Quickstart
Build, test, and ship your first agent in five minutes
This walkthrough takes a brand-new account from zero to a working, deployed agent. You'll build a simple research assistant that takes a topic and returns a structured briefing.
Don't have an account yet? Sign up at fruxon.com — it takes about a minute. The free tier covers everything in this guide.
Before you start
You'll need:
- A Fruxon organization (create one)
- An API key from at least one LLM provider (OpenAI, Anthropic, or Google work fine — the cheapest model is plenty for the demo)
Connect an AI provider
Open Settings → AI Providers in the sidebar (or jump straight in via Studio later). Paste a key from your provider's dashboard. The key is encrypted at rest and never leaves your organization.
You can connect more than one provider, and you can mix providers within a single agent — pick a fast cheap model for routing and a strong model for the final answer.
Create your first agent
Click Agents → Create Agent. Give it:
- Name — Research Briefing
- Type — Other (type is just a label for organization)
- Description — Takes a topic and returns a structured briefing.
Hit Create. You land in Studio — the visual canvas where every agent lives.
Wire up the workflow
Every agent has the same shape: an Entry Point (inputs), one or more Agent Steps (the work), and an Exit Point (outputs).
- Entry Point — Click it and add one parameter:
- Name:
topic - Type:
string - Required: ✓
- Name:
- Agent Step — Drop one onto the canvas and configure it:
- Name:
Briefing - AI Provider: pick the one you connected
- System Prompt:
You are a research analyst. Write concise, factual briefings. - User Prompt:
Write a one-page briefing on: {{input.topic}} Include: background, key players, recent developments, why it matters. Use short paragraphs. No fluff.
- Name:
- Exit Point — Map the output:
briefing←{{step.briefing.output}}
Fruxon auto-connects nodes by reading the placeholders — {{input.topic}} ties the step to the entry, {{step.briefing.output}} ties the exit to the step. There's nothing else to wire.
Test it
Click Test in the top toolbar. Fill topic with something — "the EU AI Act" — and run.
You'll see:
- The model's response, streamed live
- A full trace showing prompts, tokens, latency, cost
- Any error context if something went wrong
Iterate freely. Every save creates a revision — your history is never lost.
Deploy
Happy with it? Click Deploy and pick the revision to make live. The deployed revision is what your API calls and connectors will hit. To roll back, deploy a previous revision — there's no downtime.
Call it from anywhere
Your agent is now reachable four ways:
- REST API —
POST /v1/{tenant}/agents/{agent}:executewith an org-level API key. See the API reference. - Python SDK —
pip install fruxon. See the SDK guide. - CLI — same install ships
fruxon run,fruxon chat, and friends. See the CLI guide. - Connectors — Attach Slack, Teams, Telegram, or WhatsApp from the Connectors panel and the agent answers messages directly. See Connectors.
Where to go next
Add tools
Let your agent call APIs, query databases, run code, and use sub-agents.
Add knowledge
Upload files or connect Confluence, GitHub, web sources for retrieval.
Test rigorously
Run candidate revisions against golden datasets before you deploy.
Watch it run
Traces, costs, errors, and per-step breakdowns for every execution.
Got stuck? See the troubleshooting guide, or restart the in-app product tour from the Help menu.