Creating your first agent

Agents are AI assistants you configure inside Contextium. Each agent has a name, a role description, and a system prompt that tells it how to behave. Once created, an agent can be used directly via your AI tool (Claude Desktop, VS Code, Cursor) or shared with your team.

Persona vs custom: which path to take?

When you create a new agent, Contextium gives you two starting points.

Use a Persona if you want a head start. Personas are pre-built templates — things like a "Technical Writer", "Customer Support Agent", or "Code Reviewer" — with a system prompt already written. They're a great starting point you can customise further, and they're much faster than starting from a blank page. If you have a clear role in mind and want to get going quickly, pick the closest persona to what you need.

Go Custom if none of the personas match, or if you have a very specific use case in mind. You write the system prompt yourself from scratch. This gives you complete control over how the agent thinks and responds. It takes a little more effort up front but is the right choice for specialised roles or when you want a precise tone and approach.

Not sure? Start with a persona. You can always overwrite the system prompt entirely after you've seen what it looks like.

Step-by-step: creating an agent on the web

Starting with a persona

1. Go to Dashboard → Agents.
2. Click New Agent.
3. You'll see a grid of persona cards — browse the options and click the one that best matches the role you want.
4. Give the agent a name (something clear and recognisable, like "Support Agent" or "API Docs Writer").
5. Optionally add a short description — this appears in the agents list and helps your team understand what the agent is for.
6. Review the pre-filled system prompt. You can edit it now or come back to it later.
7. Click Create Agent.

Your agent is created and ready to use. You can attach skills to it at any time — see Attaching skills to an agent.

Starting with a custom agent

1. Go to Dashboard → Agents.
2. Click New Agent.
3. Scroll past the persona cards and click Custom (or a similar blank-slate option).
4. Enter a name and optional description.
5. Write your system prompt in the editor. This is where you define the agent's role, behaviour, and knowledge.
6. Click Create Agent.

What makes a good system prompt?

The system prompt is the most important part of your agent. It tells the AI who it is, what it knows, and how it should respond. Here are the principles that make a big difference.

Be specific about the role. Instead of "You are a helpful assistant," write "You are a customer support agent for Acme SaaS. You help users troubleshoot billing issues, account problems, and feature questions." The more concrete you are, the more consistently the agent stays in character.

Set the tone. Include a sentence about how the agent should communicate. "Be concise and professional" and "Be friendly and use plain language" produce very different agents. Match the tone to your audience.

Define boundaries. Tell the agent what it should not do. "If you don't know the answer, say so clearly and suggest the user contacts the support team" is more useful than leaving it to guess. Explicit limits help avoid confident-sounding wrong answers.

Give examples where it helps. If the agent will produce structured outputs — like a formatted report, a ticket summary, or a code review — include a short example of what the ideal output looks like.

Keep it focused. Agents that do one thing well are more useful than agents that try to do everything. If you find yourself adding many unrelated responsibilities to one agent, consider splitting it into two.

A working example of a solid system prompt:

You are a technical writer for the Contextium docs team. You help engineers turn rough notes and code into clear, structured documentation. Write in plain English, avoid jargon, and use short sentences. Format output as Markdown with headings and bullet points. If the user's input is too vague to work with, ask one clarifying question before proceeding.