1Who We Are
Qunivex is a platform for building and deploying AI agents that site owners embed on their own websites with a single <script> tag. We started from a simple frustration: every business wanted a capable AI assistant on their site, but getting one live meant wiring up models, prompts, retrieval, tools, and a secure backend — weeks of plumbing before the first real conversation.
So we built the plumbing once, and made it something you configure instead of code. Today Qunivex lets anyone stand up an agent with a custom prompt, a knowledge base, tools, sub-agent delegation, and content guardrails — and put it in front of real visitors in minutes, not months.
In short: Qunivex turns the hard parts of shipping an AI agent — secure model access, retrieval, tool execution, and moderation — into a few settings and one embed snippet.
2Our Mission
To make production-grade AI agents accessible to everyone — not just teams with an ML department. We believe the value of an AI assistant shouldn't depend on how much infrastructure you can afford to build and maintain around it.
That means an agent that is fast to create, safe by default, and honest about what it is: a tool that helps your visitors, backed by a platform that guards your data and your credentials as carefully as you would yourself.
3What We Build
Qunivex is a complete toolkit for the full life of an agent — from the first prompt to a deployed widget handling live traffic:
- Agents & projects. Configure a model, a system prompt, and a personality, then organise your work into projects with their own knowledge and settings.
- Knowledge base. Upload documents or describe your website, and your agent answers grounded in that content through semantic retrieval — no hallucinated facts about your business.
- Tools & integrations. Give your agent built-in abilities (web search, scraping, and more), author your own custom tools, or connect remote MCP servers.
- Sub-agents & workflows. Delegate specialised work to sub-agents and design the whole execution pipeline visually on a drag-and-connect workflow canvas.
- Guardrails. Add content moderation that can warn, block, or reroute a conversation at any point in the pipeline.
- One-tag deployment. Paste a single script tag on your site and your agent is live — gated to the exact domains you allow.
4How Qunivex Works
From building to going live, the flow is deliberately short:
- Create a project and configure your agent — its model, system prompt, and behaviour.
- Add knowledge, tools, sub-agents, and guardrails as your use case needs them.
- Test the agent privately in the Playground, where your own conversations never mix with real visitor traffic.
- Turn on the widget, add your domain to the allowlist, and paste the embed snippet into your site.
- Your visitors chat with the agent, and every model call happens server-side — the browser never sees a provider key.
5What We Believe
A few convictions shape every decision we make:
- Security is a design constraint, not a feature. No AI provider API key ever reaches a visitor's browser. The only client-side credential is a regenerable public key scoped to a single agent.
- Your data stays yours. Knowledge-base uploads are only ever used to answer questions inside the project they were uploaded to, and we don't use your conversations to train models.
- Isolation by default. Any code you author — custom tools and guardrails — runs in a locked-down sandbox with no host access, so one tenant can never reach another's data or the platform's own credentials.
- Honest about AI. Model output can be wrong, and we build the product so you can moderate, ground, and verify it rather than blindly trust it.
- Fast, always. An agent should be minutes of setup, and a conversation should stream back token by token — speed is part of the experience, not an optimisation we defer.
6Who It's For
Qunivex is built for anyone who wants an AI agent on their site without becoming an infrastructure team to get there:
- Businesses & support teams that want an assistant which actually knows their product, docs, and policies.
- Founders & makers shipping fast, who need something live today rather than a project for next quarter.
- Agencies & freelancers deploying agents across many client sites, each with its own knowledge and domain rules.
- Developers who want the depth — custom tools, MCP servers, workflow graphs, sandboxed code — without rebuilding the secure backend around it.
7The Technology
Under the hood, Qunivex assembles each request server-side — your system prompt, relevant knowledge-base context, conversation history, and tool definitions — and runs it through a model, streaming the response back to the Playground or widget. Model inference is served through DeepInfra, giving you access to the best open and frontier models behind one consistent abstraction, so you can switch models without rewriting your agent.
Retrieval, tool execution, sub-agent delegation, and content moderation are all orchestrated server-side, with every outbound request from tenant-controlled code routed through a hardened, SSRF-guarded egress layer. The result is a platform that stays fast and safe as your agent grows from a single prompt into a full workflow.
8Get in Touch
Questions, feedback, or want to tell us what you're building? We'd love to hear from you.