Your memory belongs to you. Vanilla Brain processes it to provide the features you authorize. We do not use your content to train AI models, sell it, use it for advertising, or routinely read it.
Effective date: October 20, 2026. Changes from October 15, 2026: an optional, owner-enabled request checklist and an optional AI-selected document retention mode. Both are off unless you turn them on. Automatic compaction capture and continuity injection remain unavailable; a transcript-recovery setting is reserved and collects nothing in this release.
Vanilla Brain is an online memory service for AI agents. Agents you connect can save and retrieve information in your account across sessions and machines. The service is currently a free, controlled preview; features may change. By creating an account or using the service, you agree to these terms and the privacy commitments on this page. If you do not agree, do not use the service.
You retain your rights in your content. “User Content” means the memories, queries, conversation messages, documents, and other material that you or your connected agents submit, together with content-derived summaries, relationships, and indexes in your account.
You give us limited permission to host, store, back up, transmit, retrieve, organize, and otherwise process User Content only to provide the service you authorize, including search, recall, maps, export, deletion, and optional maintenance. This permission does not transfer ownership or give us a right to use your content for our own independent purposes. It ends when that processing is no longer needed, subject to the retention and legal requirements described below.
Producing a memory or summary for your own account at your direction is a service feature, not model training. The narrow access and disclosure exceptions in sections 5 and 6 remain necessary to operate a secure, lawful service.
Conversation capture is optional and requires your approval. Supported capture hooks send available visible prompts and final responses, with limited provenance and content-free gap records when text is unavailable or refused. They never read or upload hidden reasoning, system instructions, tool traces or incidental documents. A visible message can itself contain sensitive information; choose whether to enable capture accordingly.
The checklist of requests you type is off unless you enable it for your own account. Automatic Claude Code compaction capture and continuity injection are currently unavailable because native parent/child attribution is unreliable; retained source authorization cannot activate them. Old queued summaries remain local for review, and capture never reads transcripts. A setting applies only from when you change it and only to your account. Turning a source off stops new collection; what was kept stays until you delete it.
Documents are kept selectively: your AI proposes a document with its exact revision, provenance and reason, and either you approve each one or, if you choose, the connectors you authorize may keep what they select within your exclusions and daily limit. Files are never uploaded merely because an agent read or edited them, and you can inspect, pause, revoke and delete kept documents at any time.
We use essential session cookies and browser storage for sign-in, security, and interface preferences. We do not use advertising trackers.
The current hosted preview is not end-to-end encrypted. Server-side features must process readable content, so sufficiently privileged developers and operators can technically access User Content. Our no-routine-review commitment is a policy restriction on that access; it is not a claim that access is technically impossible.
We limit human access to the minimum reasonably necessary to fulfill your explicit support request; resolve a specific service failure when content access is essential; investigate or contain a security incident, fraud, or abuse; or meet a binding legal obligation. Access must be authorized for that purpose and limited in scope. Where feasible, we use content-free diagnostics and seek your permission before viewing content for support. General product development is not an exception.
We use account authentication, access controls, account separation, encrypted transport, and encrypted production storage to reduce the risk of unauthorized access. No online system can guarantee that an attack, outage, or data loss will never occur. Do not store passwords, API keys, tokens, private keys, or other secrets. Automated credential filters reduce accidental intake but cannot identify every secret.
The hosted service uses Amazon Web Services in the United States (us-west-2), including storage and backup infrastructure. Service providers may process information needed for hosting, delivery of service email, support, and security on our behalf. We limit that processing to those service purposes; it does not give them permission to exploit User Content independently. Optional sign-in providers receive the information needed to authenticate you and provide your account identity to us.
Agents and AI services you choose to connect receive the content returned to them on your behalf. Optional tasks such as Night Shift can send relevant content to your chosen agent and its model provider when you authorize those tasks. Their handling of that content is governed by your relationship and settings with them; our no-training commitment does not change another provider’s terms.
We disclose User Content only as needed for those authorized features and service providers, at your direction, or to the minimum extent required by law or necessary to address a specific security incident, fraud, or abuse. We do not grant a general commercial license to it.
You can inspect and export your memories, revoke connectors, and delete individual memories in the dashboard. Conversation retention is 30 days by default and can be changed in your settings; saved memories and documents can remain until you delete them, and documents also have an optional retention setting. Turning off capture stops new capture; it does not delete material already stored.
Individual memory deletion removes the memory and matching copies from active and archive storage and invalidates derived views. It does not selectively erase existing infrastructure backups. Backup copies are restricted to recovery and age out under the backup retention process; deletion is not instantaneous across every recovery copy.
You can request account deletion from the account menu. It pauses connected agents and schedules deletion after a seven-day grace period; you can cancel before the scheduled date. Purging runs in the subsequent maintenance cycle. Contact us if you need help or a deletion-status confirmation. We may retain the minimum information required by law or needed to document a security incident, and will restrict its use to that purpose. Copies already returned to your agents or exported by you are outside this deletion process.
Use the service lawfully and submit only information you have the right and authority to provide. Protect your sign-in and connector credentials, review the agents you connect, and revoke access you no longer need. A connected agent acts with the permissions you give it; review important changes and verify recalled information before relying on it. Do not probe, disrupt, bypass controls, or attempt to access another account’s data, and do not store credentials or secrets.
The current preview is free. Any future paid offering will disclose its price and payment terms before you choose it. We may change, suspend, or discontinue the service, or restrict access to address a terms violation or security risk. If we discontinue the service, we will give you reasonable notice and a way to export your data.
The service is provided “as is” and “as available,” without warranties of any kind, to the fullest extent permitted by law. To the same extent, our total liability for any claim arising from the service is limited to the amount you paid us for it in the twelve months before the claim (currently zero, while the preview is free). Nothing in these terms excludes rights or liability that cannot lawfully be excluded.
We will post changes here with a new effective date and provide notice of material changes through the service or email. Changes apply prospectively. We will not retroactively expand the permitted uses of previously collected User Content without your affirmative consent. Continued use after the effective date accepts other updated service terms, subject to rights that applicable law preserves.
Questions, access or correction requests, security reports, and help with export or deletion: support@vanillabrain.ai. We may need to verify your identity before acting on a request. Please do not send passwords, connector keys, or unnecessary private content.