Get Running in 5 Minutes
From a blank account to a deployed, observable agent — without writing a single line of code.
What you’ll build
By the end of this guide you’ll have:A Working Agent
Clear instructions, a defined scope, and at least one connected tool
A Tested Conversation
Five real message tests covering happy path, edge cases, and tool failures
A Live Deployment
Your agent accessible on Web, Telegram, or via API
Step 1 — Pick one job
The most common mistake is starting too broad.Rule of thumb: If your agent description contains the word “anything” or “everything,” narrow it down before writing a single instruction.
Step 2 — Create the agent
1
Open the Dashboard
Navigate to bothive.cloud/dashboard and click Bots in the sidebar.
2
Click New Bot
Hit New bot in the top right corner.
3
Choose your build path
4
Name it specifically
Good:
Billing Support Agent — Bad: My BotStep 3 — Write instructions
Your instructions are the single biggest factor in agent quality. Spend time here.Step 4 — Connect tools (only what’s needed)
Step 5 — Test before deploying
Run at least five test messages in the Playground before deploying anywhere.Normal Request
“Why did my invoice fail?” — expect a clear explanation and next step
Vague Request
“It didn’t work” — expect a useful clarifying question
Tool Needed
“Check my subscription” — expect correct tool selection with safe arguments
Tool Failure
Simulate billing lookup failure — expect graceful error message with retry path
Step 6 — Deploy
Start with one surface. Add more channels after you’ve confirmed the first one works.Step 7 — Open observability
After your first real run, open the trace.1
Open the Runs tab
Go to your bot → Runs to see all executions.
2
Click a run
Inspect: user message, context loaded, tools selected, latency, errors.
3
Use HiveMind Debug
If a run went wrong, click Debug with HiveMind to get an AI-powered root cause analysis.
4
Fix and redeploy
Update instructions or tool config based on what the trace reveals.
Launch-ready checklist
Before you go live
- Agent has one clear job with a defined audience
- Instructions include what to do, what tone to use, and what to refuse
- Every tool has a reason for being connected
- At least five test cases passed in the Playground
- Memory works across a multi-message conversation
- One channel deployed and tested end-to-end
- Observability shows meaningful traces
- Billing and usage limits are configured