Nobody trains a model on your business
- We build no models. There is nothing for your data to be trained into.
- Our providers are contractually barred. A term of the agreements we are on, quoted below.
- Two providers. Anthropic and OpenAI, both on the subprocessor list.
- No shared pool. Your data is not added to any aggregated corpus, and no “improving the service” clause permits it.
Read their terms
Understanding
Anthropic
- Entity
- Anthropic, PBC, United States
- Interface
- Claude API, commercial terms
- Training
- Prohibited. “Anthropic may not train models on Customer Content from Services.”
- Retention
- Deleted within 30 days. Flagged content may be held longer.
- Holds
- SOC 2 Type II, ISO 27001, ISO 42001
Search
OpenAI
- Entity
- OpenAI, L.L.C., United States
- Interface
- Embeddings API, business terms
- Training
- Prohibited. Will not use Customer Content “to develop or improve the Services, unless Customer explicitly agrees.” We have not.
- Retention
- Up to 30 days for abuse monitoring, then removed.
- Holds
- SOC 2 Type II, ISO 27001, ISO 42001
Limits of these guarantees
Zero retention is not zero
Both providers may hold flagged content beyond the standard window. Content is deleted within 30 days, except where flagged.
Inference happens in the US
Neither provider offers a Canadian or EU inference region we can use. If your data must never be processed in the US, Apex is not the right product. Transfers are covered by the DPA.
The assistant you connect is yours
Apex serves your brain to whichever AI you point at it, over MCP. That assistant runs under your agreement with its maker. Every read of your content is logged with the client that asked.