Manual Model Download

Stuck on Meetily's Step 1 or Step 2 download screen during onboarding? Your network is probably blocking the download. This guide shows you how to grab the files in your browser, drop them into the right folder, and restart Meetily. No terminal, no commands - about ten minutes total.

A quick checklist so you don't get stuck mid-way:

• A modern browser - Chrome, Edge, or Firefox. Safari sometimes opens large files in a new tab instead of saving them; if that happens, hold Option while clicking the download link to force a save.
• Free disk space - about 700 MB if only Step 1 failed, or around 3 GB if both Step 1 and Step 2 failed.
• A stable connection - if your office or school network is the problem, a phone hotspot is the fastest workaround. Most hotspots don't apply the same filtering.
• About 10 minutes end-to-end (most of that is the encoder download).

On a Mac with more than 8 GB of RAM, Meetily picks Qwen 3.5 4B. Everyone else (Windows, or Mac with 8 GB or less) gets Qwen 3.5 2B. You don't pick this - Meetily picks for you. The model name shown on the Step 2 screen tells you which one was chosen.

A phone hotspot is the easiest fix. Most hotspots don't have the same corporate filtering, and the downloads will go through.

parakeet-tdt-0.6b-v3-int8 - not parakeet-v3-int8, not Parakeet-tdt-0.6b-v3-Int8 (wrong case), not parakeet-tdt-0.6b-v3-int8 (with trailing space). One wrong character and Meetily won't find the model. Copy-paste the name from this guide.

This block does everything in Section A in one paste - creates the folder, downloads all four files, places them. Skip the right-click downloads in A.1 and the folder setup in A.2 if you use this.

MODEL_DIR="$HOME/Library/Application Support/pro.meetily.ai/models/parakeet/parakeet-tdt-0.6b-v3-int8"
BASE_URL="https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx/resolve/main"
mkdir -p "$MODEL_DIR"

curl -L --fail -o "$MODEL_DIR/encoder-model.int8.onnx"        "$BASE_URL/encoder-model.int8.onnx"
curl -L --fail -o "$MODEL_DIR/decoder_joint-model.int8.onnx"  "$BASE_URL/decoder_joint-model.int8.onnx"
curl -L --fail -o "$MODEL_DIR/nemo128.onnx"                   "$BASE_URL/nemo128.onnx"
curl -L --fail -o "$MODEL_DIR/vocab.txt"                      "$BASE_URL/vocab.txt"

$modelDir = Join-Path $env:APPDATA "pro.meetily.ai\models\parakeet\parakeet-tdt-0.6b-v3-int8"
$baseUrl  = "https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx/resolve/main"
New-Item -ItemType Directory -Force -Path $modelDir | Out-Null

$files = @(
  "encoder-model.int8.onnx",
  "decoder_joint-model.int8.onnx",
  "nemo128.onnx",
  "vocab.txt"
)
foreach ($f in $files) {
  Write-Host "Downloading $f ..."
  Invoke-WebRequest -Uri "$baseUrl/$f" -OutFile (Join-Path $modelDir $f)
}

Meetily expects the variant it picked for you. If it's set up for 2B and you place 4B (or vice versa), the app won't recognise the file.

This block does everything in Section B in one paste - creates the folder, downloads the gguf, places it. Skip B.2 and B.3 if you use this. Defaults to Qwen 2B; if you need 4B, uncomment the two lines marked below before running.

SUMMARY_DIR="$HOME/Library/Application Support/pro.meetily.ai/models/summary"
mkdir -p "$SUMMARY_DIR"

# Default: Qwen 3.5 2B (~1.27 GB)
MODEL_FILE="Qwen3.5-2B-Q4_K_M.gguf"
HF_REPO="unsloth/Qwen3.5-2B-GGUF"
# For Qwen 3.5 4B (~2.6 GB):
# MODEL_FILE="Qwen3.5-4B-Q4_K_M.gguf"
# HF_REPO="unsloth/Qwen3.5-4B-GGUF"

curl -L --fail -o "$SUMMARY_DIR/$MODEL_FILE" "https://huggingface.co/$HF_REPO/resolve/main/$MODEL_FILE"

$summaryDir = Join-Path $env:APPDATA "pro.meetily.ai\models\summary"
New-Item -ItemType Directory -Force -Path $summaryDir | Out-Null

# Default: Qwen 3.5 2B (~1.27 GB)
$modelFile = "Qwen3.5-2B-Q4_K_M.gguf"
$hfRepo    = "unsloth/Qwen3.5-2B-GGUF"
# For Qwen 3.5 4B (~2.6 GB):
# $modelFile = "Qwen3.5-4B-Q4_K_M.gguf"
# $hfRepo    = "unsloth/Qwen3.5-4B-GGUF"

Write-Host "Downloading $modelFile ..."
Invoke-WebRequest -Uri "https://huggingface.co/$hfRepo/resolve/main/$modelFile" `
  -OutFile (Join-Path $summaryDir $modelFile)

You'll know the manual install worked when you see:

• A green 100% completion state on the previously stuck onboarding screen, with the wording "Transcription Model Ready" (Step 1) or "Summary Model Ready" (Step 2). The progress bar is full, and Proceed to Step 2 or the available forward button is active.
• If you'd already finished onboarding before starting this guide: open Meetily and check Settings → Transcript (the transcription model shows Downloaded) and Settings → Built-in AI Models (the summary model shows Downloaded).

If you don't see those, jump to Troubleshooting below.

If your network blocks the in-app downloader regularly (corporate environments, schools, certain ISPs), forward this to your IT team: please allowlist huggingface.co and cas-bridge.xethub.hf.co (HuggingFace's CDN that serves the actual file bytes). Once allowlisted, Meetily's normal in-app downloader works on the first try and you'll never need this guide again.