LLM Data Retention & Privacy by Provider 2026
Vendor policy summaries for every AI provider Meetily can connect to via BYOK, plus the local options that keep summaries on-device. Each summary is dated and links to its source - always confirm against the linked vendor policy.
Each page documents the vendor's default retention, zero-data-retention (ZDR) availability, training defaults, and how Meetily routes traffic through that provider.
Providers
llama.cpp
llama.cpp runs LLMs entirely on your own hardware. There is no remote service, no telemetry, and no training on your prompts. Its bundled llama-server exposes a...
LM Studio
LM Studio runs LLMs entirely on your own machine, so prompts and documents never leave the device, nothing is retained server-side, and its local OpenAI-compati...
Ollama
Ollama runs LLMs entirely on your local machine, so prompts and responses never leave your device, there is no remote retention period, and there is no training...
Anthropic Claude
Anthropic's commercial terms commit that prompts and outputs from the Claude API are not used to train models by default. Verified policy summary plus how to ro...
Azure OpenAI
Azure OpenAI commits that prompts, completions, and training data are not used to train OpenAI's or Microsoft's models, with a 30-day abuse-monitoring window an...
Cohere
Cohere may use API prompts and generations to train its models by default, with a dashboard opt-out, and retains logged data for 30 days on the SaaS Platform. L...
DeepSeek
DeepSeek's privacy policy states that inputs are processed and stored in China and used to train and improve its models by default, with an opt-out. No fixed AP...
Fireworks AI
Fireworks AI runs serverless inference with zero data retention by default and does not train on your prompts or API inputs without opt-in. Summary of the verif...
Google Gemini
Google's paid Gemini API and Vertex AI commit not to train on customer prompts and responses. Free-tier Google AI Studio operates under different terms. Verifie...
Groq Cloud
Groq does not retain inference data by default, does not train on customer Inputs or Outputs, and now lets any customer enable Zero Data Retention self-serve in...
Mistral La Plateforme
Mistral's paid APIs are not used to train its models, with a default 30-day abuse-monitoring retention window and Zero Data Retention available on the Scale pla...
OpenAI
OpenAI's API does not train on customer data by default and retains inputs and outputs for a short abuse-monitoring window. Zero data retention is available for...
OpenRouter
OpenRouter does not log your prompts by default and offers Zero Data Retention controls, but retention and training also depend on the downstream provider you r...
Perplexity
Perplexity's Sonar API operates under a Zero Data Retention policy: prompts and responses are not retained and are not used to train models. This page covers th...
Together AI
Together AI's terms state it does not train on customer data without explicit opt-in, and its API supports zero data retention. Verified policy summary plus how...
xAI Grok
xAI's API documentation states it never trains on your API inputs or outputs without explicit permission, with a 30-day audit retention window and enterprise ze...
Use any of these providers in Meetily via BYOK
Meetily transcription is always 100% local. For summaries, bring your own key for any cloud provider above, or run summaries locally with a local engine and keep the entire pipeline on-device.