Agent Mode

Deep reasoning with persistent memory

What is Agent Mode?

Agent Mode uses Claude Opus 4.5 for deep reasoning tasks. Unlike Normal mode, Agent mode:

  • Remembers across sessions — Persistent memory file stores context
  • Takes more time — Multi-step reasoning for complex analysis
  • Requires approval — Write operations need your confirmation (or trust the session)

When to Use Agent Mode

Use CaseWhy Agent Mode
Complex research synthesisNeeds deep reasoning across multiple nodes
Finding non-obvious connectionsBenefits from extended analysis
Multi-step knowledge organizationCan plan and execute autonomously
Building context over timeMemory persists between sessions

How to Enable

Toggle the mode selector in the Chat panel:

┌──────────────────────┐
│ ⚡ Normal │ 🔥 Agent │
└──────────────────────┘

Your choice persists across sessions.

Tool Approval

When Agent mode tries to create or modify data, you'll see an approval modal:

Options:

  • Approve — Allow this specific action
  • Reject — Block this action
  • ☑️ Trust Session — Skip future approvals this session

This prevents accidental modifications while allowing autonomous operation when you trust the agent.

Persistent Memory

Agent mode maintains memory across sessions in: ~/Library/Application Support/RA-H/agent-memory.md

What gets remembered:

  • Your preferences and patterns
  • Project context and goals
  • Important findings
  • Key decisions

View memory: Settings → Memory tab

Cost Considerations

Agent mode uses Claude Opus 4.5, which costs more per token than Normal mode (GPT-5 Mini). Use Normal mode for quick tasks and Agent mode when you need deeper reasoning.