This repository was archived by the owner on Nov 8, 2026. It is now read-only.
Repository navigation
F5 dictation and Smart Reply on the shared LFM3-Audio assistant (#649) - #654
Merged
Merged
Conversation
- stream-dictate: new `lfm3` engine (dictation_lfm.py) that sends pause-cut phrases to mseries-assistd /v1/audio/transcriptions and types each as it returns. - assistd: MSERIES_ASSIST_RESERVE_GIB makes the 6 GiB desktop reserve configurable; the lfm3 F5 engine uses the same reserve. - mseries-ai: Smart Reply drafts fall back to the assistant (LFM3) when the default Qwen endpoint has no model, so Ctrl+Alt+R works without the Qwen profile loaded. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part of #649.
What changes
lfm3engine forstream-dictate.py(dictation_lfm.py). It cuts the mic stream into phrases at pauses, sends each tomseries-assistd/v1/audio/transcriptions, and types each phrase as it returns. F5, F6 and Smart Reply share one resident model.MSERIES_ASSIST_RESERVE_GIBmakes assistd's fixed 6 GiB desktop reserve configurable (default unchanged). Thelfm3F5 engine uses the same value.Checked on the M1 Pro
/v1/complete: sensible continuation in 1.6 s.mseries-reply draft fixtures/lunch.json: a draft in 3.3 s via the fallback.cargo test: mseries-ai 24 passed, mseries-reply 78 passed.Not checked: a live F5 press with a real voice.
🤖 Generated with Claude Code