Supported subprocessors
Each company below receives only what its job needs. This page is generated from one canonical list in the repository (packages/core/src/subprocessors.ts) and checked against supported routing configurations. Optional entries do not mean that every provider receives your data.
| Company | Category | What it handles | Where, retention and training |
|---|---|---|---|
| OpenAI | AI model providers | Optional direct language-model provider when configured for model work or comparison checks. | Depends on the selected endpoint and account configurationRetention and training depend on the selected endpoint and account terms. Direct routes do not inherit the OpenRouter zero-retention request setting. |
| Together AI | AI model providers | Optional direct language-model provider when configured for model work or comparison checks. | Depends on the selected endpoint and account configurationRetention and training depend on the selected endpoint and account terms. Direct routes do not inherit the OpenRouter zero-retention request setting. |
| Fireworks AI | AI model providers | Optional direct language-model provider when configured for model work or comparison checks. | Depends on the selected endpoint and account configurationRetention and training depend on the selected endpoint and account terms. Direct routes do not inherit the OpenRouter zero-retention request setting. |
| Groq | AI model providers | Optional direct language-model provider when configured for model work or comparison checks. | Depends on the selected endpoint and account configurationRetention and training depend on the selected endpoint and account terms. Direct routes do not inherit the OpenRouter zero-retention request setting. |
| fal.ai | AI model providers | Optional media gateway for image, video, music, voice and transcription jobs when configured. | Depends on the selected model endpoint and account configurationReceives the submitted media and prompt. Retention and training depend on the selected endpoint and account terms; this route does not inherit OpenRouter zero-retention settings. |
| Anthropic | AI model providers | Runs the primary language models that read, summarise, draft, plan and check work for your assistant. | United StatesDoes not train on API inputs or outputs without permission. Retention depends on endpoint, model eligibility and account settings, including covered-model exceptions, batch processing and safety or legal holds. |
| OpenRouter | AI model providers | Zero-retention gateway for speech synthesis, web search, and secondary or comparison model calls (routing only to zero-retention endpoints on upstream hosts such as Microsoft Azure, OpenAI, Google, xAI, Amazon Bedrock and open-weight hosts when those routes run). | United StatesEvery request sends zero-retention routing (zdr) with data collection denied, and omits headers that identify you or Avery. Keeps no prompts or outputs after processing, except in-flight safety screening or legal preservation. Content promised strict direct zero retention stays on direct routes. |
| DeepInfra | Speech providers | Turns spoken audio into text during voice sessions, and runs open-weight models for spoken openers, generated pictures, opt-in specialist routes and comparison checks. | United StatesHolds audio, prompts and outputs in memory only while the request runs, writes nothing to disk, and never trains on them. Only request IDs and cost metadata are logged. |
| Speechify | Speech providers | Turns your assistant’s reply text into spoken audio on the default voice route, reached through OpenRouter. Receives only the words being spoken; your microphone audio stays off this route. | United States (via OpenRouter)Reached only through OpenRouter’s zero-retention speech endpoint. Reply text is neither kept after the audio is made nor used for training. |
| ElevenLabs | Speech providers | Turns your assistant’s reply text into spoken audio when you pick the alternate voice route, reached through OpenRouter. Receives only the words being spoken; your microphone audio stays off this route. | United States (via OpenRouter)Reached only through OpenRouter’s zero-retention speech endpoint. Reply text is neither kept after the audio is made nor used for training. |
| Microsoft | Speech providers | Fallback text-to-speech synthesis when the primary voice is slow or unavailable, reached through OpenRouter without microphone audio. Microsoft Azure also hosts zero-retention GPT endpoints behind OpenRouter. | United States (via OpenRouter)Reached only through OpenRouter’s zero-retention endpoints, which do not train on inputs or retain prompts and outputs after the reply is made. |
| Perplexity | Web search | Answers live web searches (Sonar, reached through OpenRouter). Receives the search question, including information included in that question. Your assistant is instructed to leave private details out; the search transport does not automatically remove them. | United States (via OpenRouter)Reached through OpenRouter with zero-retention routing (zdr) and data collection denied, so search questions are not used for model training. |
| Vultr | Hosting and delivery | Hosts the dedicated application server, database, encrypted file store and server backups. | United StatesInfrastructure host only. Sign-in tokens, saved credentials, secret keys and stored files are encrypted by Avery before they are written to disk; encrypted backups age out within 14 days after deletion. |
| Cloudflare | Hosting and delivery | Answers DNS lookups for our domain names and serves this public website. | Global edge networkHandles DNS queries and public website pages only. Encrypted connections to the application server terminate directly on the server itself. |
| Apple, Google and Mozilla | Hosting and delivery | Deliver push notifications to your phones and browsers when you turn notifications on. | Global delivery networksEvery notification payload is encrypted end to end to your device (Web Push RFC 8291), so the delivery service sees only the device address and timing and cannot read the title or body. |
Retention terms depend on the endpoint, model eligibility and account settings. See Anthropic’s API retention documentation (reference date: 16 November 2026), including covered-model exceptions. A routing check does not verify an upstream account agreement.
Accounts you connect yourself
Services you choose to connect in the app (such as Google, Slack, GitHub and GitLab) are your own accounts rather than Avery’s subprocessors. Avery sends requests to them only when you connect them and within the permissions you grant.
How this list stays current
When Avery’s model, speech, search or hosting configuration changes, packages/core/src/subprocessors.ts is updated in the same release and checked against supported configurations. Deployment overrides and account retention settings also need to be reconciled when publishing. Read the privacy notice for what Avery keeps, your controls, and how to export or delete your data, or write to privacy@askavery.com.