Data Processing Addendum

Last updated: November 14, 2026

This Data Processing Addendum ("DPA") applies when an organization uses Context OS on a Team or Enterprise plan and stores personal data in its team vaults. It forms part of the Terms between the organization ("you", the customer) and Aqeeli LLC ("Aqeeli", "we"), and it supplements our Privacy Policy. If this DPA and the Terms conflict about customer content, this DPA prevails.

1. Roles

  • Customer content is what your organization stores or processes in Context OS: vault files and folders, uploaded sources and exports, captured chats, voice notes and Vault Chat questions. For customer content, you are the controller and Aqeeli is your processor. Under California law, we act as your service provider.
  • Account, billing and security data (sign-in details, payments, logs) is processed by Aqeeli as an independent controller under the Privacy Policy.
  • Personal vaults on the Free, Pro or Import Pack plans are covered by the Privacy Policy.
  • In a self-hosted deployment, Aqeeli processes no customer content at all (see Section 6).

2. Details of the processing

| Item | Description | | ---------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Subject matter | Hosting, distilling, serving and syncing your organization's context vaults. | | Duration | For the term of your subscription, then until deletion as described in Section 8. | | Nature and purpose | Uploading sources (secrets are scrubbed in the browser first), AI distillation into Markdown files, storage, and serving to the people and AI tools you authorize (members, MCP clients, the browser extension, share links). Optional sync to your own storage. | | Types of personal data | Whatever your sources contain, typically names, contact details, work and project information, and conversation content. Context OS is not designed for special categories of personal data. | | Data subjects | Your members and the people mentioned in your content, such as employees, clients and partners. |

3. Our obligations as processor

We process customer content only on your documented instructions. Those instructions are the Terms, this DPA and the settings you choose in the product: folder roles, API tokens, share links, bring-your-own-key, bring-your-own-storage and the passphrase lock. In particular, we:

  • never use customer content to train AI models and never sell it;
  • make sure everyone who can access customer content is bound by confidentiality;
  • help you answer data subject requests and meet your security, impact-assessment and consultation obligations, taking into account the nature of the processing;
  • notify you without undue delay after we become aware of a personal data breach affecting your customer content;
  • make available the information needed to demonstrate compliance with this DPA, and allow reasonable audits, primarily through documentation and written answers to security questionnaires.

4. Security measures

  • Encryption in transit (TLS) and at rest. Vault files and Quarantine diffs are encrypted with a per-vault AES-256-GCM data key, which is wrapped with a master key held only on our servers.
  • An optional zero-knowledge passphrase lock. Your browser encrypts every file with a key derived from your passphrase (PBKDF2-SHA-256 + AES-256-GCM). While a vault is locked, we store only ciphertext, and AI, MCP and extension features pause.
  • Secret scrubbing. Keys, tokens, passwords, payment-card numbers and IBANs are replaced with placeholders before they are stored or sent to the AI provider. Imports, quick-start text and extension captures are scrubbed in the browser and again on our servers, and local-folder uploads in the browser. Voice-note transcripts, writes from connected AI tools and Vault Chat questions are scrubbed on our servers. An editor save that contains a secret is refused until the user confirms it, and the secret is then stored as a placeholder. Files pushed through the sync API with a secret are not stored. Email addresses and phone numbers are also redacted from imports, quick-start text, voice notes, extension captures and AI-tool writes, but kept in text that users write in the editor, sync from a local folder or ask in Vault Chat.
  • Folder roles (Owner, Admin, Member) apply in the app and to MCP and API tokens. Share links are limited to one folder and expire automatically.
  • API tokens and share links are stored only as salted hashes. AI keys and storage credentials are stored encrypted.
  • Administrative actions are recorded in an audit log, and rate limits protect sign-in and public endpoints.

5. Subprocessors

You authorize the subprocessors below. This is the same list as Section 8 of the Privacy Policy.

| Subprocessor | Purpose | Customer data processed | Location | | ------------------------------ | ----------------------------------------------------------- | -------------------------------------------------------------------------------------------- | ----------------------------------------- | | Google (Gemini API, paid tier) | AI distillation, routing, Vault Chat and voice-note routing | Secret-scrubbed vault text and prompts. Not used to train Google's models. | United States and other Google locations | | OpenAI | Server-side voice transcription, only when you use it | Voice-note audio | United States | | Neon | Managed PostgreSQL database | Account data, encrypted vault files, staged uploads and usage records | United States | | Vercel | Application hosting and serverless compute | Data processed in transit and in memory, and request logs | United States, with a global edge network | | Inngest | Background jobs (distillation, cleanup, storage sync) | Job identifiers and status, and intermediate step results that can include distilled content | United States | | Resend | Transactional email and our newsletter list | Email address, name and email content | United States | | PostHog | Product analytics of the conversion funnel | Account or demo identifiers, event names and coarse values. No vault content. | United States (or EU region) | | Stripe | Payments, subscriptions and tax | Billing details and transaction history. No vault content. | United States and other Stripe locations |

We update this list before a new subprocessor starts processing customer content. You may object to a new subprocessor on reasonable data protection grounds by writing to privacy@contextos.dev.

6. Keeping processing out of our cloud

  • Bring your own key (BYOK). AI calls for that vault go to your own AI provider account (for example OpenAI, Anthropic or Google) or to a local Ollama or vLLM endpoint, instead of our Gemini subprocessor. That provider acts under your agreement with it. The vault is still stored and served by our cloud.
  • Bring your own storage (BYOS). A copy of the vault is pushed to your GitHub or git host, or to your S3-compatible bucket. That provider is not our subprocessor. Distillation still runs on our servers.
  • Self-hosted (Enterprise). The whole stack runs on your infrastructure with docker-compose.enterprise.yml: the app, PostgreSQL, object storage, background jobs and optional local models through Ollama. The license key is verified offline. Customer content therefore never passes through Aqeeli or any of the subprocessors above. You choose your own AI provider, or run models locally.

7. International transfers

Transfers of personal data from the European Economic Area, the United Kingdom or Switzerland rely on the safeguards described in Section 9 of the Privacy Policy. These are the EU-U.S. Data Privacy Framework where the recipient is certified, and otherwise the European Commission's Standard Contractual Clauses (with the UK addendum where relevant).

8. Return and deletion

  • You can export your vaults at any time as Markdown files or a ZIP archive, or sync them to a local folder or your own storage.
  • Deleting a vault, or deleting your organization, permanently deletes its files, folders, API tokens and share links from our database. Deleted data can remain in encrypted backups until those backups expire under our providers' standard retention.
  • Staged upload text is deleted when the distillation job ends. Leftovers are purged 24 hours after the job last changed. Sources a budget-limited build could not distill are kept encrypted for up to 7 days, then deleted.

9. A signed copy

Team and Enterprise customers who need a countersigned DPA, for example for a vendor review, can request one at privacy@contextos.dev or through the contact form.