Transcription error on self-hosted Cap

[bapspatil]

Description

I've self-hosted Cap and connected my self-hosted Minio as the S3 bucket to store the video files. I added my DEEPGRAM_API_KEY, expecting transcription to work after that, but it's not and I'm currently getting these errors in the server logs:

2025-11-05T17:55:12.743690544Z Error transcribing video: Error: Failed to transcribe audio
2025-11-05T17:55:12.743732234Z     at ag (.next/server/chunks/ssr/apps_web_e25a4340._.js:1:32882)
2025-11-05T17:56:24.598063147Z [ShareVideoPage] Starting transcription for video: z0wh037c02gg9be
2025-11-05T17:56:24.778082857Z [transcribeAudio] Starting transcription for URL: https://redacted.domain.com/redacted-coolify-bucket/redacted-id/redacted-subid/result.mp4?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=redacted%2F20251105%2Feu-central%2Fs3%2Faws4_request&X-Amz-Date=20251105T175624Z&X-Amz-Expires=3600&X-Amz-Signature=redacted&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject
2025-11-05T17:56:25.980731757Z [transcribeAudio] Deepgram transcription error: Error [DeepgramApiError]: {"err_code":"Length Required","err_msg":"Fetching failed to get valid content-length.","request_id":"redacted-2fbf-483d-ba53-redacted"}
2025-11-05T17:56:25.980776927Z     at <unknown> (.next/server/chunks/ssr/apps_web_e25a4340._.js:1:9636) {
2025-11-05T17:56:25.980786547Z   __dgError: true,
2025-11-05T17:56:25.980816747Z   status: 411
2025-11-05T17:56:25.980822467Z }
2025-11-05T17:56:25.981185067Z Error transcribing video: Error: Failed to transcribe audio
2025-11-05T17:56:25.981192267Z     at ag (.next/server/chunks/ssr/apps_web_e25a4340._.js:1:32882)
2025-11-05T17:56:25.981197257Z     at async L (.next/server/chunks/ssr/apps_web_app_s_[videoId]_eae78d05._.js:1:10664)

Reproduction

1. Add Deepgram API key via DEEPGRAM_API_KEY.
2. Record or upload a video to Cap.
3. Click on the Transcription tab in the recording page.

Additional Context

• Cap version: 0.82
• Operating system, version: macOS Tahoe 26.1
• Device (optional):

Cap's debug logs: (not applicable as the error is happening on the server for transcription, not the macOS client)

[linear]

CAP-540

[xenova]

I wonder, is there any interest in using a local model for speech transcription? 👀 (as opposed to via an API)