1Overview
Qunivex is a visual studio for building AI agents: you configure a model, a system prompt, a knowledge base, tools, sub-agents and guardrails, then deploy a single embeddable chat widget with one <script> tag. No provider API key ever reaches the browser — every model call runs server-side through Qunivex.
This page is a short map of the concepts. For the fuller story behind the platform, see About Qunivex.
2Getting Started
- Create an account and start a new project — a project holds one main agent plus everything attached to it.
- Set your agent's model and system prompt.
- Try it out in the Playground before anything goes live — Playground conversations never count against your public chat quota.
- Open Deployments, turn the widget on, add your site's domain, and paste the embed snippet before
</body>.
3Agents & Prompts
Each project has a main agent with its own model, temperature and system prompt. Pick from a catalog of open and frontier models — Llama, Qwen, DeepSeek, Mistral and more — served through a single fast inference layer, so switching models never means rewriting your prompt or tools.
4Knowledge Base
Upload documents (PDF, DOCX, HTML, TXT) or describe your website, and your agent grounds its answers in that content through semantic retrieval. Sub-agents can carry their own separate knowledge base, scoped just to that specialist.
5Tools & MCP
Enable built-in tools (web search, web scraping, QR codes, current time), author your own Python or HTTP tools, or connect a remote MCP server. Custom code runs in a sandboxed, network-restricted environment, and every outbound request is SSRF-guarded.
6Sub-Agents
A lead agent can delegate to specialist sub-agents — each with its own model, prompt, tools and knowledge base — for tasks like sales, support or operations. Delegation is wired visually on the workflow canvas.
7Workflow & Guardrails
The Workflow page is a drag-and-connect graph editor: wire together your agent, tools, sub-agents, guardrails and conditional branches exactly the way you want a conversation to flow. Guardrails can block, warn, or route a message on policy — before or after the model runs.
8Deploying the Widget
Deployments generates a unique, domain-locked <script> tag for your agent. Paste it into any page and the widget loads itself — no build step, no npm install, no backend changes on your site. The only client-side credential is a public, regenerable agent key.
9Getting Help
Logged-in users can open the in-app Help panel (the ? button in the dashboard header) for quick answers, or reach out directly on the Contact page.