1Overview
Qunivex ("Qunivex", "we", "us") operates a platform for building and deploying AI agents that site owners embed on their own websites via a single <script> tag. This policy explains what data we collect, why, and what happens to it — across both sides of the product: the dashboard you use to build agents, and the chat widget your visitors talk to once an agent is deployed.
In short: we never let an AI provider API key reach a visitor's browser, chat messages are sent server-side to the model providers needed to answer them, and knowledge-base uploads are only ever used to answer questions inside the project they were uploaded to.
This summary is for orientation only — the sections below are the operative policy.
2Scope & Definitions
This policy applies to two distinct groups of people, and it's worth being precise about which one you are:
- Customers — people with a Qunivex account who create Projects and Agents, upload knowledge-base content, and deploy the widget. If you're logged into the dashboard, this is you.
- Visitors — people who chat with an Agent embedded on a Customer's website, without ever creating a Qunivex account.
A few terms used throughout:
- Agent — a configured AI assistant (model, prompt, tools, knowledge base) that a Customer builds and deploys.
- Widget — the embeddable chat interface (
widget.js) a Customer places on their own site. - Playground — the private, logged-in testing interface inside the dashboard where a Customer tries out their own Agent before deploying it.
3Information We Collect
3.1 Account information (Customers)
When you register, we collect your name, email address, date of birth, and a password (stored as a salted hash — we never store or can recover your plaintext password). Date of birth is collected solely to verify you meet our minimum age requirement (see Children's Privacy) and isn't used for any other purpose. If you upgrade a plan, payment is handled by RazorPay; we receive confirmation of payment and plan status, not your full card/bank details.
3.2 Project & agent configuration
Everything you build: Project and Agent settings, system prompts, enabled tools, custom tool/guardrail source code, sub-agent delegation graphs, workflow graphs, and appearance/branding settings for your widget. This is stored against your account and is never used to answer another Customer's questions.
3.3 Chat conversations
We log conversations that pass through an Agent — both your own test conversations in the Playground and the conversations your Agent has with real Visitors through the deployed widget — as a source: test or source: public tagged record, so the two never get mixed into the same usage counters or analytics. A logged conversation includes the messages exchanged, any tool calls made, and — best-effort — an error record if something failed. Logging failures never block a chat from completing.
3.4 Knowledge base documents
Files you upload to a Project's knowledge base (PDF, DOCX, HTML, plain text, and similar) are extracted to text, split into chunks, and embedded for semantic search so your Agent can answer questions grounded in that content. The extracted text is stored in a per-project search index; the retrieved chunks are inserted into the Agent's context only when relevant to a Visitor's or your own question.
3.5 Automatically collected data
- IP address — used transiently for abuse/rate-limiting on both the dashboard and public chat endpoints, and to validate that widget requests originate from a domain you've allowed. We do not maintain a persistent IP-to-person history log for marketing purposes.
- Usage counters — number of test/public conversations, knowledge-base storage used, tool/guardrail counts — tracked against your plan's limits.
- Basic request metadata — origin/referer of widget requests (to enforce your domain allowlist), timestamps, and general error diagnostics.
We do not run third-party advertising or analytics trackers (e.g. ad pixels, cross-site analytics SDKs) anywhere on Qunivex or in the widget.
5How We Use Information
- To operate the service: authenticate you, run your Agents, execute tools, retrieve knowledge-base content, and stream responses.
- To enforce plan limits and prevent abuse (rate limiting, domain allowlisting, sandboxed code execution).
- To bill you for paid plans and manage upgrades/downgrades.
- To provide support and investigate reported bugs or incidents — this is the only situation where we'd look at logged conversation content that isn't ours to begin with, and only in connection with the Customer account it belongs to.
- To maintain and improve the platform (e.g. diagnosing errors from best-effort logs).
We do not sell personal information, and we do not use Customer or Visitor conversation content to train any foundation model — Qunivex calls third-party model APIs (below); it does not train or fine-tune models itself.
6How Chat Data Reaches AI Models
No AI provider API key is ever exposed to a Customer's dashboard session or a Visitor's browser — every model call is made server-side, by Qunivex, on your Agent's behalf. Concretely, when a message is sent (from the Playground or the widget):
- Qunivex assembles the request server-side: your Agent's system prompt, relevant knowledge-base chunks, conversation history, and any tool definitions.
- That assembled request — which includes the message text — is sent to the model provider configured for your Agent to generate a response.
- If your Agent has tools enabled (e.g. web search), the tool's query text is sent to that tool's provider only when the model decides to call it.
- The response streams back through Qunivex to the Playground or widget; it is not cached by the model provider for Qunivex's benefit beyond serving that single request.
See the next section for exactly which providers are involved and for what.
7Sub-Processors & Third Parties
We rely on the following third-party services to operate Qunivex. Each only receives the category of data it needs to do its job:
| Service | Role | What it receives |
|---|---|---|
| DeepInfra | AI model inference | The full assembled chat request (prompt, history, retrieved knowledge-base context, tool definitions) for every message sent through an Agent, to generate the response. |
| OpenAI | Text embeddings | Knowledge-base document text at upload time (and query text at retrieval time), used only to generate the vector embeddings that power semantic search. OpenAI does not generate your Agents' chat responses. |
| Tavily | Web search tool | The search query text, only when an Agent with the web-search tool enabled decides to use it. |
| RazorPay | Payments | Billing details needed to process a plan upgrade (Qunivex does not store full card numbers). |
| Infrastructure providers | Hosting & storage | Database and application hosting for all data described in this policy, under standard hosting-provider access controls. |
Each provider is bound by its own terms and processes data only to deliver the service listed above — not for its own advertising purposes. If we add or change a sub-processor in a way that materially changes how your data is handled, we'll update this page (see Changes to This Policy).
9Data Retention & Deletion
Account data, project/agent configuration, chat logs, and knowledge-base content are retained for as long as your account and the associated Project/Agent exist, so the product keeps working (conversation history, analytics, and RAG retrieval all depend on it).
- Deleting a Project, Agent, document, or conversation from the dashboard removes it from active use; the account cleanup tools let you do this yourself for content counted against your plan's limits.
- Deleting your account removes your login credentials and, in the ordinary course, the Projects and Agents tied to it. Residual copies may persist briefly in backups before being purged on our normal backup rotation.
- To request deletion of data you can't remove yourself — including Visitor conversation data you believe identifies you — contact us using the details in Contact Us.
10Security Measures
Security is a design constraint on this platform, not an afterthought. Measures that directly protect the data described in this policy include:
- No client-side keys. Provider API keys never reach a browser — the widget's only client-side credential is a regenerable public key scoped to one Agent.
- Domain-gated widget access. Public chat requests are validated against the origin/referer of the domains a Customer has explicitly allowed for their Agent.
- Sandboxed tenant code. Customer-authored custom tools and guardrails run in an isolated, read-only container with no host filesystem access, dropped privileges, and hard resource limits — never in the main application process.
- SSRF-hardened outbound requests. Any server-side fetch of a user-supplied URL (custom HTTP tools, web scraping) validates that every resolved address is public and pins the connection to the validated address, closing the door on DNS-rebinding attacks against internal infrastructure.
- Encryption in transit. Dashboard and API traffic is served over HTTPS in production, and the session cookie is marked Secure/HttpOnly.
- Fail-closed content moderation. If the code sandbox is ever unavailable, Python-based guardrails fail closed (block) rather than silently skipping moderation.
No system is perfectly secure, and we can't guarantee absolute security — but the architecture is built so that a compromise of one Customer's Agent or tenant code cannot reach another Customer's data or the platform's own credentials.
11Your Rights & Choices
Depending on where you're located, you may have rights to access, correct, export, or delete your personal data, restrict or object to certain processing, and withdraw consent where processing relies on it (for example, under the GDPR if you're in the EU/UK, the DPDP Act 2023 if you're in India, or state privacy laws such as the CCPA/CPRA if you're in the US).
- Customers can access and export most of their own data directly from the dashboard (Profile, Projects, knowledge base, logs), and delete it via account cleanup or by contacting us.
- Visitors without a Qunivex account can exercise these rights by contacting us directly (see Contact Us) or the website owner they chatted with, who may be better placed to identify which conversation is theirs.
We'll respond to verifiable requests within the timeframe required by applicable law.
12Notice for Website Visitors
If you're chatting with a Qunivex-powered widget on a website, that website's owner — not Qunivex — is generally the party responsible for how your conversation is used (for example, whether it's reviewed to improve support). This mirrors the notice shown directly in the chat widget: "This chat may be reviewed by the website owner to improve responses and provide support. Please don't share sensitive information."
In data-protection terms, the website owner (our Customer) is typically the controller of your conversation data, and Qunivex acts as their processor — we process it to run the Agent on their behalf, under their instructions, and don't independently use it for our own purposes beyond providing and securing the service. If you have a concern about a specific conversation, the fastest path is usually the website you were chatting on; we're also reachable directly (see Contact Us) and will route or act on legitimate requests either way.
13Children's Privacy
Qunivex is not directed to children. You must be at least 18 years old to create a Qunivex account — we collect and check your date of birth at registration to enforce this, and it applies regardless of the age of majority in your jurisdiction. If we ever learn an account was created by someone under 18 despite this check, we'll close it and delete the associated personal information.
Because Agents can be embedded on any website a Customer controls, we can't control who a given Agent is deployed to speak with — Customers deploying agents on sites likely to be used by children are responsible for their own compliance with applicable children's privacy laws (e.g. COPPA). If you believe a child has provided us with personal information — through account registration or otherwise — contact us and we'll remove it.
14International Transfers
Qunivex and the sub-processors listed above may process and store data in countries other than the one you're in, including countries that may have different data protection laws. Where required, we rely on appropriate safeguards (such as standard contractual clauses or an equivalent lawful transfer mechanism) for these transfers.
15Changes to This Policy
We may update this policy as the product changes. Material changes will update the "Last updated" date at the top of this page, and where a change meaningfully affects how we handle your data, we'll make reasonable efforts to give Customers more direct notice (e.g. by email or an in-app notice). Continued use of Qunivex after a change takes effect means you accept the updated policy.
16Contact Us
Questions, requests, or concerns about this policy or your data — reach out and we'll get back to you.