Creating Your First Agent

Learn the basics of creating and using AI agents in Sciorex.

Difficulty: Beginner Time: 5 minutes

What is an Agent?

An agent is a configured AI assistant with:

• A specific personality (system prompt)
• Defined capabilities (allowed tools)
• Optional output format (schema)
• Configurable thinking depth and effort level

Think of agents as specialized team members - each one has a role and expertise.

Step 1: Open the Agents Panel

TBD: Replace with screenshot of the agents panel in sidebar

1. Click Agents in the sidebar
2. Click + New Agent

Step 2: Basic Configuration

Fill in the basic fields:

Field	Value
Name	Hello World Agent
Description	A friendly agent that helps with basic tasks

Step 3: Write the System Prompt

The system prompt defines your agent's personality and instructions:

systemPrompt: |
  You are a helpful, friendly assistant. You:

  - Answer questions clearly and concisely
  - Ask for clarification when needed
  - Provide examples when helpful
  - Admit when you don't know something

  Keep responses brief unless asked for detail.

TIP

Good system prompts are specific about:

• What the agent should do
• How it should communicate
• What it should avoid

Step 4: Choose a Model

Select the Claude model:

Model	Best For
claude-sonnet-5-0	General tasks, good balance of speed and quality (default)
claude-opus-4-6	Complex reasoning, creative tasks
claude-haiku-4-5-20251001	Simple tasks, fastest responses

For your first agent, start with Sonnet.

Step 5: Configure Thinking & Effort

Thinking Level

Controls how deeply the agent reasons before responding:

Level	Description
off	No extended thinking (fastest)
think	Basic reasoning for simple tasks
think-hard	Deep analysis for complex problems
think-harder	Deeper analysis with more token budget
ultrathink	Maximum reasoning depth

Effort Level (Opus 4.6 only)

Controls overall token spend, tool calls, and thinking depth:

Level	Description
low	Minimal effort, fast responses
medium	Balanced effort
high	High effort (default)
max	Maximum effort and tool usage

For your first agent, thinkingLevel: off keeps things simple.

Step 6: Save and Test

1. Click Save
2. Go to Chat
3. Select your Hello World Agent from the agent dropdown
4. Send a message:

Hi! Can you explain what you can help me with?

The Complete Agent File

Here's what your agent looks like as YAML:

name: Hello World Agent
description: A friendly agent that helps with basic tasks

systemPrompt: |
  You are a helpful, friendly assistant. You:

  - Answer questions clearly and concisely
  - Ask for clarification when needed
  - Provide examples when helpful
  - Admit when you don't know something

  Keep responses brief unless asked for detail.

model: claude-sonnet-5-0
adapter: claude-code
thinkingLevel: off

Agent Definition Reference

Here's the full set of fields available on an agent:

Field	Description
name	Display name
description	What the agent does
systemPrompt	Instructions and personality
model	Claude model ID
adapter	Provider adapter (e.g., claude-code)
thinkingLevel	Reasoning depth (off, think, think-hard, think-harder, ultrathink)
effortLevel	Effort level for Opus 4.6 (low, medium, high, max)
temperature	Response randomness (0-1)
maxTokens	Max response tokens
allowedTools	Tool permissions (array of ToolPermission)
mcpServers	MCP servers to enable
autoApproveTools	Tools that skip confirmation
requiresHumanApproval	Tools that always need confirmation
inputSchema	Expected input structure (JSON Schema)
outputSchema	Expected output structure (JSON Schema)
labels	Tags for organizing agents

What's Next?

Now that you've created your first agent, try:

1. Add tools: Give your agent access to files with allowedTools
2. Enable thinking: Set thinkingLevel: think for better reasoning
3. Create a specialist: Make an agent for code review, writing, or research
4. Add MCP servers: Enable sciorex-tickets or sciorex-interactions
5. Set effort level: Use effortLevel: max with Opus for deep analysis

Related

• Agents Reference - Full agent configuration options
• Models Reference - Choose the right Claude model
• Tool Permissions - Control what agents can do
• Agent with Custom Tools - Configure tool permissions
• Code Review Agent - A practical example
• Research Pipeline - Build a multi-agent flow