Subprocessor List

Echo engages the third parties below to help provide the Service. A subprocessor processes personal information or user content on Echo's behalf. This list supports the disclosures in our Privacy Policy, Terms of Use, and Memory Usage Terms. Retention windows reflect Echo's current plans and configurations and will be updated if these change.

User-connected AI accounts (your own ChatGPT, Claude, Gemini, DeepSeek, Grok, or Perplexity logins) are Connectors you control, not Echo subprocessors. The AI providers below are engaged by Echo to process content on Echo's behalf (embeddings and memory extraction).

Infrastructure and hosting

Subprocessor	Purpose	Data processed	Location	Retention (Echo's plan)
Supabase	Primary database, authentication (including account/transactional emails such as verification and password reset), storage, realtime, vector/search data	Account data, authentication data, memories, source conversations, embeddings, metadata, and stored files	US (West US / N. California)	Active rows deleted per Echo workflow; daily backups retained 7 days (Pro plan, PITR off)
Vercel	Application hosting and serverless deployment (extension backend, web app)	Request metadata, function/runtime logs, operational telemetry. Raw user content is not written to logs or deployment artifacts.	US	Runtime logs ~1 day (Pro). Build logs / deployments retained by default — kept free of user content.
Cloudflare, Inc.	Domain registration and authoritative DNS; CDN and reverse proxy; TLS termination; caching; DDoS protection and WAF; HTTP traffic analysis; cookieless Web Analytics/RUM	Visitor IP addresses; request and connection metadata; headers; URL/path; timestamps; user-agent, browser/device and performance data; security signals; and website or API content routed through or cached on Cloudflare’s network	US provider; global edge network	Web Analytics retains unsampled beacon data ~7 days and aggregated analytics longer (dashboard access currently up to 6 months); other processing per Cloudflare’s configuration and DPA

AI processing

Subprocessor	Purpose	Data processed	Training?	Abuse-log retention
OpenAI	Embedding generation (/v1/embeddings) for memory search	Memory text / search text submitted for vector embedding	No (not used for training; no app-state retention; ZDR-eligible)	Up to 30 days
Google — Gemini / Vertex AI	Memory extraction, structured generation, classification, labels, and AI-generated answers and summaries (Deep Search, Workspace briefings)	Conversation text, extraction prompts, context, generated structured outputs	No (paid services / Vertex — not used to train Google's models)	Up to 55 days (paid Gemini abuse monitoring)

Operational services

Subprocessor	Purpose	Data processed	Scope	Consent / notes
PostHog	Operational analytics; product analytics for Echo for Mac; cookieless anonymous website measurement; consent-based persistent analytics and replay	Operational metadata; cookieless page views and allowlisted acquisition or setup interactions; with consent, persistent usage metadata and privacy-protected replay on selected public pages; for Echo for Mac, app-usage events with a random installation identifier, app and macOS versions, and your Echo user identifier after sign-in. No private memory content.	Web platform, Echo for Mac, and selected website pages	PostHog DPA. Cookieless mode stores no browser identifier or persistent profile. Persistent IDs, autocapture, and replay require consent. Inputs are masked and sensitive routes are excluded from replay.
Amplitude	Product usage analytics for non-web Echo surfaces while migration completes	Usage and interaction metadata. No memory content.	Chrome Extension, MCP tools, and Echo Chat	Processor under DPA (incorporated into Amplitude ToS).
Google — Firebase Crashlytics	Crash and error diagnostics for Echo Chat (app)	Crash/error diagnostic data (device/OS, stack traces, event metadata). No memory content.	Echo Chat app	Google/Firebase data-processing terms; crash/diagnostic data retained ~90 days (Firebase default). No memory content or precise identifiers.
Stripe	Payment processing for paid plans (website checkout, including checkout opened from Echo for Mac)	Payment card details (collected directly by Stripe, not through Echo); billing/transaction metadata. No memory content.	Website checkout	Processes under Stripe’s own terms; Echo does not store full card numbers

Software distribution

Echo for Mac downloads and its signed update feed are hosted on GitHub (GitHub, Inc., US), and the Echo MCP server package is distributed through the npm registry (operated by GitHub, Inc.). When you download Echo for Mac, check for updates, or install or update the Echo MCP server, those services receive standard request information — such as your IP address, user agent, and the file requested — under their own terms and privacy policies. They do not receive your account data or memory content from Echo.

How we manage subprocessors

We require subprocessors to protect personal information under obligations consistent with our commitments and applicable data-processing terms. We update this list when we add or remove a subprocessor or materially change how user content is processed or retained.