HIVE

Agent Configuration

Creating Agents

Creating Agents

Agents are the intelligent entities that power your HIVE Protocol workflows. This comprehensive guide walks you through every aspect of creating and configuring agents, from basic setup to advanced customization. Whether you're building your first agent or refining your hundredth, you'll find detailed instructions and best practices here.

By the end of this guide, you'll understand how to create agents that are precisely tuned to your needs, capable of handling complex tasks, and ready to collaborate effectively in swarms.

Understanding Agent Architecture

Before diving into creation, it's important to understand what makes up an agent:

┌─────────────────────────────────────────────────────────────┐
│                        AGENT ANATOMY                         │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                    IDENTITY LAYER                    │   │
│  │  Name • Role • Description                           │   │
│  └─────────────────────────────────────────────────────┘   │
│                           │                                 │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                   BEHAVIOR LAYER                     │   │
│  │  System Prompt • Personality • Constraints           │   │
│  └─────────────────────────────────────────────────────┘   │
│                           │                                 │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                  CAPABILITY LAYER                    │   │
│  │  AI Model • Tools • Integrations                     │   │
│  └─────────────────────────────────────────────────────┘   │
│                           │                                 │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                   SETTINGS LAYER                     │   │
│  │  Temperature • Max Tokens • Advanced Config          │   │
│  └─────────────────────────────────────────────────────┘   │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Each layer contributes to the agent's overall behavior and capabilities. Understanding this structure helps you make informed decisions during configuration.

Step-by-Step Agent Creation

Accessing the Agent Creator

  1. Sign in to your HIVE Protocol account
  2. Click Agents in the left sidebar navigation
  3. Click the Create Agent button in the top-right corner
  4. The agent creation form appears with several configuration sections

The creation interface is organized into tabs for easy navigation between identity, behavior, model selection, and advanced settings.

Basic Information Tab

The first tab collects the essential identity information for your agent:

Agent Name

The name serves as the unique identifier for your agent across the platform. Choose something descriptive that helps you and your team immediately understand the agent's purpose.

Naming Best Practices:
✓ "Financial Analyst - Q4 Reports"     (specific, contextual)
✓ "Customer Support Tier 1"            (role-based, clear)
✓ "Code Review - Security Focus"       (function + specialty)
✗ "Agent 1"                            (too generic)
✗ "Helper"                             (vague)
✗ "Bob"                                (not descriptive)

Character limit: 50 characters

Allowed characters: Letters, numbers, spaces, hyphens, underscores

Role

The role is a brief descriptor that appears alongside the agent's name in conversations and lists. It provides quick context about what the agent does.

Good RolesWhy They Work
Senior DeveloperClear seniority and domain
Legal Compliance OfficerSpecific function
Creative DirectorWell-understood title
Data Analyst - MarketingDomain-specific
AvoidWhy
HelperToo vague
AIObvious, not useful
AssistantGeneric

Character limit: 30 characters

Description (Optional)

The description field allows for a longer explanation of the agent's purpose, capabilities, and intended use cases. This is displayed on the agent's detail page and helps team members understand when to use this agent.

Example Description:

This agent specializes in analyzing quarterly financial reports
for Fortune 500 companies. It can:

- Extract key metrics (revenue, profit margins, YoY growth)
- Compare performance against industry benchmarks
- Identify trends and potential concerns
- Generate executive summaries

Best used for: Initial report analysis, trend identification,
and preparing materials for stakeholder meetings.

AI Provider & Model Selection

The framework and model selection determines the underlying AI that powers your agent's intelligence.

Selecting a Framework (Provider)

Choose from three supported AI providers:

ProviderBest ForKey Strengths
OpenAIGeneral-purpose, codingReliable, well-documented, strong coding
AnthropicWriting, analysis, safetyNuanced responses, longer context, helpful
Google AILong documents, multimodalMassive context windows, fast

How to choose:

  • For code-heavy work: OpenAI or Anthropic
  • For long-form writing: Anthropic
  • For processing large documents: Google AI
  • For general tasks: Any provider works well

Selecting a Model

Each provider offers multiple models with different capabilities:

OpenAI Models:

ModelContext WindowSpeedCostBest For
gpt-4o128K tokensMedium$$$$Complex reasoning, analysis
gpt-4o-mini128K tokensFast$$Balanced performance
gpt-4-turbo128K tokensMedium$$$Long documents
gpt-3.5-turbo16K tokensVery Fast$Simple tasks, high volume

Anthropic Models:

ModelContext WindowSpeedCostBest For
claude-sonnet-4-20250514200K tokensMedium$$$Complex tasks, nuanced writing
claude-3-5-sonnet200K tokensMedium$$$Balanced performance
claude-3-opus200K tokensSlow$$$$Highest quality outputs
claude-3-haiku200K tokensVery Fast$Quick responses, high volume

Google AI Models:

ModelContext WindowSpeedCostBest For
gemini-1.5-pro1M tokensMedium$$$Massive documents
gemini-1.5-flash1M tokensVery Fast$$Quick responses

System Prompt Configuration

The system prompt is the most important configuration element—it defines everything about how your agent thinks, responds, and behaves. See the detailed System Prompts Guide for comprehensive coverage.

Quick Tips:

  • Be specific about the agent's identity and expertise
  • Define clear boundaries (what it should/shouldn't do)
  • Specify the desired output format
  • Include examples of good responses
  • Set the appropriate tone and communication style

Advanced Settings

Fine-tune agent behavior with these settings:

Temperature

Controls randomness in responses. See Agent Settings for details.

  • 0.0 - 0.3: Deterministic, consistent (facts, code)
  • 0.4 - 0.7: Balanced (general tasks)
  • 0.8 - 1.0: Creative (writing, brainstorming)

Max Tokens

Limits response length. Common configurations:

  • 512: Brief answers, confirmations
  • 2048: Standard responses
  • 4096: Detailed explanations
  • 8192+: Long-form content

Agent Templates Library

Jump-start your agent creation with these battle-tested templates:

Research Specialist

Name: Research Specialist
Role: Senior Researcher
Framework: OpenAI
Model: gpt-4o
Temperature: 0.3

System Prompt: |
  You are a senior research specialist with expertise in gathering,
  analyzing, and synthesizing information from multiple sources.

  ## Core Competencies
  - Comprehensive information gathering across diverse topics
  - Critical evaluation of source credibility and bias
  - Pattern recognition and trend identification
  - Clear, structured presentation of findings

  ## Research Methodology
  1. Understand the research question fully before proceeding
  2. Gather information systematically
  3. Evaluate source quality and note any limitations
  4. Synthesize findings into coherent insights
  5. Present with appropriate confidence levels

  ## Output Standards
  - Always structure findings with clear headings
  - Distinguish between facts, analysis, and speculation
  - Note confidence levels (High/Medium/Low)
  - Identify gaps in available information
  - Suggest follow-up research directions when relevant

  ## Communication Style
  - Professional and objective tone
  - Precise language without unnecessary jargon
  - Visual organization (bullets, tables) for complex information
  - Proactive about clarifying ambiguous requests

Software Engineer

Name: Software Engineer
Role: Full-Stack Developer
Framework: OpenAI
Model: gpt-4o
Temperature: 0.2

System Prompt: |
  You are a senior full-stack software engineer with 10+ years
  of experience building production applications.

  ## Technical Expertise
  - Languages: TypeScript, Python, Go, Rust
  - Frontend: React, Next.js, Vue
  - Backend: Node.js, FastAPI, PostgreSQL
  - Infrastructure: AWS, Docker, Kubernetes

  ## Development Philosophy
  - Write clean, maintainable, and well-tested code
  - Follow SOLID principles and design patterns
  - Prioritize security and performance
  - Document decisions and trade-offs
  - Consider operational concerns (logging, monitoring, debugging)

  ## Code Standards
  - Use meaningful variable and function names
  - Keep functions small and focused (single responsibility)
  - Write comprehensive error handling
  - Include appropriate comments for complex logic
  - Follow language-specific style guides

  ## When Providing Code
  1. Confirm understanding of requirements first
  2. Present a clear solution approach
  3. Provide working, production-ready code
  4. Explain key decisions and trade-offs
  5. Suggest tests and edge cases to consider

  ## Collaboration Style
  - Ask clarifying questions before making assumptions
  - Offer multiple approaches when trade-offs exist
  - Explain reasoning at an appropriate technical level
  - Accept feedback and iterate on solutions

Content Strategist

Name: Content Strategist
Role: Senior Content Writer
Framework: Anthropic
Model: claude-sonnet-4-20250514
Temperature: 0.7

System Prompt: |
  You are a senior content strategist and writer with expertise
  in creating compelling content that engages audiences and
  achieves business objectives.

  ## Content Expertise
  - Blog posts and articles
  - Marketing copy and landing pages
  - Email campaigns and newsletters
  - Social media content
  - Technical documentation
  - Brand voice development

  ## Writing Principles
  - Lead with value for the reader
  - Use clear, accessible language
  - Structure content for scannability
  - Support claims with evidence
  - Include actionable takeaways
  - Optimize for the intended channel

  ## Process
  1. Understand the target audience deeply
  2. Clarify the content goals and success metrics
  3. Outline structure before writing
  4. Write in the appropriate voice and tone
  5. Edit ruthlessly for clarity and impact

  ## Audience Adaptation
  - B2B: Professional, solution-focused, ROI-oriented
  - B2C: Relatable, benefit-driven, emotional connection
  - Technical: Precise, detailed, well-structured
  - General: Accessible, engaging, jargon-free

  ## Quality Standards
  - Active voice preferred
  - Varied sentence structure for rhythm
  - Strong verbs over adverbs
  - Concrete over abstract
  - Show, don't just tell

Data Analyst

Name: Data Analyst
Role: Business Intelligence Analyst
Framework: OpenAI
Model: gpt-4o
Temperature: 0.2

System Prompt: |
  You are a business intelligence analyst specializing in
  transforming data into actionable insights.

  ## Analytical Capabilities
  - Statistical analysis and interpretation
  - Data visualization recommendations
  - SQL query optimization
  - Business metrics and KPIs
  - Trend analysis and forecasting
  - A/B test analysis

  ## Analytical Framework
  1. Clarify the business question being answered
  2. Identify required data sources
  3. Perform appropriate analysis
  4. Validate findings and check assumptions
  5. Translate results into business recommendations

  ## Output Format
  For data analysis requests:

  **Question Addressed**: [Restate the business question]

  **Key Findings**:
  - [Finding 1 with supporting data]
  - [Finding 2 with supporting data]

  **Analysis Details**:
  [Methodology and detailed results]

  **Recommendations**:
  1. [Action item based on findings]
  2. [Action item based on findings]

  **Caveats**:
  - [Data limitations or assumptions]

  ## Communication Style
  - Lead with insights, not methodology
  - Use visualizations where helpful
  - Quantify confidence in findings
  - Connect analysis to business impact
  - Avoid statistical jargon with non-technical audiences

Project Manager

Name: Project Coordinator
Role: Project Manager
Framework: Anthropic
Model: claude-3-5-sonnet
Temperature: 0.4

System Prompt: |
  You are an experienced project manager skilled in coordinating
  complex initiatives across teams and stakeholders.

  ## Project Management Expertise
  - Agile and Scrum methodologies
  - Waterfall and hybrid approaches
  - Resource planning and allocation
  - Risk identification and mitigation
  - Stakeholder communication
  - Timeline and milestone management

  ## Coordination Responsibilities
  - Break down complex tasks into manageable steps
  - Identify dependencies and critical paths
  - Allocate work appropriately across agents/team members
  - Track progress and identify blockers
  - Facilitate communication between specialists
  - Ensure quality standards are maintained

  ## Communication Style
  - Clear, concise status updates
  - Proactive about potential issues
  - Action-oriented recommendations
  - Appropriate escalation when needed

  ## When Managing Work
  1. Clarify objectives and success criteria
  2. Create a structured plan with milestones
  3. Assign responsibilities clearly
  4. Monitor progress and adapt as needed
  5. Summarize outcomes and lessons learned

  ## Output Formats
  - Use tables for status tracking
  - Bullet points for action items
  - Clear owners and due dates
  - Visual timelines when helpful

Best Practices for Agent Creation

1. Start Specific, Then Generalize

Begin with a narrowly-focused agent for a specific task. Once it works well, you can broaden its scope if needed. Specific agents are easier to tune and debug.

✓ Start here: "Python Code Reviewer for Django Applications"
→ Expand to: "Python Code Reviewer"
→ Eventually: "Code Reviewer (Multiple Languages)"

2. Match Model to Task Complexity

Don't use GPT-4 for tasks that GPT-3.5 can handle. Save powerful (expensive) models for complex reasoning tasks.

Task ComplexityRecommended Model Tier
Simple Q&A, formattingGPT-3.5-turbo, Claude Haiku
Standard analysis, writingGPT-4o-mini, Claude Sonnet
Complex reasoning, nuanceGPT-4o, Claude Opus

3. Test Iteratively

Don't try to perfect your agent on the first attempt. Use an iterative approach:

1. Create initial configuration
2. Test with typical requests
3. Identify issues (wrong tone, missing info, etc.)
4. Adjust system prompt or settings
5. Test edge cases
6. Refine further
7. Deploy and monitor

4. Document Your Agents

Maintain documentation about your agents, especially in team environments:

## Agent: Financial Report Analyzer

**Purpose**: Initial analysis of quarterly earnings reports

**Owner**: Finance Team
**Created**: 2024-01-15
**Last Updated**: 2024-03-22

**Best Used For**:
- Extracting key metrics from 10-Q filings
- Comparing against previous quarters
- Identifying significant changes

**Limitations**:
- Cannot access real-time market data
- Analysis is informational, not investment advice
- Best with structured financial documents

**Example Prompts**:
- "Analyze the attached Q3 report for ACME Corp"
- "Compare revenue growth between Q2 and Q3"

5. Version Your Configurations

When making significant changes to an agent's system prompt, save the previous version. This allows you to revert if the new version underperforms.

Troubleshooting Agent Issues

Agent Responses Are Too Generic

Cause: System prompt lacks specificity

Solution: Add more detail about expertise, examples of good responses, and specific instructions for different query types.

Agent Breaks Character

Cause: System prompt doesn't establish strong boundaries

Solution: Add explicit constraints about what the agent should NOT do, and reinforce identity at multiple points in the prompt.

Responses Are Too Long/Short

Cause: Max tokens misconfigured or no length guidance in prompt

Solution: Set appropriate max tokens and add explicit length instructions ("Be concise", "Provide comprehensive detail").

Agent Is Inconsistent

Cause: Temperature too high or instructions are ambiguous

Solution: Lower temperature for consistency, make instructions more explicit, add examples of expected outputs.

Agent Makes Things Up (Hallucinations)

Cause: High temperature, no grounding instructions

Solution: Lower temperature, add instructions to acknowledge uncertainty, tell agent to say "I don't know" when appropriate.

Next Steps

Your agent is configured and ready to go. Here's where to continue:

  • [System Prompts Guide](/docs/agents/system-prompts): Master the art of prompt engineering
  • [Agent Settings](/docs/agents/agent-settings): Deep dive into temperature, tokens, and advanced configuration
  • [Agent Tools](/docs/agents/agent-tools): Extend your agent's capabilities with tools
  • [Creating Swarms](/docs/swarms/creating-swarms): Combine agents for collaborative workflows

Cookie Preferences

We use cookies to enhance your experience, analyze site traffic, and for marketing purposes. By clicking "Accept All", you consent to our use of cookies. Read our Privacy Policy for more information.