Togo AI engages the third parties below as Subprocessors to process Customer Data on our behalf in providing the Service. Capitalised terms have the meaning given in the Terms and the DPA.
Infrastructure Subprocessors
| Subprocessor | Purpose | Processing location |
|---|---|---|
| Cloudflare, Inc. | Event and error collection, event transport, analytical storage and query, object storage, DNS, CDN and static hosting | Global edge locations; stored at rest in the United States |
| Microsoft Corporation | Hosting for the application and its database — Azure Container Apps and Azure Database for PostgreSQL | United States |
AI Subprocessors
| Subprocessor | Purpose | Processing location |
|---|---|---|
| Deep Infra, Inc. | Model inference for analysis the Service performs for you | United States |
| OpenAI OpCo, LLC | Model inference for analysis the Service performs for you | Global |
For AI processing we commit that:
- no model provider trains its models on Customer Data (Terms §4.10);
- we send a model provider only what one analysis needs: aggregated metrics, their breakdowns and your event and property definitions — never individual events, person profiles or the identifiers that identify people or devices. A property you define as identifying is treated as one; a value you put in any other property is yours under Terms §4.3, and is sent as you sent it;
- we do not keep Customer Data at a model provider. A provider may keep a request for a limited time for its own abuse and security monitoring, and for no other purpose, as stated below;
- we use only models the provider runs on its own infrastructure.
What each AI Subprocessor keeps:
- Deep Infra holds a request in memory for the duration of inference, does not store it on disk, logs only request metadata such as its ID and cost rather than its content, and does not use it to train models.
- OpenAI keeps API requests in abuse-monitoring logs for up to 30 days, or longer where the law requires it or where it is reasonably necessary to protect its services or others from harm, and does not use them to train or improve its models.
Changes to this list
We may add or remove Subprocessors. We post updates on this page and give notice in accordance with the DPA §8, which also sets out how to object to a new Subprocessor.