OVERVIEW
Providers
Overview of AI Provider Data Policies
Our integrated AI providers adhere to strict data privacy and security protocols, ensuring that no user data is stored or used for training their models. We serve as a secure intermediary, facilitating AI interactions without exposing your data to external parties. The data is processed in a manner where only relevant pieces are shared with AI providers via a Retrieval-Augmented Generation (RAG) pipeline, ensuring privacy and compliance. Below is a table summarizing the data retention and AI policies for each integrated provider:
Provider | Stored with Pulze | Stored with Data Provider | Data Retention Policy | AI Policy |
---|---|---|---|---|
Anthropic | ✅ | ❌ | Data is not used for training unless opted in by the user (we opted out). | Data used only under specific conditions related to Trust & Safety. Opt-out Information |
Cohere | ✅ | ❌ | Logging for training is switched off. | No data is used for model training when logging is disabled. |
Fireworks | ✅ | ❌ | SOC2 and HIPAA compliant. No prompt data is logged except for specific analytics. | Logging Details |
Google Gemini API | ✅ | ❌ | Data is stored transiently and not used for product improvement. | Data may be used for operational purposes but not for model training. Terms |
Groq | ✅ | ❌ | Does not permanently retain any user data. | Customer data is not stored or used for model training. Terms |
Mistral AI | ✅ | ❌ | Logs stored for 30 days for auditing. | Data used solely for auditing and compliance. Privacy Policy |
OctoAI | ✅ | ❌ | Prompts retained for 15 days; outputs are not stored. | Data is not used for training unless opted in for fine-tuning. Terms |
OpenAI | ✅ | ❌ | Data retained for up to 30 days for safety and legal purposes. | Data is not used to improve model performance. Usage Policy |
TogetherAI | ✅ | ❌ | Storage of prompts and responses is disabled. | Data is not used for training models when logging is disabled. |
Each provider ensures that customer data is processed securely, with no impact on model training. This guarantees that the data remains private and is handled responsibly.
Web Search and Media Uploads
- Web Search: When a web search is requested, the query is not directed to a single search engine. Instead, the system intelligently distributes the query across leading search providers—similar to how the most suitable AI is selected for specific tasks. The top search results are retrieved, and their content is temporarily indexed into a secure vector database. This process allows for the analysis and extraction of the most relevant information to provide the most accurate and comprehensive answer to the query, without permanently storing the visited webpages.
- Media Uploads: Uploaded audio files and images are handled with the utmost care. They are stored securely within dedicated object storage, employing industry-standard encryption and access control mechanisms. The media remains confidential and protected within the infrastructure.
Was this page helpful?