Global settings live under Oral Archivist > Settings and are available to global administrators only. Provider keys are never exposed to site admins.
Providers
| Setting | Default | Description |
|---|---|---|
| Transcription provider | Mock | mock produces a fake transcript at no cost; openai uses Whisper with word timestamps. |
| LLM provider | Mock | Used for the cleanup, segmentation and summary passes. |
| OpenAI API key | Stored encrypted with the key from local.config.php. Leave blank to keep the current key. |
|
| LLM model | Fetched from the provider with Refresh model list. | |
| Default language | auto-detect | ISO 639-1 code such as en, passed to the transcription provider. |
Review and queueing
| Setting | Default | Description |
|---|---|---|
| Low-confidence threshold | Words below this confidence (0 to 1) are flagged for review. | |
| Auto-queue new audio/video media | Off | Process eligible media automatically on ingest, subject to the caps below. |
| Daily auto-queue cap | Maximum media auto-queued per day. | |
| Max attempts per job | A failed stage is retried until this many attempts are used. |
Costs
| Setting | Default | Description |
|---|---|---|
| Monthly budget (USD) | 0 (unlimited) | Queueing is refused and running jobs stop once the month's spend would exceed this. |
| Transcription rate (USD per minute) | 0.006 | Used for estimates and actuals. Matches OpenAI whisper-1. |
| LLM input rate (USD per 1M tokens) | 0.15 | |
| LLM output rate (USD per 1M tokens) | 0.60 |
See Costs and Budgets for how these are applied.
Audio processing
| Setting | Default | Description |
|---|---|---|
| Audio chunk length (seconds) | 600 | Long recordings are split into chunks of this length when they exceed the provider's upload limit. |
| Audio chunk overlap (seconds) | 5 | Overlap between chunks so words at the boundary are not lost. |
| LLM cleanup chunk size (characters) | 12000 | How much transcript is sent to the model per cleanup request. |
| ffmpeg path, ffprobe path | auto-detect | Detected in /opt/homebrew/bin, /usr/local/bin and /usr/bin. Set explicitly if the binaries live elsewhere. |
Public display
| Setting | Default | Description |
|---|---|---|
| Public transcript label | "AI-assisted transcript, reviewed by staff" | Shown near the public transcript. |
| Player style | Basic | One of five player bar styles. Sites can override it. |
| Page layout | Classic | How player, chapters and transcript are arranged. Sites can override it. |
| Placement on the page | After | Whether the automatic rendering appears before or after the page content. Sites can override it. |
See Layouts and Styles.
AI summary
| Setting | Default | Description |
|---|---|---|
| Summary metadata property | none | When set, the pipeline adds a summary pass and publishing writes the summary to this property on the parent item. Existing values in that property are replaced. Leave empty to disable summaries; the pass is then skipped at no cost. |
| Summary prompt | built in | System prompt for the summary pass. |
Prompts
The Cleanup prompt and Segmentation prompt are the system prompts for the two LLM passes. The defaults are tuned for the JSON shapes the module expects; if you customise the cleanup prompt, it must still instruct the model to return the documented JSON shape.