Vocando — Admin dashboard
Users
| Last name | First name | Status | Plan | Usage | Created | Organisation |
|---|
Corporate Identity the colours and elements of every page (app, console, admin, website) — click a colour to copy its value
Logo
Wordmark: VOCANDO — the microphone replaces the first O, the baseline runs from the microphone's stand to a pencil; letters #14171a on white, #e8ecef on #0f1316; microphone, line and pencil in the brand green (raster renderings — use each on its own background). Icon: white microphone, baseline and pencil on brand green; rounded 112/512 (22 %) for websites and cards, the square master for app stores. Brand mark: the simplified microphone below 48 px (navigation, favicon) — mint on dark green in dark mode. The full kit with every size is in the repository under brand/vocando/.
Colours
Light
Dark
Chart colours (admin charts: four series, one sequential scale)
Type and shapes
Elements
Live samples with the pages' own stylesheet (/app/ui.css): what you see here is what the console and the app show.
Overview every surface we run · status, page views of the last 7 days, its key figures
Metrics
Summary this week vs. previous week · server-wide
Acquisition funnel
Funnel class A
Views (page loads, counted per day and path since 2026-09-25 — no visitor data) then accounts created in the range. Approved = approved_at set (the first activation; backfilled with the creation date for accounts active before 2026-09-25). Activated = 3 active days in the first week after approval; Still active = activated accounts with a dictation this week. Dashed stages are not instrumented (M-1, G-3).
Table view
Time to first dictation class B
Table view
Sign-ups by source class B
Table view
Conversion by segment class B
Trial pool class A
Page views per day class A
Cohorts and retention share of each weekly sign-up cohort that dictated in week n after sign-up
Weekly retention by cohort class B
Table view
Quality and maintenance latency, errors, fallbacks, versions, corrections
Per account class A
Latency
Cleanup route
Devices, languages and learning
One account's figures as its owner's console showed them until 2026-10-06 (GET /v1/stats?user_id=, admins only): LLM skipped = short dictations that need no cleanup (under 50 characters, or ≤ 15 clean words); fallbacks = the model timed out and the raw transcript was pasted; client errors = error events from the account's devices. Server = audio in → cleaned text out; end to paste = measured by the client from the end of speech. Cloud cost at the provider's list price; dictations before 2026-10-01 carry no token counts.
Latency per step class A
Table view
Error and fallback rates class A
Per 100 dictations. LLM and STT shares come from the usage ledger (every account; transcript rows of opted-in accounts for weeks before the ledger started); client errors from the error event of every client.
Table view
Client versions class B
Table view
Corrections per 100 dictations class A
Console corrections plus in-place edits observed by the client (learn-from-edits), by the week they were stored. A rising line after a model or prompt change is the quality alarm.
Table view
Learn-from-edits loop class A
Outcome of every observed paste (learn_observed) in the range and the share of attempts where the field was not readable (learn_unavailable).
Table view
Paste and recording hygiene class A
Windows clients report which paste strategy worked (paste_strategy); macOS has no such field yet (Q-8). no_speech = a press that recorded no speech.
Accounts per account · class B · signals are derived from usage, plans do not exist yet (G-3)
All accounts
Click a row (or its Admin link) to open the account in the console's Admin tab. Signals: Heavy user = ≥ 2,000 words last week · At risk = last 2 weeks under 50 % of the previous 4-week average, or no dictation for 14 days · Never dictated = active account without a dictation · New = created in the last 14 days · Retains audio = opted in. Quota, plan and seat signals need G-3 / M-3.
Cloud cost every paid cloud call — speech-to-text, cleanup, meetings · per day, account and provider · from the cost ledger (D-061)
Cost per day class A
Table view
Per account class B
The selected accounts over the range. Meetings = segments and summaries.
Per provider class A
Entries and models
Provider = the company that bills (from the endpoint's address); entry = the line in the cloud config. Confirmed = calls whose cost the provider's usage log has replaced (Soniox).
Provider bills vs. the ledger class A
Cloud cleanup details tokens and cost of the cloud LLM route (D-039/D-041) · every account · from the usage ledger
Cloud spend per week class A
Table view
Per provider class A
Cloud-cleaned dictations per provider entry of the cloud config, all time; the model comes from the transcript rows (models.llm) or the selected entry. LLM p50 is the provider call's wall time.
Table view
Per account class B
All time, every account with a ledger row. Cloud share = cloud-cleaned dictations of all the account's dictations (short ones skip the cleanup and stay local). Per dictation and per 1,000 words are taken over the cloud dictations that carry a cost.
Table view
Governance GDPR evidence, access, devices, capacity, models
Retention states class B
Audio is kept only for accounts that opted in (users.audio_retention_days > 0). Transcript text is stored only with transcript_log on; usage counts (words, seconds, timings — never text) exist for every account in the usage ledger.
Table view
Audio on disk class B
Deletions class B
Every account deletion with its evidence (audit row account_deleted): rows removed per table, keys revoked, audio files and bytes removed from data/audio/<username>/. A red mark means the directory could not be removed — it then shows as an orphan above until cleaned up.
Access and approval log class B
Device inventory class B
Table view
Capacity class A
Uptime class A
Incidents
Models in use class A
Shares per dictation in the range: STT engine, prompt, client version and platform from the usage ledger (every account), LLM and gateway from transcripts.models (opted-in accounts); reachability from /v1/health. A share of ONNX rows > 0 means the MLX sidecar was down.
Learning and data assets inputs for product improvement · internal only
Corpus and correction diffs are class C (derived from transcript or audio content): internal use on opted-in accounts only; never shown to customers, never used in marketing. The counts above are the aggregate alternative.
Privacy classes used on this page
Counts and durations across accounts. Safe for any audience, incl. a public status page.
Devices, timestamps, settings, word counts. Operator and the account's own admin.
Needs stored transcript text or audio. Internal only; would contradict "no audio retained, nothing leaves the network" if exposed or marketed.