What Are Agents

Every conversation in Magus is powered by an agent. The agent is the AI “brain” behind the conversation: it reads your messages, decides how to respond, runs tools when needed, and remembers things across your chats.

The default agent

When you start a new conversation, it uses the default agent. The default agent is a capable, general-purpose assistant with a broad set of tools available. It works well for most everyday tasks: writing, research, coding help, answering questions, and more.

You do not need to think about agents at all if you just want to chat. The default agent handles everything automatically.

What makes an agent

An agent is defined by a few key things:

• Instructions: A system prompt that tells the AI how to behave, what tone to use, and what it should focus on.
• Tools: Capabilities the agent can use, such as searching the web, running code, or reading files.
• Integrations: Connections to external services, such as Telegram or Google Calendar.
• Model: The AI model the agent uses by default (or it can auto-select based on the task).
• Knowledge: Data sources and memories the agent can draw on.

Custom agents

Custom agents let you create specialized assistants tailored for specific purposes. Instead of a general helper, you might create:

• A code review agent that knows your conventions and can read your error logs.
• A writing assistant with specific style guidelines baked in.
• A research agent connected to RSS feeds and web search.
• A customer support agent integrated with your Telegram bot.

Custom agents save you from explaining the same context every time. The instructions, tools, and integrations are always there, ready to go.

Where to find agents

Visit your agents page to see all the agents you have created, browse examples, and create new ones. From there you can also set which agent a conversation uses.

Assigning an agent to a conversation

When creating a new conversation, or from the conversation settings panel, you can select which agent to use. The conversation will use that agent’s instructions, tools, and integrations for every message.

Mentioning agents with @

You can bring a custom agent into any conversation by typing @handle in your message, where “handle” is the agent’s unique handle. For example, if you have an agent with the handle researcher, typing @researcher find recent papers on BEAM concurrency will route that message directly to the researcher agent.

How it works:

• The mentioned agent receives the message in its own home conversation, processes it with its own instructions and model, and sends the response back to your current conversation.
• The main conversation agent does not reply to the message. Only the mentioned agent responds.
• You can mention multiple agents in one message. Each one will receive the message independently.
• If you mention an agent while already in that agent’s home conversation, the agent simply handles the message normally (no separate dispatch needed).