---
title: "QWED Dashboard: review what your AI verification decided"
description: "QWED Dashboard shows every decision your QWED server recorded, with its evidence: the ledger, each record, signature and audit-chain checks. Open source, Apache-2.0."
url: https://qwedai.tech/platform/dashboard
---

[Platform](https://qwedai.tech/ecosystem) · QWED Dashboard

# See what QWED decided, and why.

QWED Dashboard shows every decision your QWED server recorded: what was checked, what the verifier said, whether the action went ahead, and the record behind it. It runs against your own server.

[Run it with sample data](#run) [View on GitHub](https://github.com/QWED-AI/qwed-dashboard)

Early, version 0.x

-   **Open source** Apache-2.0, on GitHub.
-   **Your server** It talks only to the QWED server you run.
-   **Try it in a minute** Sample data runs in your browser; nothing is sent anywhere.

## The overview

[Image: The dashboard's overview: counts of records by verdict, a chart of verdicts per hour, and a list of what needs attention.]

The overview, with the dashboard’s sample data. No real organisation, key or record is shown.

## What it shows today

Everything comes from what your QWED server stored. The dashboard reads it; it does not decide anything.

-   ### The decision ledger

    Every verification your organisation ran, newest first: what was asked, what the verifier said, the verdict, and whether the action was allowed to go ahead. Filter it, search it, export it as CSV.

-   ### Each record, as stored

    The request, the verifier’s words and the result exactly as the server kept it, with a copy you can download when someone asks for evidence.

-   ### Signatures, checked in your browser

    Where the server signed a result, the record page checks that signature with the server’s public keys, and says whether it matches and has not expired.

-   ### The audit chain

    Checks the signature of every audit entry and the link between each one and the next, so an entry that was changed or removed shows up.

-   ### What needs attention

    Records that were blocked, could not be proven, or were not admitted, grouped so a person can review them first.

-   ### Keys, people and agents

    API keys and when each was last used, members and their roles, registered agents, and security events such as a leaked key being revoked.

[Image: The decision ledger: rows of records, each with its time, verdict and admission, request, verifier and API key.]

The decision ledger.

[Image: One record: VERIFIED and ADMIT, the claim, the stored result and its checks.]

One record, as the server stored it.

## The words on a record, in plain English

Written for the people who read the records, not only the people who write the code.

VERIFIED

The check ran to a proof. It can prove a claim true or prove it false; the reason says which.

UNVERIFIABLE

The check could not reach a proof either way, so nothing is taken on trust.

Admission: ADMIT or DENY

Whether the action was allowed to go ahead. Anything not proven safe is denied.

proof\_ref

A fingerprint of the evidence behind a result, so the same evidence can be found and checked again later.

Attestation

The server’s digital signature on a result: it shows who issued it and that it has not been changed since.

Audit chain

Each stored entry is signed and linked to the one before it, like numbered pages, so a missing or edited page shows.

## What it never does

A review tool is only useful if it never puts words in the verifier’s mouth.

-   It never works out a verdict or an admission itself. It shows what the QWED server decided and stored.

-   Where the server stored no value, it says “not recorded” instead of filling one in.

-   A request, or a message a caller supplied, is labelled as the caller’s and kept apart from what the verifier said.

-   It does not run or replay the actions that were verified.

In development · no dates promised

## What comes next

The checks already decide. These are for the decisions a person still has to make.

-   ### A review queue

    Outputs that could not be proven wait for a person, instead of going ahead.

-   ### Signed human overrides

    A human override is recorded as its own signed status, with who made it, when and why. It is never shown as VERIFIED.

-   ### An audit and proof view

    For auditors and leadership: every decision with its evidence and its record, in one place.

## Run it with sample data

You need Node.js 22.22.2 or later. Then open `http://localhost:5173` and choose **Explore with sample data**. No QWED server is needed for that.

Terminal

```
$ git clone https://github.com/QWED-AI/qwed-dashboard
$ cd qwed-dashboard
$ npm ci
$ npm run dev
```

To connect your own QWED server, copy `.env.example` to `.env` and set `QWED_API_URL`. A new server has no user yet: [create the first owner](https://github.com/QWED-AI/qwed-dashboard#first-sign-in-on-a-new-server).

### Questions

Is it free?

QWED Dashboard is open source under the Apache-2.0 licence. You can run it, read it and change it.

Does it send our data anywhere?

It talks only to your own QWED server. No hosted QWED service sits in the path. With sample data, nothing leaves your browser.

Can I use it today?

Yes, from source: it is early, version 0.x, and has no packaged release yet. A release follows the dashboard work in qwed-verification.

Who is it for?

Anyone who has to answer for what an AI system did: the engineers who run QWED, and the reviewers, auditors and leaders who read its records.

## Running AI decisions that need a person’s sign-off?

We are looking for design partners: teams who will use the dashboard early and tell us what their reviewers and auditors need from it.

[Talk to us](https://qwedai.tech/contact)
