GETTING STARTED

A live agent in five minutes

Five steps, start to finish. Everything is free during beta, no card required. By the end you will have a working assistant embedded in your product that can read your docs and call your tools.

Steps

Create a project

Sign in to the console at /admin and create a project. Each project has its own agent, its own key, and its own settings. Copy the project key, it looks like era_... and goes in the widget snippet below.

Add a model provider key

Open Settings and paste a key from Anthropic, OpenAI, or Google (Gemini). Agentifys never charges you for LLM tokens, you pay your provider directly. Keys are encrypted at rest, tried in priority order, and health-checked. Add more than one for automatic failover.

Paste the embed snippet

Drop this one script tag into your product's HTML, just before the closing </body>. Swap in your real project key. The widget is vanilla JS with zero dependencies and renders in a Shadow DOM, so it will not collide with your styles.

html

<script
  src="https://agentifys.ai/static/widget.js"
  data-api-key="era_your_project_key"
  data-title="Assistant"
  data-primary-color="#2F6FED"
  data-placeholder="Ask me anything…"
  data-position="bottom-right"
></script>

Set your allowed origins

This step is for the browser-direct install you just pasted. In Settings, under Identity & Access, add every origin your product is served from. A request whose origin is not listed is refused with ORIGIN_NOT_ALLOWED and a 403.

An entry is one origin: scheme, host, and port, no path and no trailing slash. The match is exact — no wildcards, no subdomain matching — so https://example.com and https://www.example.com are two different entries and you need both if you serve both. Type the scheme yourself: a value without one is saved as https://, which will not match a dev server running on http.

text

http://localhost:5173
https://www.example.com
https://app.example.com

If your backend proxies Agentifys instead of the browser calling it directly, leave this list empty. Server-side requests carry no origin header at all, and an absent origin never matches a non-empty list, so a single entry here refuses every call your server makes.

Say hi

Load the page. The launcher appears in the corner. Open it and ask a question. You will see the agent stream its answer, show live tool status as it works, and ask for approval before any write action. That is the whole loop.

A working model key is required. With no healthy key, chat is rejected with PROJECT_NO_API_KEY. Add at least one provider key in step 2 before you test.

Where to next

You have an agent that talks. Now make it act. Read the core concepts to see how identity, tools, and knowledge fit together, then follow the Connect tools guide to give it actions, or upload your docs so it answers from your own content.

PREVIOUSDocs home NEXTCore concepts