Sub-processors
The third parties that process personal data for Cusp, what they do, and where.
Обновлено
These are the companies that process personal data on our behalf to provide Cusp. Our Data Processing Addendum explains how we choose and oversee them, and how we notify customers of changes: we update this page at least 30 days before a new sub-processor starts processing customer personal data. To get notified, email privacy@cusp.run with the subject "Sub-processor updates".
Services that run only on your device, and providers you connect with your own keys (BYOK), aren't our sub-processors: with BYOK, your agreement with that provider applies.
Infrastructure and operations
| Sub-processor | Purpose | Data | Location | Transfer safeguard |
|---|---|---|---|---|
| Railway Corporation | Hosting of the website, web app, cloud API and Hub; managed PostgreSQL and Redis; object storage for hosted artifacts; cloud runners | All cloud data: accounts, synced sessions, hosted artifacts, usage records, Hub data | European Union (Amsterdam, the Netherlands): data and the services that handle requests. Railway is a US company, so its staff may access the data from the United States for support and operations | Railway's DPA: EU SCCs with the UK Addendum, or the EU-US Data Privacy Framework and UK Extension where Railway is certified |
| Cloudflare, Inc. | DNS, TLS, delivery and attack protection for cusp.run, nibusercontent.com and their subdomains | All traffic to our website, web app, API, Hub and hosted artifacts: IP addresses, request headers, and the content of requests and responses, which Cloudflare decrypts and re-encrypts on the way to our servers. It caches only public website pages | Cloudflare's global network: the data centre nearest each visitor, which may be outside the UK and the EU | Cloudflare's DPA: EU SCCs with the UK Addendum; Cloudflare is certified under the EU-US Data Privacy Framework and UK Extension |
| Stripe Payments UK Ltd / Stripe, Inc. | Payments, subscriptions, invoices, tax calculation | Name, email, billing address, payment details, subscription and usage amounts | United Kingdom, United States | EU-US Data Privacy Framework and UK Extension; SCCs in Stripe's DPA |
| Apple Inc. | Push notifications to the iOS app (Apple Push Notification service) | Device push token, notification title and identifiers | United States | Apple Developer Program License Agreement |
| Eleven Labs Inc. | Speech-to-text for cloud dictation | Live voice audio, the transcribed text, and hint words from the session (project, branch, file, symbol and crew member names). Nothing is stored by us | United States and other countries; its DPA doesn't fix a location, and EU data residency is offered only on its Enterprise plan | ElevenLabs' DPA (elevenlabs.io/dpa): EU SCCs with the UK Addendum. Our account is opted out of training on our audio (since 2026-10-23); zero retention needs its Enterprise plan |
| Perplexity AI, Inc. | Managed web search for the agent | The search query (which may include words from the session) and the results. Nothing is stored by us beyond returning them | United States and other countries; its DPA doesn't fix a location | Perplexity's DPA: EU SCCs with the UK Addendum. Under its API terms, queries and results aren't used for training |
Managed model gateway
These apply only when you use Cusp's managed gateway (our keys), not when you use your own keys.
| Sub-processor | Purpose | Data | Location | Transfer safeguard |
|---|---|---|---|---|
| OpenRouter, Inc. | Routes managed-gateway requests to model providers | The full content of each request and response (prompts, code, files and tool output the agent sends), plus a session identifier | United States | EU SCCs, as set out in OpenRouter's privacy policy |
| Model providers reached through OpenRouter | Run the model you selected | The full content of each request and response | Depends on the model and provider. See AI and Model Providers | Varies by provider and model |
The provider that serves a request depends on the model you choose. For example, Claude models are served by Anthropic (or by Amazon Bedrock or Google Vertex AI), GPT models by OpenAI or Microsoft Azure, and Gemini models by Google. Open-weight models (such as DeepSeek, Qwen, GLM and Kimi) may be served by inference hosts such as Together AI, Fireworks AI, DeepInfra, Groq, Cerebras, Novita, SiliconFlow or the model's developer. Some of these process data outside the UK, the EU and the US, including in China. Every managed-gateway request tells OpenRouter to exclude providers that collect or train on the data. Team and Enterprise workspaces can also restrict which providers and models may be used, and can require zero data retention.
Services you choose to connect
These aren't sub-processors, because you decide to connect them and they act under their own terms, but they receive data when you use them:
| Service | When | What they receive |
|---|---|---|
| GitHub, Inc. | You sign in with GitHub, install the Cusp GitHub App, or the agent works with GitHub | Your GitHub identity; repository content and events for the repositories you selected |
| Search providers (Brave, Tavily, Exa) | You add a search key and the agent uses web search | Search queries |
| MCP servers and Hub extensions | You install them | Whatever the agent sends them |
Planned
Not yet in use. We'll update this page and give notice before they process customer personal data:
| Sub-processor | Purpose |
|---|---|
| Resend | Transactional email (sign-in links, invitations, receipts), EU region |
| Sentry | Error monitoring for our servers, and crash reports from the apps if you've opted in to send them; EU region. Receives error details only, with messages scrubbed, and no user identity or request data |
| Better Stack | Uptime monitoring and status page |
| Parallel Web Systems Inc. | Additional managed web search provider. Enabled only after zero data retention or a written training opt-out is confirmed, because it trains on queries by default |