Settings
A complete reference for every option in the Harvestry Settings panel.
Opening Settings
Open the Settings panel by clicking the gear icon (⚙) in the main window's toolbar, beside the Import button, or use the menu bar: Harvestry → Settings (⌘ ,).
Settings is organized into tabs: General (language, screenshots, export location, and updates), Tools (transcription model and video downloader), Consolidation, Markdown, Licensing, Cache, and Privacy. The sections below follow the options most people change.
Tools Tab: Transcription Model
Configure the Whisper speech recognition models used for transcription.
Whisper Models
Five model sizes are listed: Tiny (~75 MB), Base (~150 MB), Small (~490 MB), Medium (~1.5 GB), and Large Turbo (~1.6 GB). Each row shows the model's download status:
- Not Downloaded — A Download button is shown. Click it to begin downloading from Hugging Face.
- Downloading — A progress bar replaces the button. Downloads can be cancelled.
- Ready — A green checkmark. The model can be used for transcription.
- Update Available — An update badge and an Update button. The local model weights are outdated.
Default Model
One downloaded model is marked as the default. This model is pre-selected when you start processing a new lecture. Click Set as Default on any downloaded model row to change the default. Individual lectures can override this selection.
Spoken Language for New Lectures
Below the models, Spoken language for new lectures is set to Auto-detect by default. If your lectures are all in one language, choose it here: every lecture you import from then on starts with that language instead of detecting it, which avoids the occasional wrong guess (see Choosing the spoken language). You can still change the language for any single lecture before processing it, and lectures already in your library keep their own setting.
Update Check
Harvestry checks for model updates automatically every 24 hours by comparing the local snapshot SHA against the Hugging Face Hub HEAD. Click Check Now to trigger an immediate check.
General Tab: Screenshots
Configure how Harvestry captures screenshots during processing.
Max Interval Slider
The Max Interval slider controls the maximum time that can pass between consecutive screenshots, regardless of scene activity.
- Range: 15 seconds – 120 seconds
- Default: 30 seconds
- Lower values (15–20 s) produce more screenshots, useful for fast-paced content like code walkthroughs.
- Higher values (60–120 s) produce fewer screenshots, useful for slow presentations where visual changes are infrequent.
This setting applies to all future processing runs. Existing processed lectures are not affected; they must be reprocessed to use the new interval.
General Tab: Export Location
Controls where Harvestry saves exported study documents: each lecture's folder with its index.html, stylesheet, extracted audio, and images.
Save exports to iCloud Drive
When iCloud Drive is available, this toggle chooses between two locations:
- On (the default): exports go to Harvestry's folder in iCloud Drive and sync across your Macs.
- Off: exports go to
~/HarvestryLibrary/and stay on this Mac only.
Changing the toggle only affects where new exports are saved. Existing exports stay where they are. To move one, right-click the lecture in the sidebar and choose Move Export to iCloud Drive or Move Export to Local Storage (see Moving an export).
If iCloud Drive isn't available on your Mac, the toggle isn't shown, and exports are always saved to ~/HarvestryLibrary/.
Current location and Open in Finder
Below the toggle, the current export folder's path is shown with an icon: a cloud for iCloud Drive, a disk for local storage. Click Open in Finder to open that folder. Harvestry doesn't offer a way to pick a custom export folder; the location is always iCloud Drive or ~/HarvestryLibrary/, depending on the toggle.
General Tab: Updates
Check for updates automatically is on by default: Harvestry looks for a new version each time it opens, and once a day while it's running. When there is one, it tells you — nothing is installed until you choose to. Turn the switch off to check only when you ask. Check Now, like Harvestry → Check for Updates…, checks straight away.
Tools Tab: Video Downloader
Manage the yt-dlp and ffmpeg binaries used for URL download and video format conversion.
yt-dlp Status
The yt-dlp row shows the currently installed version and one of two states:
- Up to date — Green checkmark. The installed version matches the latest release on GitHub.
- Update available — Gray arrow. A newer release is available.
ffmpeg Status
The ffmpeg row shows the installed version and the same two states as yt-dlp. Harvestry installs a native Apple Silicon build of ffmpeg, so it works whether or not Rosetta is installed. If you already have ffmpeg from Homebrew, Harvestry can use that instead.
Download Quality
Chooses the highest resolution URL imports download:
- 1080p (recommended) — the default. Fast downloads that are ready to process right away.
- Best available (up to 4K) — sharper slides from high-resolution videos. Downloads are larger and slower, videos above 1080p are converted before processing, and each lecture uses more iCloud storage.
The setting applies to new downloads. To get a sharper copy of a lecture you've already imported, change the setting and re-download it. See Video Quality and Format.
Check for Updates
One Check for Updates button refreshes both yt-dlp and ffmpeg simultaneously. The check queries:
- yt-dlp: GitHub releases API at
yt-dlp/yt-dlp - ffmpeg: the latest native Apple Silicon release from
ffmpeg.martin-riedl.de. Each download is checked against its published checksum and its developer signature before it's installed.
If updates are found, an Update button appears next to each row. Updates install to ~/Library/Application Support/io.archetyp.Harvestry/ without requiring your password.
Consolidation Tab
Choose the consolidation provider, configure API keys and the Ollama endpoint, set the language the notes are written in, and edit the system prompt.
Provider
Pick between Claude API, ChatGPT API, and Ollama (Local). The panel below the picker changes to show only the settings that apply to the selected provider. Ollama is Pro-only.
Claude API Key
Paste your Anthropic API key into the API Key field. The key is stored in the macOS Keychain. It is never stored in plain text in files, preferences, or logs. The field shows only the last four characters after saving for confirmation.
Click Remove to delete the key from the Keychain. This disables Claude consolidation for all lectures.
ChatGPT API Key
Paste your OpenAI API key into the API Key field. As with the Claude key, it is stored in the macOS Keychain and never written in plain text. Once saved, Harvestry fetches the model list from your account and filters it to the chat-capable models; click refresh beside the picker to pull the list again.
An OpenAI API key is billed separately from a ChatGPT Plus subscription — a Plus subscription on its own does not provide API access.
Notes Language Pro
By default the consolidated notes are written in the transcript's own language. Set this to any supported language to have the model write them in that language instead — the only way to get pairs other than X → English, since Whisper's own translation targets English only. Changing it marks existing notes as stale and offers to re-run consolidation.
Ollama Endpoint
The default Ollama endpoint is http://localhost:11434. Change this field if you run Ollama on a different port or a remote machine. Harvestry appends /api/tags and /api/generate to this base URL.
System Prompt
The system prompt is sent to the LLM before the transcript. It instructs the model on how to structure the output. The default prompt requests structured Markdown with topic headings, bullet points, and a Key Takeaways section.
Edit the prompt directly in the text editor. Changes are saved immediately. Click Reset to Default to restore the original prompt.
The same system prompt is used for every provider. Per-lecture prompt overrides are not currently supported.