Subprocessors
Last updated: October 15, 2026 · Original disclosure: September 14, 2026
A subprocessor is a third-party company Becki uses to operate the service. We list them publicly here so you know exactly who has access to which categories of your data, and why.
For each subprocessor we describe its purpose, the data involved and the processing region. Training, retention and security terms depend on the applicable service and agreement. No-training is not zero retention. This page does not claim that Becki has a provider-specific zero-data-retention arrangement.
Current subprocessors
| Subprocessor | Purpose | Data received | Region |
|---|---|---|---|
|
OpenAI Becki-managed API processing only Privacy |
Managed API text processing. Query rewriting, dead-end relevance, project and cross-project memory synthesis, and supersession judgments. This is separate from models or provider accounts you connect to Astir or another AI tool. | Text submitted for the applicable task, including saved memories, conversation context, queries, candidate entries and related task metadata. This can include transcript text when it forms part of the submitted context. OpenAI states that API data is not used for training unless the customer opts in. Abuse-monitoring logs may contain customer content and are generally retained for up to 30 days, with stated exceptions. Application-state retention depends on the API feature and settings. No Becki-specific zero-data-retention arrangement is asserted. See API data controls. | Processing locations are governed by the applicable OpenAI API terms and configuration. No Becki-specific data-residency guarantee is asserted here. |
|
Anthropic, PBC Becki-managed processing only Privacy |
General Messages and retained service-side AI requests. Processes text requests routed through Becki's managed general Messages proxy and retained service-side AI routes. This is separate from models or provider accounts you connect to Astir or another AI tool. | Text submitted for the applicable task, including saved memories, conversation context, queries, candidate entries and related metadata. This can include transcript text when it forms part of the submitted context. Anthropic states that commercial/API inputs and outputs are not used for training by default, subject to its stated exceptions. Standard API retention is generally up to 30 days, with exceptions and other arrangements. That is not a Becki-specific zero-retention assurance. See commercial training policy and the provider's applicable retention terms. | Processing locations are governed by Anthropic's applicable service terms. No Becki-specific data-residency guarantee is asserted here. |
|
Supabase Inc. Privacy |
Authentication, primary application database (Postgres), file storage for release artifacts, and serverless edge functions. Hosts Becki's account directory and the synced cross-tool vault. | Account email + display name; synced vault text and embeddings; commitments, project and meeting metadata; and text received by service-side functions for applicable tasks. Storing a sealed queue row does not mean plaintext was never received during ingestion or that downstream processing is always local. | United States (AWS us-east). Underlying disk encryption: AES-256. |
|
Paddle.com Market Ltd Reseller / Merchant of Record Privacy |
Billing for paid plans. Paddle is Becki's reseller and Merchant of Record. It runs checkout, subscription billing, payments, refunds, and sales tax on Becki's behalf. Becki never sees or stores your full card number. | Name, email, and billing address you enter at checkout; subscription and transaction status. Your card and payment-instrument details are handled by Paddle and its payment processors, not by Becki. | United Kingdom (Paddle.com Market Ltd), with global payment-processing subprocessors. |
|
Fly.io, Inc. Privacy |
Hosts Becki's marketing site and access backend, and the self-hosted embedding service used by server-side memory processing. Local embedding is a separate option where supported. A model's name or artifact download host does not establish a third-party inference relationship. | Email addresses of users who sign up; per-user access tokens; HTTP request logs (IP, user-agent, request path). For the embedding service: vault-entry text and search-query text in server-side MCP requests or desktop fallback requests. The embedding service processes text, not local recording files. | United States |
|
Apple Inc. Privacy |
Code signing and notarization of the Becki Mac application. Distribution mechanics, every Mac that runs Becki contacts Apple's notarization service to verify the app's signature on first launch. | Application binary metadata at notarization time. End-user devices contact Apple for ticket validation; Apple receives standard verification metadata about the binary, not user content. | United States |
|
Resend (Resend, Inc.) Privacy |
Transactional email delivery for signup confirmations, download link emails, and account messages. | Recipient email address; subject line; message body of transactional emails; delivery + bounce metadata. | United States |
Optional integrations you control
These services only see your data if you explicitly connect them in the app. They are not subprocessors of Becki by default, they're third-party services where you are the customer:
| Service | Activated when | Data received |
|---|---|---|
| GitHub / GitLab | You add a project from a GitHub or GitLab repository. | Read-only access to commits, pull requests, and issue metadata you authorize. Becki never writes to your repos. |
| macOS Calendar (EventKit) | You grant Becki Calendar permission for meeting auto-detection. | Event details, attendees and conference links available under the permissions you grant. Selected event details can be used as context for a task you authorize. Calendar access alone does not authorize recording. |
| Other AI tools you connect via MCP (Claude Code, Cursor, Zed, etc.) | You install Becki's MCP server into the AI tool of your choice. | Vault chunks the AI tool retrieves through Becki, scoped to your account. The AI tool acts on your behalf with your authentication. |
How we choose subprocessors
- Only when needed for a specific feature we couldn't reasonably build ourselves (LLM inference, hosted Postgres, deliverable email). Where we can bring a capability on-device (e.g. embedding), we do.
- Each subprocessor must offer commercially reasonable security (TLS in transit, encryption at rest, access controls).
- Review the service-specific data-use, training, retention and security terms. Do not treat a no-training policy as a guarantee of no retention, monitoring or lawful disclosure.
- We prefer subprocessors with public privacy policies and a track record of honoring them.
Notification of changes
If we add or replace a subprocessor that materially changes who has access to your data, we will update this page and notify users via the app or email at least 14 days before the change takes effect, or sooner if required by your jurisdiction. Removals from this list (subprocessors we no longer use) are reflected in the updated date alone.
To object to a new subprocessor, contact support@becki.io within the notice period. If we cannot reasonably accommodate the objection, you may terminate your account and request deletion of your data per the Privacy Policy.
Contact
Questions about subprocessors: support@becki.io.