LingLu topic minutes

Configure LingLu

In the LingLu API area of setup or advanced settings, select a service and enter an available model, endpoint and authentication. Ollama can use a local service; an example model name does not mean it is installed. Inspect LingLu configuration separately from ordinary translation.

Adjust scenario, temperature, output length and trigger frequency for the meeting. Frequent updates add service calls. Test the topic tree and report with a short recording before using them for a long session.

123

1. 1Choose the service for meeting analysis.
2. 2Enter a model available through the service.
3. 3Choose a scenario appropriate for the meeting.

Record, review and export

1. During recording: expand the live topic tree in the meeting area and inspect topics and linked passages.
2. After recording: inspect the saved session, topics, summary, decisions and actions, and review them against the transcript.
3. Delivery: choose a report theme, preview it and export; inspect the actual output path.

12

1. 1Choose the report theme.
2. 2Use the panel actions to preview or export.

Scenarios and report themes

Choose from generic, personal recording, dialogue, gathering, work meeting, public meeting, business or academic scenarios, or use automatic selection. Choose a scenario explicitly if automatic selection is unsuitable. Personal recordings do not need artificial business KPIs.

Choose a terminal-grade, keynote-stage, editorial, minimal-print or sumi-ink report theme. A theme changes presentation, not meeting facts. Review the preview before exporting.

12

1. 1Check the summary against the transcript.
2. 2Check actions, owners and dates.

Reports and meeting data

The LingLu topic tree, transcript and minutes are different data. Confirm the session and export scope. Cloud analysis sends the required text to the selected service; for local analysis, verify that the service address actually points to your machine.

Local cache and background transcription

With local SenseVoice or FunASR transcription and Linglu enabled, audio is cached in approximately 30-second WAV chunks and transcribed in the background, one chunk at a time. Live subtitles keep their existing recognition path; Linglu uses the chunk transcript for its report. Cloud realtime transcription retains its existing path and does not use this local chunk workflow.

1. In Setup Wizard, open the Linglu section and expand Audio cache location. Leave it empty for the default session location, or choose an absolute folder on a drive with enough space. Save the wizard settings. Changes apply to new recordings.
2. Start recording. The Linglu panel shows waiting, transcribing, transcribed and retry counts. Transcribed does not mean the report has finished: check the generated notes too.
3. Choose Stop. Any nonempty final chunk is processed, even when it is shorter than 30 seconds. Keep the app running until transcription, report processing and saving finish.
4. If transcription needs attention, review the error, resolve it and choose Retry processing. The original cached audio is retained. Report request failures have a separate retry state.

If processing falls too far behind or free space is low, capture stops and admitted audio is retained for processing. Review the stop message before starting another recording. Do not move the active cache folder or power off while work is pending.

After the original recording, raw chunk transcript and report are saved and checked, completed temporary chunk data can be moved into a sibling folder named 可删除. This is archival, not deletion, and does not free disk space. Pending or failed chunks stay in the active cache. Check saved session results before manually removing completed cache. Formal recordings and reports are retained.

Can I close the window before Linglu finishes?

You can close the interface and leave the app working. If Linglu is recording, processing or saving when you close the main window, a task reminder appears. Live meeting minutes, AI minutes and Growth Journal reports are also included in the exit wait.

1. Keep using the app: choose Cancel to stay on the current page.
2. Work in the background: choose Close interface · Keep working. The main window hides, but the app stays in the tray. Recording also continues unless you stop it first.
3. Exit: choose Save and exit. The app stops recording, includes the final tail and waits for Linglu processing and saving. If work remains, exit is deferred; leave it running in the background and try later.
4. Check results: click the tray icon or choose Open from its menu, then review the saved session and report.

Keep the computer on for background processing. Shutdown or forcing the process to end in Task Manager interrupts it. A hidden window does not mean the app has exited. If saving fails, review the message and check the saved recording. If Linglu needs attention appears, the report request has failed three times. Review the service error and choose Retry processing after the service recovers. Hiding keeps pending work in this running app; it does not start a separate process or replace saving before shutdown.

When the notes response is incomplete

This response check requires the Windows r512 or later installer preview.

If the service reaches its output limit, refuses the result, returns tools instead of text or provides no completion marker, this result is not applied as completed notes. The original transcript, previous result and pending batch remain available in this running app. A valid-looking JSON response alone does not prove completion.

1. Review the service message. The notes stage may be paused while transcription has already finished; those are separate stages.
2. Check the saved Linglu service and model in Setup. Select a model that can finish this task, or correct the service so it returns completion metadata.
3. Use Retry failed work in the progress card. Retry is bound to this recording and uses its pending text, rather than asking you to record it again. Saving configuration alone does not replay work.
4. Keep the app and computer running until processing and saving have finished. You can hide the interface to retain pending work. Shutting down and resuming automatically after a restart is not supported.

Known incomplete results pause for your decision instead of making the same request repeatedly. Temporary transport failures retain the existing bounded retry policy. Completion metadata cannot prove that a summary includes every fact; continue reviewing citations and original speech.

Understand background progress for this recording

This progress card requires the Windows r512 or later installer preview. It covers local chunked transcription with Linglu enabled. Cloud realtime recognition keeps its own status.

1. Find the card below the main recording controls.The docked report and independent Linglu window use the same kind of progress view. Transcription and notes may continue after recording stops.
2. Read sealed duration and four separate counts.Waiting, transcribing, transcribed and need-retry counts describe audio chunks. Sealed duration means the chunk cache has passed its save checks; it does not confirm that the full original recording or report has finished saving.
3. Expand “Audio and notes details”.Compare audio that entered the chunk writer with sealed audio. Unsealed audio is not confirmed durable. Stopping seals the final nonempty short tail. Transcription, text cleanup and notes finish separately; they are not added into a completion percentage.
4. Follow the actual reason shown.Backlog, low disk space, disk-check failure and unconfirmed saving have different guidance. The 32-chunk value is a protection threshold for pending audio. Draining admitted audio can exceed it after capture stops; it is not remaining time or total capacity.
5. Retry only failed work for the current recording.The button appears only when the backend confirms a retry is possible and carries the recording task identity. A failed status read marks the previous readings as stale: select “Refresh status” first. If the model worker ended or saving is unconfirmed, review the original recording, fix the configuration and import the audio yourself if needed.

Refreshing, retrying or closing the report does not clear pending or failed cache. A disappearing card is not proof that every file saved successfully: review audio and text in the session browser. The unfinished-task prompt still governs closing the main interface or waiting for saving. Background work needs this computer to remain on.