Run diagnostics
This report stays in memory until the next run or reload. It includes request steps, timing, status codes, and redacted errors. Review it before sharing.
Audio source
Add an .mp3 file directly from your device, then choose “Transcribe and write Markdown”. WAV, M4A, MP4, MPEG, MPGA, OGG, WebM, and FLAC are also supported.
With Groq, a public URL is sent to Groq first. If Groq cannot fetch a redirecting or oversized file, this browser tries to download it for upload or chunking. Browser downloads need access from the media host (CORS).
Open media link to save the audio fileSupported files within the chunk size limit upload as they are. Larger MP3 files are cut at audio frame boundaries without decoding or reencoding. Other large local files still need browser conversion and may use substantial memory. Groq URL requests remain subject to its account file limit.
Existing transcript
A Markdown file must contain a “## Full transcript” section. No audio is sent or reprocessed.
This rewrites the summary for an already-transcribed file, the way the original script’s --resummarize flag worked.
Stored only in this browser (localStorage). Last saved: never. Export or import everything from Settings JSON.
Appearance
Saved with your other settings and shared through the JSON export.
OpenAI-compatible endpoints
Set each API's base URL (including /v1). The app calls /audio/transcriptions and /chat/completions beneath it. The two endpoints can be different providers. Direct browser access requires CORS.
Choose a model and response format your endpoint supports. Segment stitching is best with verbose_json; some models only support json.
If a summary fails, the app tries up to three rounds. A countdown shows the pause between rounds, starting at 10 seconds and allowing up to 60 seconds if the service asks for more time.
Private-use note: keys are stored unencrypted in this browser and included in JSON exports. A static Cloudflare Pages deployment does not hide them. Only enter endpoints and keys you trust; do not use this on a shared device.
Chunking and language
Two-letter code such as en or sv, or blank to auto-detect.
Summary prompt
Sent to the summary API with every run. Edit it to change the sections, tone, or language rules.
Leave blank to send no system message.
Placeholders: {transcript} (required), {language_hint}, {language}. The transcript is inserted last and is clipped to the context limit below.
Summary and output
Uses the same summary request and language. The title appears in the result and Markdown; the download file name stays based on your source or chosen file name.
Write YAML without the --- delimiters. Placeholders: {title}, {source}, {date}, {duration}, {transcription_model}, {summary_model}, {summary_status}, {language}. Values are quoted automatically: use title: {title} without extra quotes. Add your own fields, such as author: "", tags: [], or aliases: [].
Longer transcripts are clipped to this limit for the summary request.
Shows a downloadable report after each run, including failed requests. The app does not add audio, transcript text, prompts, or API keys to it. Provider error messages may contain details you should review before sharing. Turn this on, repeat the problem, then open Run diagnostics.
Saved results live in this browser's persistent storage. Audio and raw API responses are not saved there. Settings JSON contains preferences and API keys, not saved results. Provider requests are sent directly from this page; browser CORS rules may prevent direct calls.
About
Pod-2-notes v0.6.4. This is the app release version, not the settings schema version. It changes with releases, not when you save or refresh settings. Refresh while online to check for new deployments.
Transcripts and summaries saved on this device remain available after a reload and while offline. They are separate from Settings JSON. The browser may remove them if site data is cleared or storage space runs low; download important Markdown files too.
Loading saved results…