Troubleshooting

Find the error by the text you actually saw. Each entry says what it means and what to type.

scribe and ascli are shorter aliases for anyscribe — every fix below works typed as any of the three.

Start here

Whatever the problem, run this first:

anyscribe doctor

It checks your dependencies, config, install, and whether an update is available. If you're asking for help or filing a bug, paste its output — it contains everything needed to diagnose the issue and no secrets.

If the question is about a specific past run — did it actually transcribe, what happened to the one that failed — check the activity log instead:

anyscribe logs

That shows recent transcriptions plus any recovery artifacts: audio that was downloaded before a run failed, kept so you don't have to download it again.

Install and startup

command not found: anyscribe (or 'anyscribe' is not recognized)

anyscribe is installed, but the folder pip put it in isn't on your PATH.

The instant workaround — this always works, no setup:

python -m anyscribe onboard              # same as: anyscribe onboard
python -m anyscribe transcribe "<url>"   # same as: anyscribe transcribe "<url>"

To fix the anyscribe shortcut properly, first confirm it's installed:

python3 -m pip show anyscribe

Then add your Python scripts directory to PATH:

```powershell # Find where pip installed the scripts: python -c "import sysconfig; print(sysconfig.get_path('scripts'))" # Add that path permanently (replace with the output above):

```

Then restart your terminal.

"I upgraded and my API keys are gone"

anyscribe used to keep config and keys in ~/.anyscribecli/; it now uses ~/.anyscribe/. Normally the move happens by itself on your first transcription — but if the first thing you ran after upgrading created an empty new folder (anyscribe ui or anyscribe config, say), the old keys can be stranded in the legacy folder.

Fix — run the one-time migration. It never overwrites anything already in the new folder, and it reports how many keys it moved, never the keys themselves:

anyscribe migrate --dry-run    # preview — writes nothing at all
anyscribe migrate              # do it

Safe to run twice: a second run just reports there's nothing left to do.

"Permission denied"

Usually the app folder is owned by root, from an install run with sudo.

ls -la ~/.anyscribe/                      # check the owner
sudo chown -R $(whoami) ~/.anyscribe/     # hand it back to yourself

ModuleNotFoundError: No module named 'mcp.server.fastmcp'

Only affects the MCP server (anyscribe-mcp), not the regular commands.

The MCP library released a version 2 that renamed the piece anyscribe imports. anyscribe releases before 0.16.3 didn't ask for a specific version, so a fresh install pulled the new one and the two stopped fitting together.

pip install -U "anyscribe[mcp]"           # 0.16.3+ pins the version range

Then restart whichever app hosts the server (Claude Desktop, Cursor) so it launches the fixed one. See Agents and MCP for the full setup.

Input and downloads

no matches found when you paste a URL

Your shell — zsh especially — is treating the ? in the URL as a wildcard. This happens before anyscribe ever runs.

Fix: wrap the URL in double quotes.

anyscribe transcribe "https://www.youtube.com/watch?v=VIDEO_ID"

Or run anyscribe transcribe with no URL and paste it at the prompt, where no quoting is needed.

yt-dlp download failed: Video unavailable

The video is private, deleted, age-restricted, or blocked in your region.

anyscribe auto-updates yt-dlp when it's more than 60 days old — YouTube changes formats often and old versions break — so this is rarely a stale-tool problem. Still:

  1. Try a different video, to confirm anyscribe itself is working.
  2. Update yt-dlp by hand: pip install --upgrade yt-dlp
  3. If it's age-restricted, yt-dlp needs browser cookies (advanced).

Instagram: rate-limit reached or login required

The reel is behind a login wall. Point anyscribe at a browser you're already logged into — it borrows the cookies, no password involved:

anyscribe config set instagram.browser firefox

Supported: firefox, chrome, safari, brave, edge, chromium, vivaldi, opera. Firefox is the most reliable on macOS; Chrome's cookie encryption makes extraction flakier.

Already configured a browser and still seeing this?

  1. Open Instagram in that browser and confirm you're logged in.
  2. Open the reel URL in that same browser and confirm you can watch it.
  3. If it plays in the browser but not through anyscribe, the cookie store may be locked by the running browser — quit the browser and retry.

Instagram: private account

The poster's account is private. Cookies from a browser logged into an account that follows them will work; cookies from any other account won't.

Instagram: video unavailable / post not found

The reel was deleted, made private, or is region-locked. Nothing to fix on your side.

Providers and API keys

OPENAI_API_KEY not set (or any other missing-key error)

The key that provider needs isn't in ~/.anyscribe/.env.

anyscribe config set openai_api_key YOUR_KEY   # set it directly
anyscribe onboard --force                       # or re-run the wizard

WARNING: quality 'X' wants Y but no Y_API_KEY is set

Not an error — anyscribe went ahead using your configured provider instead. Your quality tier prefers a provider whose key you don't have.

Either add the missing key:

anyscribe config set deepgram_api_key YOUR_KEY

Or pick a tier you have a key for. anyscribe config shows which keys are present in its Key column.

Provider error or general API failures

anyscribe providers test                          # is the key valid?
anyscribe "url" --provider openai                 # does another provider work?

If the test fails, the key is wrong, expired, or the account is out of credit. If the test passes but transcription fails, try another provider to isolate whether it's the service or the file.

"Next run" isn't the provider I set

quality outranks provider. If quality is a tier (accuracy/balanced/cost/free), that tier picks the provider and your provider line is ignored.

anyscribe config    # look at the (…) after the model — `quality: balanced` means the tier won

Fix: set the provider through the CLI, which writes quality: custom in the same save — that's what makes the choice stick:

anyscribe config set provider elevenlabs

Hand-editing config.yaml is the usual cause: writing provider: without also writing quality: custom leaves the tier in charge.

Unknown model 'whisper-2' for openai.

That model isn't in the provider's list. anyscribe rejects it before downloading or spending anything, and the error names the valid options.

anyscribe config --json          # each provider's current model and alternatives
anyscribe providers list --json

Two cases that look like bugs but aren't:

OpenRouter: model not found / 404

The slug you pinned isn't available on OpenRouter. OpenRouter is the one provider anyscribe doesn't validate models for — it forwards any name unchanged — so a typo or a retired slug only fails once the request reaches the API. Old pins to models OpenRouter has since removed are the most common cause.

Find the pin:

anyscribe config --json     # resolved.model, and providers[].default_model
anyscribe config show       # provider_models.openrouter

Set a current slug:

anyscribe config set provider_models.openrouter openai/gpt-audio-mini

Any audio-capable OpenRouter slug works — see openrouter.ai/models for what's live today.

A leftover OPENROUTER_MODEL= line in ~/.anyscribe/.env is not the cause — that variable is no longer read at all. Delete the line and use provider_models.openrouter instead.

Transcription results

"Already transcribed" — nothing got re-transcribed

Not an error. anyscribe found an existing transcript whose source: frontmatter matches this URL or file, so it handed you that file back instead of re-downloading and re-paying for it. You'll see Already transcribed: <path> — use --force to re-transcribe. — in JSON that's "cached": true, and in a batch the row is marked CACHED.

If you genuinely want a fresh run (you switched providers, or the source was re-uploaded):

anyscribe "url" --force

If you want the old one gone entirely: anyscribe rm <path-or-slug> first, then transcribe again.

Transcription came out in the wrong language

Auto-detection guessed wrong.

anyscribe "url" --language en        # force it for one run (or es, fr, hi, …)
anyscribe config set language hi     # or make it the default

Timestamped or diarized output came out as plain paragraphs

The model that ran doesn't return segment timestamps, so there's nothing to build [mm:ss] markers from. On OpenAI that's gpt-transcribe (the default), gpt-4o-transcribe, and gpt-4o-mini-transcribe; sargam and openrouter never return them at all.

On OpenAI this shouldn't happen on its own — anyscribe swaps in whisper-1 when your output format is timestamped or diarized, and prints switched to whisper-1 — gpt-transcribe can't produce timestamps. So if you got paragraphs, one of these is true:

  1. You passed -m. An explicit per-run model always wins, which suppresses the swap. Re-run without -m, or with -m whisper-1.
  2. The provider isn't OpenAI. Sarvam and OpenRouter have no timestamped model to fall back to. Switch: anyscribe "url" -p deepgram --force
  3. Your output_format is clean. There were never going to be timestamps: anyscribe config set output_format timestamped

Check what actually ran:

anyscribe config --json     # resolved.provider, resolved.model, resolved.notes

To force Whisper on every OpenAI run: anyscribe config set provider_models.openai whisper-1. Deepgram and ElevenLabs give word-level timestamps if you want finer granularity than Whisper's segments.

A long video is taking forever

Anything over ~30 minutes is split into chunks and transcribed piece by piece, then merged. That's normal, not a hang.

Rough guide: cloud providers run at roughly real-time speed. Local transcription on CPU is 2–5× real-time.

Audio left over after a failed run

When a transcription fails after the audio downloaded, anyscribe keeps that audio rather than throwing it away, so a retry doesn't re-download it.

anyscribe logs

Recovery artifacts appear in their own section, separate from the activity log. Re-run anyscribe "url" to retry (add --force if it partly wrote a transcript), or just delete the file if you've moved on — nothing else references it.

Local (offline) transcription

faster-whisper is not installed

The local provider is opt-in and needs a one-time install:

anyscribe local setup --model base

That installs faster-whisper into the same Python environment as anyscribe and downloads the model. The first download needs internet (~150 MB for base); everything after that is fully offline.

--model is deliberately required — anyscribe will never pick a model size for you. base is the recommended starting point.

Checking what's actually installed

anyscribe local status

Reports faster-whisper's version, whether ffmpeg is present, which models are cached, disk usage, and how anyscribe was installed. It always exits 0, so it's safe to run before setup.

Model weights look corrupted

Delete and re-download in one step:

anyscribe model reinstall base --yes

Web UI and menu-bar tray

The "Setup needed to start transcribing" banner won't go away

The banner appears when either of two things is missing, and it disappears on its own — it re-checks every 30 seconds — once both are satisfied:

  1. ffmpeg (and ffprobe) aren't installed. The banner shows the exact install command for your platform, with a copy button.
  2. No transcription provider is set up at all. Add an API key under Settings → Providers.

Using local transcription only? That counts. Once faster-whisper is installed (anyscribe local setup --model base, or the Set up local transcription button on the Local provider card in Settings), the provider requirement is satisfied without any API key.

You can also dismiss the banner with the ✕ for the current session.

The Remove key button is missing for a provider

That provider's key is coming from your shell environment — an export OPENAI_API_KEY=… in your shell profile, say — not from ~/.anyscribe/.env. anyscribe can't edit your shell config, so it hides a button that wouldn't work. Unset it where you set it.

The tray companion needs extra packages / the tray won't start

The tray needs the optional [tray] extra (pystray, Pillow, and pyobjc on macOS). The one-line installers include it by default — if you see this message, you almost certainly installed with a bare pip install anyscribe:

pip install -U "anyscribe[tray]"
anyscribe tray

The tray is running but there's no icon in the menu bar

If the server answers at http://127.0.0.1:8457 but you can't find the waveform icon, check whether a menu-bar manager is running — Hidden Bar, Bartender, Ice, Dozer. macOS inserts new status items at the left of the strip, which these apps hide by default. Expand the manager (its chevron) and drag the anyscribe waveform to the always-visible side.

Very old versions had a bug where the icon never appeared at all — fixed in 0.13.2. Run anyscribe update if you're on anything older.

An anyscribe tray is already running.

A tray is already active and its pidfile (~/.anyscribe/tray.pid) still points at a live process. This is intentional — anyscribe refuses to start a second tray rather than collide with the first.

Fix: use the existing one (check your menu bar), or quit it first (menu → Quit, or Ctrl+C in its terminal) and relaunch.

port already in use

anyscribe ui handles this itself. If the port is taken it looks at the next ten, starts on the first free one, and tells you which:

Port 8457 busy — using 8458.

Nothing to do. You only see an error if all eleven are busy, and only then do you need to name a port yourself:

anyscribe ui --port 9000

anyscribe tray behaves differently, on purpose. If an anyscribe ui server is already listening on its port, the tray attaches to that server instead of starting a second one — that's expected, not a bug. The tray doesn't roll forward to another port (it would then be pointing at a server nothing else knows about), so if an unrelated process holds the port, give the tray a different one:

anyscribe tray --port 9000

Removing the tray completely

  1. Quit the tray if it's running (menu → Quit).
  2. Remove the login auto-start (macOS): anyscribe uninstall-service --yes
  3. Optionally drop the extra packages: pip uninstall pystray Pillow

uninstall-service only removes the login item — it doesn't stop a running tray or uninstall anyscribe.

Last resort

Config or workspace looks corrupted

anyscribe doctor     # find out what's actually wrong first

The nuclear option resets your config and API keys. Your transcripts are safe — they live separately in ~/anyscribe/:

rm -rf ~/.anyscribe
anyscribe onboard

Check anyscribe config show first if you set a custom workspace path, and back up that folder if the transcripts matter to you.

Still stuck?

Run anyscribe doctor and include its full output when you open an issue. It covers dependencies, config, install type, and version — everything needed to reproduce your setup, and nothing sensitive.