What You'll Build
By the end of this guide, you'll have:
- An AI agent running in a secure sandbox
- Real deliverables produced from your first conversation
- Optionally: an automated workflow that runs on schedule
Total time: ~5 minutes. No code. No credit card.
Step 1: Pick Your Agent (30 seconds)
Go to the Agent Marketplace and browse 90+ agents across 19 verticals.
Not sure where to start? Try one of these:
| If you're in... | Try this agent | What it does |
|---|---|---|
| Sales | SDR Agent | Research leads, score against ICP, write outreach |
| Marketing | SEO Content Strategist | Keyword analysis, content brief, on-page audit |
| Legal | Contract Reviewer | Read contracts, flag risks, generate redlines |
| Finance | CFO Agent | Cash flow forecast, financial models, board reports |
| HR | Recruiter | Screen resumes, score candidates, design interviews |
Click Try Agent on any agent card to start.
Step 2: Start a Conversation (1 minute)
The agent opens in a chat window — just like ChatGPT, but with real tools.
Type your first message. Examples:
- SDR Agent: "Research leads in the fintech space with 50-200 employees in the US"
- Contract Reviewer: "Review the sample contract and flag any risk clauses"
- SEO Writer: "Analyze the keyword 'ai agent platform' and create a content brief"
- CFO Agent: "Build a 13-week cash flow forecast from the sample data"
The agent will start working — you'll see tool calls (Read, Bash, WebSearch) as it processes your request.
Step 3: Watch It Work (2 minutes)
Unlike a chatbot, BeaverStudio agents use tools:
- Read: Opens files in its workspace (leads.csv, contracts, financial data)
- Bash: Runs commands, processes data, generates reports
- WebSearch: Searches the internet for real-time information
- Write: Creates deliverables (reports, emails, analyses)
Each agent has a seed workspace — pre-loaded with domain-specific skills, data templates, and methodology. The SDR agent has lead qualification skills and ICP scoring. The contract reviewer has clause analysis and risk assessment skills.
You'll see the agent's work in real-time: tool calls stream as they happen, and the final response includes the deliverable.
Step 4: Get Your Deliverable (30 seconds)
The agent produces real output:
- SDR Agent: A scored lead list with personalized outreach emails
- Contract Reviewer: A risk assessment with flagged clauses and redline suggestions
- SEO Writer: A structured content brief with keyword targets and outline
- CFO Agent: A cash flow forecast with scenario analysis
This isn't a summary — it's a file you can use. The agent writes it to its workspace, and you can download or copy it.
Step 5 (Optional): Save as Automated Workflow
If the agent did something you want repeated:
- Open the Daily Monitor (from the desktop or Start Menu)
- In the agent sidebar, describe what you want automated: "Track competitor pricing changes weekly"
- The system records the tool calls (trace)
- Type
/minion graduate weeklyto save it as a scheduled workflow - The workflow runs automatically and shows results in your Daily Monitor
Cost per run: ~$0.0003. The workflow uses a lightweight LLM interpret step instead of full agent reasoning.
What's Next?
- Build a custom agent: Use the Agent Builder to create an agent for your specific use case
- Deploy a team: Browse Agent Teams for multi-agent setups (Sales Machine, Content Factory, etc.)
- Set up your command center: The Daily Monitor shows recurring agent output as a real-time dashboard
- Explore all verticals: 19 industry verticals, each with specialized agents and skills
FAQ
Do I need an API key? No. Demo agents run on our infrastructure. For production use, you'll subscribe to a plan.
Is my data secure? Every session runs in an isolated E2B sandbox — a disposable Linux container. Nothing persists between sessions unless you configure it.
Can I connect my own tools? Enterprise plans include custom integrations (Salesforce, HubSpot, Slack, Jira). The Agent Builder can create agents with custom tool definitions.
What models do the agents use? BeaverStudio routes to the optimal model per task: Qwen for fast operations, Claude for complex reasoning, GPT for specific capabilities. You don't need to choose — the router handles it.
Ready? Browse agents and try one now — no signup required.