Last updated: 17 September 2026
Parallel Partners uses third-party AI service providers, including OpenRouter, Anthropic Claude, Google Gemini, and OpenAI.
We use open-source models, which have reusable code and licences. We use open-weight models, which have publicly available parameters. We also use proprietary models that their providers control. We select models based on the requirements of each task and the data protections available for each endpoint.
We also use open-source and open-weight models that we host locally.
These services power features such as summarising and analysing user-provided content, drafting documents and recommendations, and helping people work with information in Parallel Partners products.
When you use an AI feature, Parallel Partners may process the relevant content and context you provide through these services. We use that data only to provide the feature you requested. We do not sell customer data, use it for advertising, or use it to train, fine-tune, or improve general-purpose AI or machine learning models.
AI requests can go directly to the selected model provider or pass through OpenRouter. When used, OpenRouter routes each request to the selected provider. OpenRouter states that it does not retain prompts unless prompt logging is specifically enabled. Its Zero Data Retention policy explains how it identifies endpoints that do not retain request data or train on it.
We do not opt customer data into prompt logging. We use OpenRouter's data-policy controls to route requests only to provider endpoints consistent with our no-training commitment.
When a feature uses Claude, Parallel Partners sends necessary prompts and context directly to Anthropic or through OpenRouter. This transfer lets Anthropic generate the requested response.
Anthropic states that it does not train on chats or coding sessions from commercial services by default. Training requires explicit participation in its Development Partner Program or feedback submitted for that purpose. We do not opt customer data into either use. Anthropic's commercial model-training policy explains these rules.
When a feature uses Gemini, Parallel Partners sends necessary prompts and context directly to Google or through OpenRouter. This transfer lets Google generate the requested response.
Google's Gemini API terms distinguish between unpaid and paid services. Under the paid-service terms, Google does not use prompts or responses to improve its products. Google only logs them for a limited period for safety, security, or legal requirements. We route customer data only under terms consistent with our no-training commitment. The Gemini API Additional Terms of Service explain these rules.
When a feature uses an OpenAI model, Parallel Partners sends necessary prompts and context directly to OpenAI or through OpenRouter. This transfer lets OpenAI generate the requested response.
OpenAI states that it does not use API data to train or improve its models by default. OpenAI can use the data for training only after a customer explicitly opts in.
By default, OpenAI retains abuse-monitoring logs for up to 30 days. These logs can contain prompts and responses. OpenAI can retain the logs longer for legal or safety reasons. Eligible customers can apply for additional retention controls. We do not opt customer data into model training. OpenAI's API data controls explain these rules.
If you connect a Google service, we only access the data you authorise and only for the feature you choose to use. We request the minimum Google API permissions needed for that feature. You can review or remove that access through Manage Google access at any time.
We use and transfer raw or derived Google API data according to the Google API Services User Data Policy. This policy includes the Limited Use requirements. We do not use Google data for advertising or sell it. We allow people to read it only with your permission, for security, or where the law requires it.
An agentic system uses AI agents to complete tasks and make decisions. We keep records that make it possible to examine how each system works.
For systems that we build with operators, we record intent and requirements in a structured format. We record material decisions. These decisions include the model choice, the provided context, and each human approval. We trace the original operator intent through to the outcomes.
These records support accountability and help limit undocumented drift. They also keep business judgment and proprietary context under operator control. We continue to strengthen these practices. Our goal is to keep agentic systems inspectable and operator-owned over time.
Built in parallel.