v0.6.8 — Live words, voices and dictations of any length

New

• See the words while people speak. During a Vietnamese (or auto-language) meeting, the words appear in grey within about a second of being said. The real transcript replaces them a little later. This runs on your computer and sends nothing. You can turn it off on the Meeting start screen.
• Wispra recognises speakers by their voice. Name a speaker once, or let them introduce themselves, and Wispra labels their voice in later meetings. Voices are kept only on this computer and encrypted, and only a voice's numbers are stored, never the audio. You can see every voice it knows on the Learned tab, forget one, forget all, or turn the feature off there. It labels a speaker only when it is sure; no name is better than a wrong one.
• Dictate for as long as you like. There is no time limit on a dictation anymore. Long recordings are sent in parts, whether you use Wispra Cloud or your own Groq key, and the text comes back whole.

Improved

• Nothing you said is lost. Every dictation is saved on your computer while you speak, before anything is sent. If it cannot be turned into text — no connection, a server error, or nothing heard — it stays under History → Recordings not transcribed yet. From there you can Try again or Delete it.
• Your vocabulary is spelled the way you wrote it. Names and terms in Custom vocabulary now come out exactly as you entered them in meetings, file transcriptions and dictations (for example "GitHub", "CapCut", "Nguyễn Văn A"). Wispra only respells words it actually heard; it never adds a word.

Changed

• Updates now come from a new download location (github.com/sinhgiang/wispra-releases). Nothing to do: this version moves there by itself, and later versions arrive as usual.

v0.6.7 — Long Cloud dictations, no invented sentences

Fixed

• Long dictations now work with Wispra Cloud. Dictating for more than about two and a half minutes on Wispra Cloud used to fail: the recording was too large for the server to accept in one go. Long recordings are now sent in parts, cut where you pause, and the text comes back whole and in order. A recording that still cannot be sent says so plainly, with its size.
• Errors from Wispra Cloud now say what went wrong. When the server refused a request — the monthly limit, a busy transcription service — Wispra showed a generic "Transcription failed". You now see the server's own reason.
• No more invented sentences in transcripts. On silence or background noise, a sentence nobody said ("Các từ và các nội dung cần giữ nguyên.") could appear in Meeting transcripts, dictations and file transcriptions. It came from the hints Wispra gave the speech model. Those hints are now only your own vocabulary words, and any sentence that just repeats them is removed. Nothing that was not said is added to your text.

Improved

• More room for the Meeting transcript. A finished recording's title and date now sit on the left with the tabs on the same row, so the transcript table starts higher. Copy transcript has moved into the table, at the end of the Transcript column.
• Unlimited Wispra Cloud accounts are shown as unlimited. The Account page used to show "Wispra Free · 0.0 / 30 min" for accounts without monthly limits. It now shows Wispra Cloud · Unlimited, with what you used this month and no limit bar. Free and Pro accounts show their limits as the server reports them.

v0.6.6 — The meeting table fills in while you talk

New

• The meeting table appears the moment you press Start. While you record, the Meeting tab shows the same four columns you know from finished recordings — time and speaker, topic, transcript and action items — and the transcript fills the middle column as people speak.
• Topics and action items appear while the meeting is still going on. When the conversation moves on to a new subject, the part that just finished gets its topic title and its action items a minute or two later — no need to wait until you press Stop. The part still being spoken is marked "Named when this part ends". After Stop, only what is left gets named.
• Speaker names from what is said. People who introduce themselves — including speakers who refer to themselves by their own name, as is common in Vietnamese ("Sơn đã test…") — are labelled with that name, but only when it is clear. You can click any label, including "You" and "Others", to rename that speaker everywhere, also while recording.

Improved

• More room for the transcript. Below the table there is now just the question box and Send. The chat's answers appear once you have asked something.
• Careful with your AI allowance. While recording, only finished parts of the meeting are sent to the AI, and each part only once. If the provider's daily limit is reached, Wispra switches to your backup model when you have one, and otherwise stops naming topics until the recording ends and tells you so.

Fixed

• Transcript printed two or three times. After switching to another tab and back during a session, a new recording could transcribe the same audio several times, so every passage appeared twice or three times in a row. Each recording now has exactly one recorder, and a repeated piece of audio is ignored.

v0.6.5

New

• Backup AI when the daily limit is reached. When Groq's gpt-oss-120b reaches its daily limit on your own key, mind maps, Transcript topics, summaries and posts carry on with Groq's smaller gpt-oss-20b (it has its own daily allowance), then with Cloudflare Workers AI if you set it up. Whatever a backup model wrote says so. A per-minute limit is still just waited out, and transcription always stays on Groq.
• Cloudflare Workers AI as a free backup. Settings → Account → "Backup when the AI reaches its daily limit": paste a Cloudflare Account ID and an API token with the Workers AI permission. They are tested before saving and never shown again. Use a Cloudflare account on the Workers Free plan: Wispra counts its own use and stops for the day before the free allocation (10,000 neurons, reset at 00:00 UTC) is used up.

Improved

• Older recordings no longer start AI by themselves. Only recordings made after this update get their Transcript topics and action items automatically. An older recording shows its transcript as before, with a "Create topics and action items" button, so simply opening old recordings no longer uses up the AI's daily allowance. Anything already made is kept.

v0.6.4

Fixed

• Daily limits are now called daily limits. When your AI key had used up its allowance for the day (Groq's free plan allows 200,000 tokens per day), Wispra said "Stopped by the AI provider's per-minute limit" and kept waiting. It now tells a daily limit from a per-minute one. A daily limit is no longer waited out: the Mind map, the Transcript topics, the Summary and the Website / social posts stop at once and say "The AI provider's daily limit is reached", with how much was used out of how much and when it resets. Wait until then and press Continue or Try again — or choose "Use Wispra Cloud" on the Account tab.
• A per-minute limit is still waited out and the request sent again, as before.
• Summary now also waits out the per-minute limit, and says why when it stops, instead of "No summary available for this session".

v0.6.3

New

• Transcript in four columns. A finished recording's Transcript tab now shows time and speaker, topic, transcript and action items side by side. Topics are written in bold with their time range and always cover the whole recording. Action items sit next to the topic they came up in, and a topic without any stays empty — nothing is made up. A By topic / List switch shows all action items as one list instead, and Wispra remembers your choice. Click an action item to jump to where it was said. The topics and action items are created by AI right after you press Stop (or the first time you open an older recording), in the session's Summary language.
• Speaker names. Names said in the recording ("I'm Linh", "over to Sơn") label the paragraphs they speak. When recording in "Both" mode, paragraphs are marked You or Others from which side the sound came from. Click a label to rename that speaker everywhere. There is no voice recognition and nothing about anyone's voice is stored.
• Choose where transcription and AI run. Settings → Account now lets you pick Use Wispra Cloud (counted toward your account's monthly minutes and AI allowance) or Use my own Groq API key (straight to Groq; nothing counted there). Switching never deletes a saved key.

Improved

• Mind map, Website and social posts are created only when you ask. Opening one of those tabs no longer starts the AI. An empty tab shows a button — "Create mind map", "Create website content", "Create Facebook post"… — and nothing happens until you press it. A mind map that is being built or was stopped part-way still shows its progress or Continue.
• Posts wait out the AI provider's per-minute limit. When the provider answers "too many requests" (for example while a mind map is being built with the same key), Website and social posts now wait and try again, and the tab says it is waiting. If the limit is still reached after a few minutes, the message says so instead of "check your connection/API key".
• The Settings window opens wider (1250 × 700), so the four columns fit; narrower windows fold the columns.

Fixed

• Transcribe tab with Wispra Cloud. Transcribing a file while using Wispra Cloud failed with "Not signed in". It now works when you are signed in. Files over 4 MB can't go through Wispra Cloud yet — the tab says so and suggests using your own key.

v0.6.2

Fixed

• Mind map stopped part-way with "The AI provider refused the request". Sometimes the AI's answer for one part of a long recording comes back as broken JSON (shown as "HTTP 400 — Failed to generate JSON"). Wispra treated that as a refusal and stopped. It now asks for that part again, then asks in a different way, and if needed cuts the part in two — so the map gets finished.
• If the answer is still unusable after all that, the message now says so ("The AI returned a result that could not be used") instead of telling you to check your API key or plan. The parts already done are kept, and "Continue" goes on from the part that failed.

If a mind map stopped with that error in 0.6.1: open its Mind map tab after updating and press Continue — the finished parts are still there.

v0.6.1

Fixed

• Mind map of a long recording could spin without ever finishing. With an AI key that only allows a small amount per minute, the map was thrown away at the first "too many requests" and started again from zero each time. Building the map is now a background job: it waits for the provider's limit and goes on, keeps every part it has finished, and continues from there after a failure or after Wispra was closed. You can leave the tab, open another recording or dictate while it runs.
• The Mind map tab shows real progress ("5 of 9 parts done"), and a small mark on the recording in the list tells you when a map is being built, is ready, or was not finished.
• A map that cannot be finished now stops with the reason and a "Continue" or "Try again" button instead of spinning.

Improved

• Wispra Cloud: a clear message when this month's AI allowance is used up. Summary, the Website / Facebook / Instagram / LinkedIn / X tabs, the chat about a recording and the Mind map now say that the allowance is used up, how much was used, when it resets, and what you can do — instead of a generic "check your connection/API key". Nothing is retried on its own; "Try again" asks once.
• Dictation keeps working when the allowance is used up: your words are typed as transcribed, without the AI cleanup, and Wispra tells you once.
• A question you asked in a recording's chat is put back in the box if it could not be answered, so you can send it again later.
• Settings → Account shows how much of the monthly AI allowance has been used, when the server reports it.

v0.6.0

New

• Mind map for meetings. A finished recording gets a "Mind map" tab next to Summary. Wispra builds the map with AI the first time you open it, then keeps it with the recording. Long recordings are outlined part by part, so the middle of a multi-hour meeting is not skipped. Click a numbered circle to open a branch; click a point and "Show in transcript" to jump to that part of the transcript. Export as PNG or copy as a Markdown outline. The map is written in the session's "Website & social posts" language.
• Cloud sync (opt-in, off by default). Settings → Account can keep a copy of your History, Meetings and learned words in your Wispra account, updated a few seconds after something changes.
• Connect your AI assistant. A private connection link in Settings → Account lets ChatGPT, Claude, Grok, Perplexity, Cursor, Claude Code or Codex read your synced dictations and meetings (read-only). The link is masked, can be given an expiry, and can be changed at any time.

Improved

• Summaries of long meetings are much more thorough: one section per topic instead of a few bullets.
• The window opens wider so a meeting's tabs fit on one line; the Copy button now sits next to the session title.

Fixed

• When writing a Website / Facebook / Instagram / LinkedIn / X post failed, the tab kept asking the AI again and again. It now stops, shows the error and offers "Try again".
• The garbled-text filter no longer drops a whole sentence when only part of it is gibberish.