Creating an agent takes about two minutes. You give it a name, a role, and instructions — Qorven handles the rest.

​
From the web dashboard

1

Open the Agents page

Go to Agents in the sidebar and click New Agent in the top right.
2

Give it a name and role

The name is how you and Prime will refer to this agent. The role description helps Prime know when to delegate work to it — for example: “Specialist in financial analysis and spreadsheet data. Ask me to review reports, analyse numbers, or summarise financial documents.”
3

Write the instructions

This is the most important step. Write in plain English — describe what this agent should do, how it should respond, what tone to use, and any specific rules it should follow. See writing good instructions → for tips.
4

Choose a model

Select which AI model powers this agent. You can leave this as the install default, or pick a specific model — for example, a reasoning model for analysis tasks or a fast model for quick responses.
5

Enable tools

Choose which tools this agent can use. A research agent needs web search and scraping. A developer agent needs file access and code execution. Limiting tools to what an agent actually needs is also a good security practice.
6

Save

Click Create Agent. Your new agent is immediately available — Prime can delegate to it, and you can talk to it directly by selecting it in the chat interface.

​
From the CLI

qorven agents create \
  --name "Support Agent" \
  --role "Handles customer support questions and looks up order information." \
  --model claude-opus-4-7 \
  --tools research,read_document,email
List all agents: 
qorven agents list
Edit an existing agent: 
qorven agents edit <agent-id>

​
Practical examples

Name: Morning Brief
Role: Delivers a morning summary of news, email, and tasks.
Instructions: “Every morning at 8am, compile a briefing for me covering: unread emails that need a response, any news in [my industry], and any tasks on my list for today. Keep it concise — bullet points, no fluff.”
Tools: email, research, memory_read
Schedule: Set a cron schedule for 8:00 AM in your timezone
Name: Support
Role: Answers product questions and helps customers with common issues.
Instructions: “You’re a friendly support agent for [Company]. Answer questions about our product using the information in your memory. For billing or account issues, tell the customer to email billing@company.com. Always be warm and concise.”
Tools: memory_search, read_document
Channel: Connect to your website webchat or a Telegram/WhatsApp number
Name: Reviewer
Role: Reviews pull requests and checks code quality.
Instructions: “When asked to review code, check for: bugs, security issues, unclear variable names, missing error handling, and anything that doesn’t follow our style guide. Be direct — point out problems but also say what’s good.”
Tools: filesystem, glob, grep, read_document

​
What makes a good agent

  • One clear job. An agent that does “everything” does nothing well. Give each agent a focused role.
  • Good instructions. The instructions are the most important part — they determine everything the agent does. Put real thought into them.
  • The right tools. Only enable tools the agent actually needs. This keeps the agent focused and limits what it can do by accident.
  • A useful role description. Prime uses the role description to decide when to call on this agent. Make it specific enough that Prime knows exactly when to delegate.

​
Where next

Write better instructions

How to write instructions that make agents work reliably.

Tools reference

Every available tool and what it does.

Scheduling

Run agents on a schedule — daily briefings, weekly reports, and more.

Channels

Connect your agent to Telegram, WhatsApp, Slack, and more.