Creating Agents

Agent Builder walkthrough — configuration form covering model, tools, skills, prompts, memory, guardrails, and sharing

Creating Agents

The Agent Builder is a multi-step form that lets you configure every aspect of an agent — from model selection to guardrails and permissions.

Navigation

Agent Builder Steps

1. Basic Information

Name and description. Use a name that reflects the agent's role (e.g., "SPL Reviewer", "Data Quality Checker").

2. Model Configuration

Select the AI model (OpenAI, Anthropic, Google, xAI). Adjust temperature and max tokens. Model availability depends on your plan.

3. System Prompt

Define the agent's role, tone, constraints, and behavior. Be specific — describe what the agent does, how it should respond, and what it should never do. You can auto-generate a prompt from a description.

4. Capabilities

Toggle web search, tool access, and tool choice mode (auto, required, none). These control how the agent decides when to use tools.

5. Response Configuration

Set response format (text, JSON, markdown) and streaming behavior.

6. MCP Tools

Select which MCP tool configurations the agent can access — Splunk MCP, Exa, Regex for Splunk, GitHub, custom MCP servers. Attach only the tools the agent needs.

7. Agents as Tools

Enable agent handoff — the agent can transfer to specialized sub-agents during a conversation. Configure which agents are available as handoff targets.

8. Tool Optimization

Fine-tune how the agent selects and uses tools. Adjust tool selection strategy for precision vs. recall.

9. Skills

Open the Skills tab to attach modular capabilities permanently. Toggle skills on to load them for every conversation with this agent. Available skills include Search Splunk, Deep Research, Web Research, CIM Modeling, Incident Response, and more. See Agent Skills for descriptions and usage.

10. AI SDK Tools & Memory

Configure memory scope (conversation, session, persistent), history limit, and context window size. Enable persistent memory so the agent remembers across conversations.

11. Orchestration & Guardrails

Set max turns per conversation. Define input/output guardrails to constrain agent behavior. Add match-on patterns for routing.

12. Permissions

Define tool permissions as JSON — control which tools the agent can call and under what conditions.

13. Sharing Settings

Set visibility: private (only you), team (your organization), or public (all users).

Tips for Effective Configuration

  • Start from a template — pick a purpose-built agent or template as a starting point, then customize.
  • Narrow the tools — fewer tools = more focused behavior. Don't attach Splunk MCP to a research agent.
  • Iterate on the prompt — test with real questions, then refine constraints and examples in the system prompt.
  • Use guardrails — set max turns and output guardrails to prevent runaway conversations.
  • Test edge cases — ask ambiguous or adversarial questions to find gaps in the prompt.