Repository navigation
Releases
1.1.6
Transcript cleanup
Before each stage summary, the summary's model now tidies that stage's transcript: punctuation, filler words and verbal tics, repetitions and false starts, and words or names that were clearly misheard. Sentences are never merged, split, reordered or removed, and the plugin checks every changed sentence itself, keeping the original where a number, a negation, a question or the person spoken of changed, or where too much was added or cut.
- This is the new default. Choose Settings → Transcript cleanup → Correct misheard names only for the previous behaviour, or Leave as recognized to send nothing.
- The text as recognized is kept in the meeting's
meeting.json. Restore recognized text in the card menu, or the command Restore this meeting's recognized text, puts it back. - The card shows how many sentences were tidied.
- Tidying runs with reasoning switched off where the service allows it, returns only the changed sentences, and replaces the separate names check, so it adds little to a meeting's cost.
Token usage
meeting.json now adds up the tokens each kind of request used (cleanup, summary, overview, minutes), as the service reported them, including cached input and reasoning tokens.
Full Changelog: 1.1.5...1.1.6
1.1.5
1.1.5 asset update: The main.js release asset was rebuilt from commit 58bb471 to generalize and localize settings examples. The version and existing 1.1.5 tag remain unchanged; the tag and its original build provenance refer to the earlier build. Existing installations of 1.1.5 will not receive an automatic update.
Full Changelog: 1.1.4...1.1.5
1.1.4
1.1.2
1.1.1
1.1.0
Meeting Live is no longer tied to one speech service, and the interface is now available in English as well as Chinese.
Setup
- Accounts. Pick a service, paste its API key, done. One key covers every role the vendor supports, for example live transcription and recorded files.
- Language model. Models are listed from the service. Endpoints, temperature, reasoning effort and other details sit under Advanced.
Speech recognition
| Role | Services |
|---|---|
| Live transcription | Volcengine, Alibaba Cloud Bailian, AssemblyAI, Deepgram, or a local sherpa-onnx server |
| Recorded files | Volcengine, Alibaba Cloud Bailian, AssemblyAI, Deepgram, any OpenAI-compatible service (OpenAI, Groq), or a local Whisper server |
- Speaker labels where the service provides them, shown in the transcript and passed to summaries and minutes.
- Catch-up pace. Buffered audio is sent back after a network drop at the pace each service allows, with up to five minutes held for recovery.
Language models
Presets for DeepSeek, OpenAI, Anthropic, Gemini, OpenRouter, Qwen, Kimi, GLM, Volcengine Ark, SiliconFlow, MiniMax, Grok, Mistral, Groq, Ollama, LM Studio, Azure OpenAI, and any OpenAI-compatible endpoint.
Other changes
- Interface language. Follows Obsidian's language. Text written into notes can be set to Chinese or English.
- Long transcripts are condensed first when they exceed the model's input limit.
- Consent is asked per destination; services on this computer are marked as local.
- Fixes for Windows file handling, retry after a corrected key, and error reporting.
Upgrading from 1.0.x
Settings migrate automatically. You don't re-enter keys, and your prompts are kept.
1.0.0
First release.
- Embedded meeting card (
meeting-livecode block) with local recording, Volcengine real-time transcription and periodic AI stage summaries. - Meeting notes from the transcript, summaries and your own notes.
- Generate notes from an existing recording in the background.
Desktop only. Requires a Volcengine speech recognition account and an OpenAI-compatible LLM endpoint. See the README for data and privacy details.