Privacy Policy
This page describes what we actually do with your data — not what a lawyer wishes we did. It's short because our data practices are short. Effective date: August 30, 2026.
Who we are
Nexus AgentWorks operates managed business AI agents — voice support, email operations, scheduling, lead qualification, invoice follow-up, and document collection — on behalf of small businesses. Questions about anything on this page go to [email protected] and a human answers.
What we collect
We collect as little as the service needs to function:
- Sign-in profile. When you log in with Google or GitHub — or with a single-use e-mail confirmation code when your team invites you without a provider account — we receive your email, username, display name, and your provider account ID. We use these to identify your account. We never post anywhere or touch your Google/GitHub data beyond the login profile.
- Team invites. When a workspace owner invites a teammate, we hold the invited e-mail address directly (no sign-in provider in between) until the invite is accepted or revoked. Confirmation codes, and failed sign-in attempts including code attempts, are logged in the same audit trail as every other authentication event.
- Your knowledge-base documents. Files you upload so your agents can answer questions from them. These are stored per-tenant in isolated Cloudflare Durable Objects and KV storage. Current stage capacity is 100 MB per tenant.
- Agent operation logs. What each agent did, decided, and why — the same audit trail you can see, which we use to operate and debug your service.
- Usage and billing records. Metered rows (cost, AI-answer counts) in a hash-chained ledger, so your invoices can be traced to actual usage.
- Support email. If you email us, we keep the thread.
Why we use it
Three reasons, and that's the whole list:
- To run your agents. Answer your customers' questions from your knowledge base, within your tenant. (Contractual necessity — it's the service you're paying for.)
- To bill accurately. Meter usage against the pricing in your agreement. (Contractual necessity.)
- To keep the service safe. Abuse prevention and security. (Our legitimate interest — and yours.)
We do not use your data to train AI models — ours or anyone else's. Nothing from one customer's workspace is used to serve another customer.
What we don't do
- We don't sell personal data. Not to anyone, ever. There is no "partners network," no data broker arrangement, no exception.
- We don't run ad trackers. No advertising trackers, no data brokers, no cross-site profiling, and no analytics inside the product. The one third-party script on this marketing site is Cloudflare Web Analytics (static.cloudflareinsights.com), which Cloudflare Pages injects into every page: it is cookieless, sets no identifier, does not follow you across sites, and reports aggregate page views only.
- We don't contact your prospects. Our agents are structurally incapable of deciding on their own to contact a person. No outbound contact happens until a real lead exists in your pipeline — that's a hard rule in our code, not a suggestion.
- We don't mix customer data. Tenant isolation is architectural, not a policy promise: each customer's workspace is its own isolated unit. Your knowledge base, logs, and memory are not shared with, or reachable from, any other customer's agents.
AI processing and third parties
To generate answers, your customers' questions may be sent to third-party language-model providers — OpenRouter or OpenAI, or your own provider key if you bring your own model (BYOM). Your knowledge-base documents are used for retrieval-grounded answers within your tenant only. We don't send your data anywhere else, and we don't select providers that train on customer payloads.
Sub-processors. Depending on which services your workspace uses, customer data may be processed by the following sub-processors, each only for the purpose listed:
- OpenRouter — language-model inference for chat answers and scheduled agent runs
- Anthropic — language-model inference for chat answers and agent tool use
- OpenAI — realtime voice calls (speech-to-speech) and text-to-speech
- Google — language models, speech recognition, text-to-speech, and realtime voice
- Cartesia — text-to-speech voice synthesis
- ElevenLabs — text-to-speech voice synthesis
- Amazon Web Services (Amazon Polly) — text-to-speech voice synthesis
- Telnyx — telephony: phone numbers and voice calls
- Resend — transactional email delivery (tickets, reports, replies)
- Exa — web search results retrieval for the workspace web.search tool; receives the search query string only, no knowledge-base content, no customer payloads
Cloudflare hosts the site and the workspace backend (edge serving, storage, and databases). This list is derived from our provider registry: when we add or retire a provider, this page changes with it. Voice calls are transcribed to text for your workspace; call audio is not stored by us.
Human oversight
Our agents escalate to your team's humans whenever they're uncertain. They don't make fully-unattended money or legal commitments. Where an agent's output reaches someone outside your team, a human sign-off sits in between.
How long we keep things
Account data, knowledge-base documents, and agent logs are kept while your account is active. Pending team invites expire after 7 days and are deleted for good the moment they're accepted or revoked. Usage and billing ledger rows are kept as long as we need them for accounting and tax purposes. When you ask us to delete your data, we delete it — after a 7-day pause in which you can take the request back, and immediately if you tell us not to wait. The only survivor is what we're legally required to keep for billing records.
One thing outlives deletion, and we'd rather name it than surprise you. The free pilot is one per business. If you confirm a phone number to start a pilot, we keep a one-way fingerprint of that number — a keyed hash, from which the number cannot be recovered — together with a count of the free weeks used and the internal ids of the workspaces that used them. We keep that indefinitely, including after you delete your account, because a record that a deletion erases could not do its job. It holds no phone number, no name and no email address, and we use it for nothing else: not marketing, not analytics, not profiling. It is also what lets us greet you as a returning customer instead of a stranger if you come back. Ask us and we'll tell you what that record says about a number you can confirm.
Export and deletion
Self-serve, from your workspace (Billing → Your data): Download my account data gives you your sign-in profile, your tickets and notifications, and your documents as one JSON file; Download workspace data (workspace owners) gives you the knowledge base with its text, bookings, agent configurations, evolution history, usage and the member roster — the same bundle our nightly backup keeps, so what you can download is exactly what we hold. Connected-service credentials are never in an export. Delete my account on the same panel asks you to type DELETE first, and then does two things at once: your workspace stops working for you immediately — your phone number stops answering, your agents stop running, nothing goes out on your behalf — and everything is scheduled for deletion 7 days later. Nothing is destroyed during those 7 days. Sign in again before the deadline and Restore my workspace puts it all back as it was; Delete now on the same page skips the wait if you meant it. When the 7 days are up, or the moment you choose Delete now, the deletion is real deletion, not a flag.
If you'd rather not use the buttons, email [email protected] from your account address and we will do it by hand.
Your rights
Wherever you live, you can ask us for access to your data, a copy of it, a correction, or deletion — the export and deletion section above is how. If you're in the EU/EEA or UK, the legal bases we rely on are listed in "Why we use it" above, and you can object or complain to your local regulator. If you're a California resident: we do not sell or share personal information as those terms are defined by the CCPA, so there's nothing to opt out of. We don't discriminate against anyone for exercising their rights.
Security
Per-tenant isolation in isolated storage, no cross-tenant data paths, and an append-only hash-chained ledger so usage records can't be quietly rewritten. If something ever goes wrong with your data, we tell you promptly and straight.
Changes to this policy
If we change what we do with your data, we'll update this page and change the effective date above. We won't sneak a new data practice in under an old policy.
Contact
Privacy questions, export requests, deletion requests: [email protected]. A human answers.
Related: Trust & Operating Principles · Terms of Service