yanlan-transcribe

name	yanlan-transcribe
description	Transcribe a local audio recording into plain text, Markdown, or structured JSON with the Yanlan CLI and MiMo ASR. Use when an Agent needs to convert MP3, WAV, M4A, WebM, OGG, Opus, AAC, FLAC, or MP4 audio files into text, including voice memos, interviews, meetings, lectures, and podcasts. Do not use for live microphone recording, speaker diarization, summarization, or transcript correction.

Yanlan Transcribe

Use the deterministic Yanlan CLI for one local audio file at a time. The command uploads the audio to the configured MiMo-compatible ASR API; it does not send the file to a Yanlan server.

Workflow

1. Resolve the input audio path and verify that the user requested transcription of that file.
2. Check for Node.js 20.19 or newer with node --version.
3. Check whether MIMO_API_KEY is available without printing its value. If it is absent, ask the user to set it locally and stop.
4. Choose an output beside the audio unless the user supplied another path. Use .txt by default, .md for a readable document, or .json when downstream automation needs segments and metadata. If that path already exists, choose a numbered filename instead of replacing it. Use --force only when the user explicitly requested replacement of that exact output file.
5. Check the input size. The default mimo-chat data-URL protocol rejects files over 40 MiB before upload. For a larger file, split or compress it first, or use --protocol openai-transcriptions only when the configured provider explicitly supports that endpoint.
6. Run the pinned release:

npx --yes github:ranxi2001/yanlan-ai#v0.6.2 transcribe "/absolute/path/recording.mp3" --output "/absolute/path/recording.txt"

7. Confirm the output file exists and is non-empty. Report its path, selected language, and model. Do not paste a sensitive transcript into the response unless the user asks to see it.

Options

Use automatic language detection unless the user requests a bias:

yanlan transcribe recording.m4a -o recording.txt --language auto
yanlan transcribe interview.wav -o interview.md --language zh --format markdown
yanlan transcribe lecture.mp3 -o lecture.json --language en --format json

Set MIMO_BASE_URL for a compatible gateway. Remote gateways must use HTTPS; plain HTTP is accepted only for loopback development endpoints (localhost, 127.0.0.0/8, or [::1]). Use MIMO_ASR_MODEL, MIMO_ASR_PROTOCOL, and MIMO_ASR_PATH only when the provider requires non-default values. Prefer environment variables over --api-key so credentials do not enter shell history or process listings.

Failure handling

• On 401 or 403, ask the user to verify the local Key and account access without exposing the Key.
• On timeout or connection failure, preserve the audio and any existing output, then report the endpoint host and error without credentials.
• Never use the input audio path as the output path. The CLI rejects identical paths and hard links, and it rejects existing outputs unless the user explicitly approved --force.
• On an unsupported extension, do not rename the file to bypass detection; convert it with a trusted audio tool or ask for a supported format.
• On the 40 MiB data-URL limit, preserve the source and do not bypass the guard. Split/compress the audio or use a provider-supported multipart transcription endpoint.
• On an empty transcript, report the failure instead of inventing content.
• Do not summarize, correct terminology, or infer speakers unless the user separately requests those tasks after transcription.