Yogurt's AI model management, resource consumption, and user configurability for transcription-only workflows.
Raw Developer Origin & Technical Request
GitHub Issue
Nov 13, 2026
## Summary
Add a way to run Yogurt without the notes model at all: no LLM download, no LLM jobs, no retries. Recording, live transcription, speaker names, search, and the MCP server would all keep working; Summary and Ask would be visibly disabled rather than erroring.
## Use case
Some users want Yogurt strictly as a private, bot-free meeting transcriber and pipe the transcripts into their own AI tools (Claude, Cursor, etc. — especially now that the MCP server makes this easy). For that workflow the local notes model is pure cost:
- On a 25 GB M5 Mac, Yogurt selected `gemma-4-E4B` and then the **14.4 GB `gemma-4-26B`** — a large download plus a model that held ~14 GB of memory when loaded.
- There is currently no setting to opt out: the first notes model retries its download at 1 min / 5 min / then every 10 min effectively forever, and the model runs jobs (enhance, names, index) after every meeting.
- The only workaround is out-of-band (e.g., replacing the `.gguf` with a stub while keeping the `.ok` marker, so `isVerified` passes but `engine.load` fails), which is fragile and leaves meetings with failed notes jobs.
## Proposal
A "Notes & AI" setting with an explicit **Off / Transcription only** choice:
1. Skip the notes-model download entirely during setup and afterwards (no retry loop).
2. Do not queue `enhance`, `names`, `readahead`, or `folder` jobs. (`index` could keep running — the embedding model is small and search is still useful; or gate it behind the same tog...
Developer Debate & Comments
No active discussions extracted for this entry yet.
Adjacent Repository Pain Points
Other highly discussed features and pain points extracted from FrigadeHQ/yogurt.
Frequently Asked Questions
Market intelligence mapped to Yogurt's AI model management, resource consumption, and user configurability for transcription-only workflows..
What problem does Yogurt's AI model management, resource consumption, and user configurability for transcription-only workflows. solve?
How is the developer community reacting to Yogurt's AI model management, resource consumption, and user configurability for transcription-only workflows.?
What architecture is tied to Yogurt's AI model management, resource consumption, and user configurability for transcription-only workflows.?
Engagement Signals
Cross-Market Term Frequency
Counts stored text matches for foundational terms like MCP server and index in analysed public discussions. These counts do not measure adoption.
Stored text matches, not adoption or market-share measurements.
- MCP server: 59 records
- index: 47 records
- names: 25 records
- enhance: 6 records
- gemma-4-E4B: 3 records
- gemma-4-26B: 3 records
- retry loop: 3 records
- Transcription-only mode: 1 records
- opt out of the notes model: 1 records
- LLM download: 1 records
- LLM jobs: 1 records
- 14.4 GB: 1 records
- memory when loaded: 1 records
- readahead: 1 records
- folder jobs: 1 records
- .gguf: 1 records
- .ok marker: 1 records
- isVerified: 1 records
- engine.load: 1 records
SaaS Metrics