Support
ParlaAI FAQ — Common Questions Answered
Everything you need to know about ParlaAI — from installation to daily use.
🔵 General Information
About ParlaAI
What is ParlaAI?
ParlaAI is a local, privacy-first AI assistant for investigative analysis, document review, audio/video transcription, and multilingual translation. All processing happens on your machine — no data is sent to Parla Digital servers.
What does Parla mean?
Parla means "to speak" in Italian. "We Speak Computers" means you don't have to.
Does ParlaAI require an internet connection to use?
No. After the initial setup and model downloads, ParlaAI runs offline. Internet access is only needed during installation to download the AI models.
Is my data private?
Yes. All files, transcriptions, translations, and AI conversations stay on your device. Nothing is uploaded to the cloud or shared with any third party.
What operating systems are supported?
Windows 10 and Windows 11 (64-bit). Mac and Linux versions are in development.
⚙️ Installation & Setup
Getting Started
What do I need before installing?
Nothing — the installer handles everything automatically. It will detect and install Python, Ollama, and FFmpeg if they are not already on your machine. All you need is a Windows 10 or 11 computer with 20 GB of free disk space and an internet connection for the initial setup.
How do I install ParlaAI?
Once downloaded, double-click the installer and then follow the on-screen instructions. Internet is required for the initial installation but is not required after install.
How long does installation take?
Installation typically take 20 to 40 minutes depending on your internet speed and hardware.
Do I need a GPU?
No. ParlaAI runs entirely on CPU and is fully functional without any GPU. A GPU is optional but significantly speeds up transcription and AI responses and is highly recommended. The installer automatically detects your hardware and selects the best AI models for your setup. NOTE: An Nvidia 8GB VRAM GPU or higher will offer better performance and better AI chat quality versus CPU alone.
Does ParlaAI support AMD (Radeon) GPUs?
GPU acceleration in ParlaAI is built on NVIDIA CUDA. AMD GPU acceleration is not currently supported — on a machine with only an AMD GPU, ParlaAI will fall back to CPU processing. All features remain available.
What is the HuggingFace token for?
The token is optional and only used to download the speaker identification model (pyannote/speaker-diarization-3.1) during setup. After that, the model is stored locally and the token is never used again. You must accept the model's terms on the HuggingFace website before the download will work. Speaker identification can be skipped entirely and enabled later by re-running the installer.
Where do I get a HuggingFace token?
Create a free account at huggingface.co.
Go to Settings → Access Tokens → New Token.
Accept the terms for pyannote/speaker-diarization-3.1 and pyannote/segmentation-3.0.
Paste the token into the installer when prompted.
✨ Features
What ParlaAI Can Do
What file types are supported?
Transcription: MP3, MP4, WAV, M4A, FLAC, OGG, MKV, AVI, MOV, WEBM
Translation: DOCX, XLSX, XLS, PDF, RTF, TXT, MDAI
Chat / Document Analysis: DOCX, XLSX, PDF, RTF, TXT, MD, CSV
What languages can ParlaAI translate?
English, Arabic, Bulgarian, Chinese, Czech, Danish, Dutch, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Portuguese, Romanian, Russian, Spanish, Swedish, and Ukrainian.
Can ParlaAI identify who is speaking in a recording?
Yes, using the Speaker Identification feature. It can label distinct speakers (SPEAKER_01, SPEAKER_02, etc.) in transcripts but cannot identify speakers by name. Accuracy varies with audio quality and number of speakers. This feature requires an optional setup step during installation.
What is the difference between "Download Report" and "Download Translated Document"?
Download Report — a plain text file containing the translated content along with metadata (filename, file hash, language pair, timestamp). Good for record-keeping.
Download Translated Document — the original file with text translated in place, preserving the original layout and formatting (DOCX, XLSX, etc.). For PDFs, the output is a DOCX.
What is a Workspace?
Workspaces let you organize files and conversations by case, project, or topic. Files uploaded to a workspace stay associated with it, and the AI chat can search and reason across all documents in that workspace simultaneously.
Can I process multiple files at once?
Yes. Use the Bulk Jobs tab to process entire folders of audio/video files or documents in one operation.
⚡ Performance
Speed & Resources
Why is processing slow?
AI models are computationally intensive. On CPU-only hardware, transcribing one hour of audio can take 1–3 hours. Translation of large documents (100+ pages) can take 30–60 minutes. A GPU significantly reduces these times.
Why does the first task take longer than later ones?
AI models are loaded into memory on first use and kept loaded for subsequent tasks. The first transcription or translation will include model loading time; later tasks skip that step.
The app seems stuck. What should I do?
Large files take time — check the status message in the interface. If the status has not changed for more than 30 minutes and the file is not unusually large, try restarting the application. Do not close the app while processing is in progress.
🤖 AI Accuracy & Limitations
Understanding AI Output
Important
AI models can and do make mistakes. All outputs — including transcriptions, translations, and AI-generated analysis — should be treated as a starting point, not a final product. Always have a qualified human review results before use in any legal, investigative, or professional context.
Can I trust the transcription output?
Transcription accuracy depends on audio quality, accents, background noise, and the number of speakers. Results can contain errors, misidentified speakers, and missed words. Do not use raw transcription output as evidence without human review.
Can I trust the translation output?
Machine translation handles common vocabulary and sentence structure well, but regularly misses nuance, idioms, cultural context, and domain-specific terminology. A qualified human translator should review any translation used in an official or legal context.
Can I trust the AI chat responses?
The AI assistant can reason, summarize, and make connections across documents, but it can also hallucinate facts, misread context, or draw incorrect conclusions. Treat all AI responses as suggestions that require verification, not authoritative answers.
Auto-detect identified the wrong source language. What should I do?
Language auto-detection works by analyzing statistical patterns in the text. Very short input — typically fewer than 5–6 words — may not provide enough context for reliable detection. If you notice an incorrect detection, manually select the source language from the dropdown and re-run the translation. For best results, always specify the source language explicitly when working with short or isolated phrases.
🔒 Privacy & Security
Data & Confidentiality
Does ParlaAI send any data to the internet?
No. After initial setup, ParlaAI has no outbound network connections during normal operation. No files, transcripts, translations, AI conversations, or any other data are transmitted to any server.
Are my files stored permanently?
Files are stored locally on your machine. You can delete them at any time from within the app or directly from the filesystem.
Does ParlaAI send any telemetry or analytics?
No. ParlaAI has no telemetry, usage tracking, or analytics of any kind.
Can I use ParlaAI on an air-gapped machine?
Yes. Once models are downloaded during setup, the app runs fully offline with no outbound connections required. It is suitable for use on air-gapped or restricted networks.
💳 Pricing & Licensing
Purchasing
Is there a subscription fee?
No. ParlaAI is a one-time purchase. There are no monthly or annual fees of any kind.
What does the license cover?
Each license covers one machine and includes all features plus one year of software updates.
Can I move my license to a new machine?
On a case-by-case basis. Contact suppport@parladigitalsolutions.com and our team will assist you.
Is there a free trial?
Yes — every download includes a free 7-day trial so you can evaluate ParlaAI before purchasing. Visit the Download page to get started.
Do you offer volume pricing?
Yes. Bulk pricing is available for teams and organizations. We support government procurement processes and can provide invoicing. Request a quote and we will respond within one business day.
🔧 Troubleshooting
Common Issues
The installer failed partway through. What do I do?
Re-run the installer. The installer detects what is already complete and skips those steps, so it is safe to run multiple times.
ParlaAI shows a blank screen after launching.
The backend is still starting up. Wait 10–15 seconds — it will load automatically. On first launch after a restart this can take up to 30 seconds.
Speaker identification is not working.
The speaker identification model may not have been downloaded during setup. Re-run the installer and enter your HuggingFace token when prompted. Make sure you have accepted the model terms at huggingface.co before running the installer.
Translation says "language pair not supported."
Not all language combinations are available. Check the supported languages list above. Translation currently requires one language in each pair to be English or a closely related language.
The "Download Translated Document" button does not appear.
The translation may have failed silently. Common causes include password-protected files, unsupported file encoding, or very large files. Try re-uploading the file and translating again.
My GPU was not detected. How do I fix this?
Make sure your NVIDIA drivers are up to date. Then delete the venv folder inside the ParlaAI resources\app\ directory and re-run the installer — it will reinstall PyTorch with CUDA support.
© 2026 Parla Digital Solutions. All rights reserved.