Anvella — Terms of Service
Last updated: 2026-10-05
1. What this service is
Anvella is an AI intake and operating console: a web-based control panel, an AI agent that answers your messages, and a set of tools it can use on your behalf (reading files you point it at, browsing the web, controlling desktop applications, and connecting to other services you configure) — all gated behind explicit permission you grant, one capability at a time.
It can be run two ways, and this document covers both:
- Self-hosted. You install and run Anvella on your own machine. You are the operator and the only user.
- Operator-installed. Someone else (the "operator") runs an instance and hands you a login code to sign in as a customer of that instance. Your data is kept in your own workspace, separate from the operator's and from any other customer's — see the Privacy Policy for what that isolation actually means.
Whichever way you reached this page, "the service" below means whichever Anvella instance you are signed into.
2. Accounts and access
- Operator access is the console token created when Anvella is installed. It reaches the machine's settings, model configuration, connected accounts, and file browser.
- Customer access is a login code, issued by an operator. A code signs you into your own workspace and nothing beyond it — not the operator's settings, not another customer's data, not the file browser.
- Login codes are stored hashed (PBKDF2). The operator sees a code once, when issuing or reissuing it, and cannot look it up in the console afterwards. Additional copies can exist: encrypted under the account password for an email-and-password account (so the code can be recovered with the password, until it is reissued), encrypted for a remembered device (opened by the
anvella_devicecookie), and in the email that sends you your code, which can stay in the sending mailbox's sent items. Reissuing a code makes the old one stop working. If you lose one, the operator can issue you a new one. On The Eye, an account with an email address can also get a new code itself after confirming a code sent to that mailbox, unless keys are saved under its current code (a new code would drop them, so that case still goes to the operator). Either way you get a new code — see the Privacy Policy for how codes are stored. - You're responsible for keeping your login code and any PIN you set to yourself. Anything done through your account is treated as done by you.
3. Acceptable use
You may use this service for your own lawful work. You may not:
- Share, sell, or publish your login code, or use it to let someone else impersonate you.
- Use the service to access accounts, systems, or data you're not authorized to use.
- Direct the assistant to send messages, submit forms, make purchases, or take any other action you would not be permitted to take yourself.
- Attempt to bypass, disable, or interfere with the permission and confirmation prompts described below — they exist so that consequential actions require your explicit say-so, and working around them defeats the point.
- Use the service to generate or distribute content that is unlawful, infringing, or that violates a third party's rights.
What the assistant does on your behalf, you remain responsible for. It is a tool you are directing, not an independent actor.
4. Browser and desktop control
Two of the assistant's capabilities are turning on a real browser session or a real desktop-automation session under your account. Both are off by default and are never turned on without you explicitly enabling them through the console. When you turn either on, the console shows you — every time, not just the first time — a plain-language explanation of exactly what that capability does and does not permit before it takes effect. Turning either off takes effect immediately, with no confirmation step.
Independent of that toggle, every individual action the assistant attempts through the browser or the desktop is checked against a policy before it runs: reachable sites and applications are limited to ones you've explicitly allowed; actions with real consequences — sending something, spending money, deleting something, or changing a saved record — stop and ask you to approve that specific instance, unless you've separately issued a bounded, expiring grant covering exactly that kind of action. Approving one instance never authorizes the next.
Your responsibility, stated plainly. You choose the sites and applications on the allowlist, and you approve the consequential actions. The assistant acts on third-party sites and applications at your direction and on your behalf; Anvella makes no warranty about outcomes on systems it does not operate, and is not liable for what those systems do with an action you approved. Keep the allowlist narrow, read the approval prompts, and turn the capability off when you are not using it. Language-model output — including decisions about which action to take — can be wrong: verify anything consequential before it runs, and do not grant standing approvals for actions you would not perform yourself.
5. Third-party AI providers
Anvella does not run its own language model. Depending on how the operator has configured this instance, a message you send may be routed to one or more of the following third-party providers for a reply: Groq, Cloudflare (Workers AI), Mistral, DeepSeek, Google Gemini, OpenAI, Anthropic (Claude), OpenRouter, Z.ai, Cerebras, or OpenCode Zen (a gateway that itself routes to several of the foregoing). Only the providers the operator has actually configured and enabled are ever used — an instance with no Claude key configured, for example, never sends anything to Anthropic.
Where an account uses its OWN subscription through a locally-authenticated command-line tool, requests go to that subscription's provider under the same rule.
Whichever provider handles a given turn receives the content of that turn (and relevant conversation history) as its input, and processes it under that provider's own terms and privacy policy, not this one. Anvella does not control what a third-party provider does with a request once it's been sent to them, and this document cannot make promises on their behalf. See the Privacy Policy for more on what is and isn't sent.
6. Your content
You keep ownership of what you send to and store in Anvella — notes, messages, files you point it at, and anything the assistant produces for you. You're responsible for having the rights to anything you provide to it, including anything routed to a third-party provider under Section 5.
7. Disclaimers
The service is provided "as is," without warranty of any kind, express or implied, including warranties of merchantability, fitness for a particular purpose, or non-infringement. Nobody involved warrants that the service will be uninterrupted, error-free, or that any AI-generated output is accurate, complete, or safe to act on without review.
This service is not certified as HIPAA, PCI-DSS, SOC 2, GDPR, or ISO 27001 compliant. Running software with encryption at rest is not the same thing as regulatory compliance. If you intend to handle regulated data through this service, that assessment is yours to make.
The assistant can take actions with real consequences when you permit it to. The confirmation prompts described in Section 4 exist for a reason and are not decoration — you are responsible for reviewing what you approve.
8. Limitation of liability
To the maximum extent permitted by law, and subject to any consumer-protection law that cannot be excluded (including liability for death, personal injury, or fraud, which is never excluded here), liability for anything arising from your use of this service is limited to direct damages, and excludes lost profits, lost business, lost data, or indirect or consequential loss. Nothing in this document is intended to remove any statutory right you have as a consumer that cannot lawfully be waived.
9. Termination
You may stop using the service and, if you have a login code, ask the operator to revoke it, at any time. An operator may suspend or terminate a login code for a material breach of these terms — for example, sharing it — and should tell you why when they do.
Ending your access does not, by itself, delete data already stored in your workspace. See the Privacy Policy for how deletion works.
10. Changes to these terms
This document may be updated as the service changes. The "Last updated" date at the top reflects the most recent revision. Continued use after a change is posted means you accept the updated terms; if a change is material, the operator should give you a more direct heads-up than a changed date.
11. Contact
Questions about these terms: through the address given to you at sign-up or point of sale, or however the operator running this instance has told you to reach them.
12. Arbitration agreement
ANY DISPUTE RESOLUTION. Except for claims subject to small-claims court jurisdiction, any dispute arising out of or relating to these Terms, or any breach thereof, shall be resolved by binding arbitration administered by JAMS (or such other arbitrator as the parties may agree) under its Comprehensive Arbitration Rules and Procedures, excluding any right to have a jury trial. The arbitration shall be conducted by a single neutral arbitrator. The arbitrator's award judgment may be confirmed in any court having jurisdiction. The parties may agree to mediate before arbitrating. Each party shall bear its own costs and attorneys' fees unless the arbitrator awards otherwise. The Federal Arbitration Act (FAA) governs the interpretation and enforcement of this arbitration provision. THIS ARBITRATION PROVISION WAIVES YOUR RIGHT TO PARTICIPATE IN A CLASS ACTION OR CLASS-WIDE ARBITRATION. BY USING THE SERVICE, YOU IRREVOCABLY WAIVE THE RIGHT TO HAVE A COURT OR JURY HEAR ANY CLAIM AS A CLASS REPRESENTATIVE OR IN A PRIVATE ATTORNEY GENERAL ACTION.
13. Governing law
These Terms and any separate agreement whereby we provide you Services shall be governed by and construed in accordance with the laws of South Carolina, United States, without regard to its conflict of law principles. Any dispute arising out of or relating to the Services shall be brought exclusively in the federal or state courts located in Richland County, South Carolina.
14. Severability
If any provision of these Terms is found to be unenforceable or invalid, such provision will be limited or eliminated to the minimum extent necessary so that the remaining provisions of these Terms will continue in full force and effect.
15. Entire agreement
These Terms, including the Arbitration Agreement (Section 12) and the Limitation of Liability (Section 8), constitute the entire agreement between you and Anvella AI with respect to the use of the Service, and supersede any and all prior or contemporaneous understandings or agreements, whether written or oral, relating to such subject matter.