CyberScribe Release Notes

What's new in each release, written for Game Masters. Newest first.


v0.4.2 — 2026-09-18

Help for players finding their character's voice.

New

  • Character Settings examples and live preview — the player Character
    Settings page now has a Preview card showing the rendered contents of your
    own character bio and narration sample, alongside an Example card with a
    worked example to write against. A "See Example" button at the top jumps
    straight to it.

v0.4.1 — 2026-09-18

More reliable transcription, styled player summaries, and a new preset.

New

  • User Summary styles — the player "Character's Perspective" card is now
    "User Summary" with a style picker. First-person character voice stays the
    default (and recommended), with all seven GM summary styles available under
    "Other styles" — including Structured Session Report, also recommended. The
    card shows which style was used and remembers it on revisit; regenerating
    with a new style overwrites.
  • Traveller preset — a Mongoose Traveller 2e summary style joins the
    game-system presets, with task checks, Effect, Boon/Bane, careers, starships,
    starports, and trade vocabulary.

Improvements

  • Cleaner character voices — first-person summaries no longer quote dice,
    DCs, damage counts, or initiative scores from table chatter; out-of-character
    talk stays out of the story.
  • Clearer task queue — the admin Background Task Queue now shows when each
    transcription job last made progress, so a stalled job is obvious at a glance.

Fixes

  • Interrupted transcriptions resume — if the worker restarts mid-job, the
    session is automatically re-queued instead of sticking forever (previously
    only a manual admin cancel could unstick it). Jobs that fail repeatedly stop
    after three attempts and surface an error.
  • Long recordings diarize in pieces — multi-hour audio is now diarized in
    overlapping windows, so peak memory stays bounded and speaker labels stay
    consistent across the whole file. Transcription also streams one file at a
    time and warns in the log if a phase runs far slower than realtime.

v0.4.0 — 2026-09-13

First-person recaps for players, a campaign-wide NPC compendium, and cleaner
summaries.

New

  • Player first-person summaries — your players can now generate a recap told
    from their character's point of view. Each player sets a character bio and an
    example of how they sound in Character Settings, keeps private notes per
    session, and generates a first-person perspective alongside the GM summary.
    The recap draws on their notes for interiority, stays within what the
    character actually witnessed or was told, and never reveals GM-only material.
  • Campaign NPC compendium — CyberScribe now pulls the named NPCs out of
    each session and collects them into a browsable compendium per campaign,
    shared by GMs and players. Each session page has an "Extract NPCs" card (or
    enable auto-extract in GM, campaign, or session settings), each session lists
    its own NPCs, and the compendium page offers live search, most-recent /
    most-seen sorting, and links from every appearance back to its session.
    GMs can rename entries, merge duplicates, and edit or delete blurbs to keep
    the cast list tidy.

Improvements

  • Style-appropriate die rolls — story-style presets (Narrative,
    Player-Focused, Quest-Focused, and the player first-person view) now render
    rolls, checks, and damage as outcomes rather than numbers, while tactical
    presets (Combat-Focused, Structured, Bullet Points, Session Prep) keep the
    numeric detail.
  • Roomier prompt boxes — the prompt-preset text areas across all settings
    pages and the Prompt Lab are now roughly three times taller, so long prompts
    are easier to read and edit.

v0.3.2 — 2026-08-22

A more forgiving Prompt Lab, prompt exports, and small polish.

New

  • Export Prompt — the Summary Prompts page for a session now has an "Export
    Prompt" button that downloads the exact instructions CyberScribe sends to the
    AI — the full five-level prompt cascade plus your transcript, notes, roster,
    and previous summary — as a Markdown file. Handy for reproducing summaries in
    another LLM or keeping an archive of what was sent.

Improvements

  • Prompt Lab auto-saves everything — every field in the Prompt Lab (all eight
    prompts, the test transcript and notes, even edited generated summaries and
    hooks) saves itself about a second after you stop typing, with a subtle
    "Saving… / Saved" indicator near the top of the page. Navigating away no longer
    loses your work — and the Save buttons are gone because you don't need them.
  • Populate now sticks — the Populate buttons in the Prompt Lab persist what
    they fetch, so your sample transcript/notes/prompts survive a page refresh.
  • Auto-save pauses during generation — typing while a test summary or plot
    hooks are generating can't clobber the job; edits resume saving when it
    finishes.
  • Tidier settings pages — card spacing across Admin Settings, Game Master
    Settings, Campaign/Session Settings, and the game-system forms now matches the
    Docs grid instead of cards sitting edge-to-edge.

v0.3.1 — 2026-08-22

Cleaner AI prompts and small workflow fixes.

Improvements

  • Smarter, system-agnostic summaries — campaign and session prompt presets now use generic examples (Thorn/Mira/threshold instead of setting-specific names) and explicit "don't invent" guidance, so generated summaries stay closer to what actually happened and don't leak terms from other game systems.
  • Clearer "Number of Speakers" — the upload card now explains that telling CyberScribe how many people were in the room gives you cleaner "Speaker 1 / Speaker 2" labels.

Fixes

  • Prompt Lab Populate — the "Populate from Session" buttons now reliably copy your latest transcript and notes into the sandbox (and "Populate Prompts" restores all eight prompt fields).
  • More complete backups — dump-db / populate-db now includes the Prompt Lab sandbox and each session's speaker count; transient queue state is still excluded by design.

v0.3.0 — 2026-08-21

A sandbox for your AI prompts, and a fuller set of in-app docs.

New

  • Prompt Lab — a private sandbox to try out your AI prompts without touching
    your real settings. Paste a sample transcript and notes (or pull them from your
    latest session), adjust the prompts at any level, and generate a test summary or
    plot hooks to see how they read before you commit. It's all isolated to you —
    your real campaigns and other GMs are unaffected.
  • In-app documentation — the Release Notes, Privacy Policy, and FAQ are now
    browsable from the Docs page, with a Quick Links section to jump straight to the
    guides.

Improvements

  • Cleaner docs pages — consistent card spacing, a clearer numbered Guides list,
    and tighter wording in the Privacy Policy and FAQ.

v0.2.2 — 2026-08-21

A few small refinements to smooth out the summary workflow.

Improvements

  • Edit Transcript is now always available — even for a session with no upload
    yet, so you can paste in a transcript from elsewhere.
  • You can see which AI model is in use — the summary and plot-hook settings
    pages now show the model name alongside temperature and the rest.

Changes

  • Tighter default summaries — the summary temperature dropped from 0.4 to
    0.25, so summaries stay closer to what actually happened at the table.

v0.2.1 — 2026-08-21

Big, heavy audio files can finally be uploaded without fear.

New

  • Resumable chunked uploads — large recordings are sent in small pieces, so if
    your connection hiccups the upload picks up where it left off instead of
    starting over.
  • Upload progress bar — live progress, speed (MB/s), an ETA, and cancel /
    retry controls.
  • Higher upload limit — now 2 GB by default (configurable), so multi-hour
    recordings fit comfortably.

v0.2.0 — 2026-08-20

A brand-new transcription engine under the hood.

New

  • More accurate speech-to-text — CyberScribe now runs on the Parakeet TDT
    model for crisper transcripts.
  • Automatic speaker detection — your recording is diarized so the transcript
    shows Speaker 1, Speaker 2, and so on instead of one long block.
  • Optional "Number of Speakers" — tell it how many people were in the room at
    upload time for even cleaner labels.
  • Live progress — watch transcription and diarization progress in real time.

Fixes

  • Very long recordings no longer crash mid-transcription.
  • Speaker labels now land more accurately on line boundaries.

v0.1.2 — 2026-08-19

New

  • Rename a Campaign — change a campaign's name right from Campaign Settings.

Fixes

  • Prompt presets whose names contain a colon now load correctly.

v0.1.1 — 2026-08-19

More clarity and a fresh preset.

New

  • Getting Started walkthrough — a step-by-step guide from an empty dashboard
    to your first generated summary, including how to add player accounts and
    characters.
  • Discord docs — a dedicated page on webhooks, public vs. hidden channels, and
    the auto-post workflow.
  • Mage: The Ascension summary preset.

Improvements

  • Clearer labels — "Auto-Generate" toggles now say on Transcription
    Completion
    , and the two Discord webhook fields now say which channel each feeds.
  • Session page reflow — the page now flows top-to-bottom in the order you work
    (details → players → upload → transcripts → notes → summary → hooks).

v0.1.0 — 2026-08-16

The first public release of CyberScribe.

What's inside

  • A full GM dashboard to organise Game Systems → Campaigns → Sessions.
  • Audio → transcript → AI summary + plot hooks in one workflow.
  • A 5-level AI prompt cascade with reusable prompt presets.
  • Player accounts and player characters linked to your campaigns.
  • Discord integration to post summaries and plot hooks to a channel.
  • A refreshed, animated interface with a full set of documentation.