live UI shows only this project's data: standards mapping says it comes later; Catalog's counts name their base (71 engine test types, 142 in the catalogue); the release bar card shows the real bar.

Details

A display name for the greeting (TV_UI_USER or user: in tv-ui.yaml); all six steps fit on a phone; a refused run says "Not started"; the Context assistant answers "what isn't covered yet?".

Fixes

Install and setup

AreaChange
tv newWrites tv-ui.yaml, so tv ui works in a new project; its .gitignore names tv_runs/ and .tv-ui/.
WindowsA second tv ui or tv console on a port in use stops with an error instead of sharing the port.
Entry pointpython -m truvelocity_tevv runs the tv command.
ConfigureThe connection check builds an agent that runs inside TruVelocity instead of failing for having no address.
PagesThe no-key page names the command that started it; Conduct offers presets that match the agent.
Get started

From install to demo

Install or upgrade

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

Make the demo project

tv demo-project

Go into it

cd truvelocity-demo

Open the UI

tv ui --allow-runs --open
Not yet

Coming in later releases

Adding an agent from the UI · sign-off by several approvers · editing settings and calibrating evaluators in the UI · per-metric targets · comparing runs and shadow runs from Conduct · mapping evidence to OWASP, NIST AI RMF, ISO/IEC 42001 and the EU AI Act · sending alerts by email or chat.
Quality

How this release was checked

Lint (ruff, format, mypy strict over 188 files, the core import rule) and the full suite with the pilot tests: 1060 passed, none failed. The UI was reviewed screen by screen as a client would see it and checked in a browser at desktop and phone width; the demo project was generated from a built wheel in a clean environment and a run started from Conduct finished.