Skip to content

[TROUBLESHOOT]

Youssef Larbi edited this page Jul 30, 2026 · 1 revision

Troubleshooting

Transcript extraction fails with 429 or connection timeout

YouTube is rate-limiting your IP address. Set up a Webshare proxy account and add WEBSHARE_USERNAME and WEBSHARE_PASSWORD to your .env file. Free tier at https://webshare.io includes 10 proxies and 1GB per month.

AnkiConnect not reachable

Check that Anki is open and AnkiConnect is installed (code: 2055492159 in Tools -> Add-ons -> Get Add-ons). After installing AnkiConnect, Anki must be fully restarted, not just closed and reopened.

If you are on WSL, see the WSL Setup page.

Run this to test connectivity:

curl http://localhost:8765

If you get "connection refused", AnkiConnect is not listening.

pipeline.db errors about missing columns

Your database was created before a schema update. Delete pipeline.db and run again. The schema is recreated automatically. You will not lose any Anki cards, only the processing history.

Translation model downloads every time

The argostranslate model is being installed to a path that changes between sessions. This is a known issue in WSL environments where the home directory path is inconsistent. Add this to your .env:

ARGOS_PACKAGES_DIR=/mnt/c/Users/YourUsername/.argos-translate

Replace YourUsername with your actual Windows username. This pins the model directory to a stable Windows path.

Cards have no example sentences

This usually means dictionaryapi.dev returned no examples for those words in the target language. Coverage varies by language. For English vocabulary, Merriam-Webster provides better example coverage. For other languages, many words will only have the transcript example sentence.

Same word appears multiple times as different cards

This happens if you run the pipeline on the same video twice without pipeline.db tracking the first run, or if the video ID in the GUID changed. Delete duplicate notes in Anki via Browse -> select duplicates -> delete.

Processing is very slow

Each word requires at least one API call with a 0.5 second delay between calls. A 100-word vocabulary extraction takes 2-12 minutes depending on cache state. Words already in the SQLite cache are instant. Second runs of similar videos are significantly faster.

Translation mode adds additional latency per word for the translation step.

Clone this wiki locally