Troubleshooting

Click a problem to see the fix. The in-app Help page has a short version of the most common ones.

Setting up

These messages appear under the form when Connect or Test fails on the Models page. The provider's own words are under Details.

"… rejected this key"

The key is wrong, revoked, or cut short when you copied it. Copy it again from your provider: Anthropic keys start with sk-ant-, OpenAI keys with sk-, Google AI Studio keys with AIza.

"… doesn't recognise this model"

The provider doesn't offer that model name to your account. Pick one from the menu, or check the exact spelling if you typed it yourself.

"Can't reach Ollama / LM Studio … is it running?"

Start Ollama or LM Studio (in LM Studio, start its local server), then press Retry. If it runs on another machine or port, change the server address. If it says no models are installed yet, run ollama pull <model>, or download one in LM Studio.

"… refused the request. Check the address, and that the model is installed." means the server answered but wouldn't run that model. Check the model name is one the server lists.

"Couldn't reach … Check your internet connection or proxy."

Polygluttony couldn't get through to the provider at all. Check your connection, VPN or proxy. For an Other endpoint, check the URL. It usually ends in /v1.

"… says this account is out of credit"

Your key works, but the account behind it has no credit left (or billing isn't set up). Add credit in your provider account, then press Try again or Test. The same message can stop a run; see "Stopped: …" below.

"… is rate-limiting this key" or "…'s servers had a problem"

The provider is busy or limiting how fast you can send. Wait a minute and try again. If you're sure the key and model are right, Save anyway saves the model without the test.

"… sent back an error. Details has what it said."

The provider refused for a reason Polygluttony doesn't recognise. Open Details to read its own message. It often names the problem: a setting on your account, a region, a model you need to request access to.

"… uses …, which needs an API key"

A step's model has no key saved, for example "Tidy up uses Claude Opus 5.5, which needs an API key. Add it on the Models page." On Models, the step shows Needs an API key · Add it. Click Add it, paste the key and press Connect.

"Look up names online" is greyed out

It needs a model that can search the web. On Models, turn on Look up names online for your model, or give it a Claude, Gemini or OpenAI model with Use a different model. Local and Other models can't search. See Connecting a model → Look up names online.

Exporting and importing a setup

Nothing changes when an import fails, except in the "prompts couldn't be" case below. See Moving to another computer.

"This isn't a Polygluttony export."

The file wasn't made by Export setup…, or it was damaged on the way. Pick the .zip exactly as Polygluttony saved it, without unzipping and re-zipping it. If that still fails, export again on the other computer.

"This export was made by a newer Polygluttony."

The file comes from a newer version than the one on this computer. Install the latest release, then import it again.

"The edited "…" prompt in this export can't be used"

One of the edited prompts in the file won't work in this version, for example because it's missing a placeholder this version needs. On the other computer, fix that prompt or press Restore default on the Prompts page, then export again.

"Couldn't save a backup of your current setup first"

Before replacing anything, Polygluttony saves your current setup to its data folder so you can undo the import. That failed, so nothing was replaced. The message says why, often a full disk or a folder you can't write to. Fix that, then import again.

"Your connections were imported, but the prompts couldn't be"

Your models, keys and settings are in, but the edited prompts couldn't be written to Polygluttony's data folder. The message says why, often a full disk or a folder you can't write to. Fix that, then import the same file again.

"Your edited "…" prompt can't be exported"

One of your edited prompts is broken, for example empty or missing a required placeholder. Click Open Prompts in the message, fix the prompt or press Restore default, then export again.

I want my old setup back after an import

Press Undo in the Setup replaced message. If it's gone, use Import setup… and pick the newest setup-backup-<number>.zip in Polygluttony's data folder. The last 3 are kept.

Folders

"No subtitle files here"

Polygluttony reads .ass files directly in the folder you open, not in subfolders. .srt and other formats aren't read: convert them to .ass first, in Aegisub or Subtitle Edit for example. Files that already look like translations (.eng.ass, .warning.) are skipped too.

The source language is wrong

On the Episodes page, click Edit next to the settings line and change Source language. Languages written in the Latin alphabet can't be detected from the text, so those folders start with the last pair you used. See Languages.

"A glossary can only be built for Chinese subtitles for now."

The Glossary page says this when the folder's source language isn't Chinese. If it should be, change Source language in the folder settings on the Episodes page.

The glossary "has a mistake on line …"

glossary.json was changed outside Polygluttony and is no longer valid. Polygluttony won't touch it, and translations don't use it until it's fixed ("The glossary can't be read, so this run won't use it."). Click Fix it on the glossary row, then Open in your text editor and fix that line; the page reloads by itself. Or choose Restore an earlier version… to go back to a copy from before.

Translating

The gold button is greyed out

The reason is shown next to it:

• "No model is ready to translate. Set one up in Models." No model is set up, or the Translate or Verify step's model still needs its API key.
• "Tick the episodes to translate." Nothing is ticked. Each time you open a folder, only the episodes without a translation are ticked. To translate others again, tick them (Select: All), or use ··· → Translate again on a row.
• "Source and target languages are the same." Change one in the folder settings.
• "Another folder is translating." or "A glossary job is running." One job runs at a time. Wait for it, or stop it.

"Couldn't start: …"

The run was refused before anything was sent. The sentence after it says why:

• "Connect a model first, on the Models page." or "… uses …, which needs an API key." Press Open Models and fix it.
• "Another operation is already running." A glossary job or another run is going. Wait for it to finish.
• "Custom "…" prompt exists but can't be read" A prompt you changed can't be read anymore (damaged file, wrong permissions). Open Prompts, select it, and fix it or press Restore default. See Custom prompts.

Then press Try again.

"Stopped: … rejected the API key used for Translate."

The run hit an error that no retry would fix, so it stopped instead of failing every remaining batch. The sentence names the provider and the step:

Message	Fix
"… rejected the API key used for …"	Copy the key again and replace it on Models.
"… says the account used for … is out of credit."	Add credit in your provider account.
"… doesn't recognise the model "…" used for …"	Pick another model for that step on Models.
"… refused the request for … Check the address, and that the model is installed."	A local server: check its address and that the model is installed.

Episodes that finished are kept; the rest show Not finished. Fix the problem (Open Models takes you there), then press Try again. It carries on from the lines already done.

"Stopped. 3 episodes weren't finished."

You pressed Stop. Finished batches are kept, and Resume 3 episodes carries on from the lines already done, as long as the episode file, the languages and the tone haven't changed since. Otherwise those episodes start again from the top.

"The glossary couldn't be built: …" (or "updated")

With Build a glossary first (or Add … to the glossary first) ticked, a run reads the episodes for the glossary before it translates. That step failed, so nothing was translated. The usual reasons:

• "The model refused the request. Check it on the Models page." The Find names model's key was rejected, or the provider doesn't know the model. The terms found before that are saved. Open Models takes you there.
• "The model didn't answer. Check it on the Models page." Nothing came back: the provider is down or unreachable, or the local server isn't running.
• "Find names uses …, which needs an API key." (or Tidy up) Add the key on Models.

Fix it, then press Try again. To translate without the glossary step for now, untick it. Details are in Logs.

"Stopped while building the glossary. … terms kept; nothing was translated."

You pressed Stop while the glossary was being built. The terms found so far are kept. Episodes the glossary didn't finish reading are offered again next time, so press the gold button when you're ready. ("Stopped while updating the glossary" works the same way.)

"… but part of the episodes couldn't be read"

Some of the glossary's requests failed. The glossary went ahead with what came back, and the translation ran. The episodes it couldn't read are offered again the next time you translate with the tick on ("Update glossary, then translate …"). Logs says which requests failed and why.

An episode says "Needs a look"

It was saved as ….warning.eng.ass with issues left after its retries. Fold out the row to see them, or press Review to step through the flagged lines in the episode window. When you're happy, Mark episode done renames it to the clean name. See Reading the results.

An episode "Failed"

The row shows the reason, and Logs has the details. The usual ones:

• "… kept asking us to slow down, and no line got translated." See the next item.
• "Couldn't reach …, so no line got translated." Check your connection, or that the local server is running.
• "… kept failing on its side, so no line got translated." The provider is having trouble. Try again later.
• "No reply could be used, so no line got translated." The model's answers didn't follow the line format. Try fewer Lines per request, or a stronger model.
• "This file has no dialogue lines." There's nothing to translate in it.
• "Couldn't write …" The folder is read-only or the disk is full.

Press Try again in the banner, or ··· → Translate again on the row.

"Waiting — … asked us to slow down", lots of retries, or timeouts

The provider is rate-limiting you, or answers are taking too long. Now and then is normal: Polygluttony waits and backs off by itself. If it keeps happening, open the step's Tuning on Models:

• lower Requests at once;
• for timeouts, raise Response timeout or lower Lines per request.

Local models in particular need patience: the first request also loads the model.

"… lines weren't translated" / the translated file is missing lines

Those lines couldn't be translated, so they were left out and flagged UNTRANSLATED, and the episode is a .warning. one. In the episode window, press Retranslate line, or Write it myself. If many lines are missing, translate the episode again with fewer Lines per request. See Reading the results.

Review won't change an episode

The episode window says why. The common ones:

• "… was changed outside Polygluttony since it was written." You edited the translation in another editor, and Polygluttony won't write over that. Keep editing there, or translate the episode again.
• "The original file changed since it was translated." or "The lines of … aren't stored any more." Translate the episode again to review it.
• "This episode is being translated…" Wait for it to finish.
• "This episode isn't finished." Resume it from the Episodes page first.

The translated file has no embedded fonts

The [Fonts] section isn't copied into translations. Keep the fonts installed, or attach them when you mux. See Folders & files.

"Old ….warning file is still here"

A new translation replaces the other version: a clean ep01.eng.ass replaces ep01.warning.eng.ass, and the other way round. When the old one couldn't be removed (open in another program, or no permission), the row says so. Delete it yourself.

No desktop notification when a run ends

Polygluttony only notifies you when it isn't the active window. It asks for permission once; if you said no, turn notifications on for Polygluttony in your system's settings.

Installing and starting

macOS says Polygluttony can't be opened

Expected on the first launch. Use System Settings → Privacy & Security → Open Anyway. Steps: Platform notes → macOS.

Linux: the AppImage doesn't start

Make it executable (chmod +x). If it still won't start, install FUSE 2 (libfuse2 or fuse2). Run it from a terminal to see the error. See Platform notes → Linux.

Still stuck?

Open an issue and include:

• the version, shown at the bottom right of the window (3.0.0, for example);
• your OS;
• the provider and model of the step that went wrong;
• the relevant lines from Logs, at the bottom of the Episodes page (select and copy them). Leave out anything you'd rather keep private.