---
title: Configuration Optimization
description: Optimizing Verdent for your workflow
---

Optimize Verdent for VS Code to balance speed, cost, and thoroughness based on your workflow requirements. Configure model presets, execution modes, and context management strategies for optimal performance.

### Optimization Areas

- Model preset selection for cost and performance balance
- Execution mode configuration for workflow efficiency
- Context management for large codebases
- Per-project and workspace-level customization

---

## Tuning Settings for Your Workflow

### Model Preset Selection

Choose the appropriate model preset based on task complexity and budget:

| Preset | Cost | Cost Efficiency | Best For |
|--------|------|----------------|----------|
| **Efficiency** | ~0.3x Balance (70% reduction) | 3.2x more efficient than Sonnet | Simple tasks, quick fixes, rapid prototyping |
| **Balance** (Default) | 1x baseline | 1x baseline | General development tasks, everyday coding |
| **Performance** | 1-2x Balance | 0.5x efficiency (2x cost, 1M context) | Complex architectural decisions, intricate refactoring |

**Additional Models:**

- **Claude-4.5-sonnet:** Balanced general-purpose model (1x baseline)
- **Claude-4.5-sonnet-1m:** Extended context for complex tasks (0.5x efficiency, 2x cost when input exceeds 200k tokens)
- **Claude-4.5-haiku:** Fastest, most economical (3.2x more efficient than Sonnet)
- **GPT-5 / GPT-5-codex:** Reasoning and creativity focus (Beta, 1.3x more efficient than Sonnet)
- **Minimax-m2:** Fast and cost-effective (Beta, free until November 10th)

### Execution Mode Configuration

<Tabs>
  <Tab title="Skip Permissions (Speed)">
    - Full autonomy without approval prompts
    - No interruptions for permissions
    - Highest risk - use only in automated environments
  </Tab>
  <Tab title="Auto-Run (Balanced)">
    - Commands require approval, file operations automatic
    - Reduced permission prompts
    - Lower risk with maintained safety for dangerous operations
  </Tab>
  <Tab title="Manual Accept (Thoroughness)">
    - Every operation requires approval
    - Complete visibility and control
    - Slowest but safest - best for learning and production code
  </Tab>
  <Tab title="Plan (Review)">
    - Strategic planning before execution
    - Read-only analysis phase
    - Review and approve before any changes
  </Tab>
</Tabs>

### Think Hard Mode

Enable for complex reasoning tasks:
- **Thoroughness:** Maximum reasoning depth for sophisticated problems
- **Speed:** Slower due to extended thinking budget
- **Best For:** Architectural decisions, complex debugging, intricate analysis

---

## Speed vs Thoroughness Tradeoffs

Balance execution speed against analysis depth based on task requirements.

<Tabs>
  <Tab title="Speed-Critical">
    **Configuration:**
    - **Model:** Efficiency preset (Claude-4.5-haiku)
    - **Permission Mode:** Auto-Run Mode
    - **Execution Mode:** Agent Mode for direct execution
    - **Use Cases:** Quick fixes, routine operations, simple tasks

    **Benefits:**
    - Fastest response times
    - Minimal interruptions
    - 70% cost reduction vs Balance preset
  </Tab>
  <Tab title="Thoroughness-Critical">
    **Configuration:**
    - **Model:** Performance preset (Claude-4.5-sonnet-1m)
    - **Permission Mode:** Manual Accept Mode
    - **Execution Mode:** Plan Mode for strategic review
    - **Think Hard:** Enabled for sophisticated reasoning
    - **Use Cases:** Complex features, production code, architectural decisions

    **Benefits:**
    - Maximum context window (1M tokens)
    - Complete visibility and control
    - Deep reasoning for complex problems
    - Review before execution
  </Tab>
  <Tab title="Hybrid Approach">
    Most users benefit from context-switching between configurations:

    - **Quick fixes:** Efficiency + Agent Mode
    - **Complex features:** Balance/Performance + Plan Mode
    - **Exploration:** Plan Mode for strategy, then Agent Mode for execution

    **Strategy:** Adapt configuration to task type instead of using one-size-fits-all approach.
  </Tab>
</Tabs>

<Tip>
Use Auto-Run Mode for trusted projects to reduce permission prompts while maintaining safety for dangerous operations. This streamlines workflow without sacrificing critical protections.
</Tip>

---

## Workspace-Level Configuration

Verdent supports per-project and per-workspace configuration for team-wide consistency.

### Project-Specific Configuration

**Project Rules (AGENTS.md):**

- **Location:** Project root directory or workspace folder
- **Scope:** Applied only to the current project
- **Version Control:** Commit to git for team-wide standards
- **Content:** Coding standards, testing requirements, architectural patterns

**Example:**

```markdown filename="AGENTS.md"
# Project Rules

## Code Standards
- Use TypeScript strict mode
- Generate unit tests for all new functions
- Follow MVC pattern

## Testing
- All endpoints require input validation
- Run tests before committing
```

### VS Code Workspace Settings

Configure extension settings at workspace level:

**Location:** `.vscode/settings.json` in workspace root

**Example:**

```json filename=".vscode/settings.json"
{
  "verdent.enableCheckpoints": true
}
```

### Configuration Priority

When configurations conflict, Verdent applies this priority order:

1. **Project Rules (AGENTS.md)** - Highest priority (project-specific)
2. **Workspace Settings** - VS Code workspace-level settings
3. **User Rules (VERDENT.md)** - Global user preferences
4. **Default Settings** - Verdent's built-in defaults

---

## Project-Specific Customization

### Context Management for Large Projects

<Tabs>
  <Tab title="Subagents">
    - Delegate complex operations to subagents with separate context windows
    - Only subagent results consume main context, not entire process
    - Prevents main context from filling with implementation details
  </Tab>
  <Tab title="File Selection">
    - Explicitly add only relevant files using `@filename`
    - Avoid loading entire project into context unnecessarily
    - Target specific components for focused analysis
  </Tab>
  <Tab title="Task Chunking">
    - Break large tasks into smaller, completable chunks
    - Start new sessions for distinct features or components
    - Natural breakpoints prevent context depletion
  </Tab>
  <Tab title="Plan Mode">
    - Review approach before consuming execution context
    - Reduces wasted context on incorrect implementations
    - Plan Mode has separate context management
  </Tab>
</Tabs>

### Performance Optimization

**Enable Checkpoints Selectively:**

The `verdent.enableCheckpoints` setting uses git for version control:
- May impact performance on very large repositories
- Enable only when checkpoint functionality is needed
- Disable for maximum performance on large codebases

<Tip>
Enable Checkpoints selectively for critical tasks only, excessive checkpointing slows workflows without adding value.
</Tip>

**Model Selection:**
- Use Efficiency preset (Haiku) for simple, isolated tasks
- Reserve Performance preset (Sonnet-1M) for context-heavy operations
- Balance preset for general work

**Execution Mode:**
- Auto-Run Mode reduces context consumed by permission prompts
- Skip Permissions Mode maximizes efficiency for automated environments

<Note>
Multi-workspace scenarios automatically apply appropriate project rules when switching workspaces. No manual configuration switching required.
</Note>

<Tip>
Multi-workspace configurations let you optimize settings per project type, development vs production, monorepo vs microservices.
</Tip>

---

## See Also

<CardGroup cols={2}>
  <Card title="Configuration Settings" icon="gear" href="/docs/verdent-for-vscode/configuration/settings">
    Complete configuration options and model selection
  </Card>
  <Card title="Resource Monitoring" icon="chart-line" href="/docs/verdent-for-vscode/resource-management/monitoring">
    Monitor performance and optimize token usage
  </Card>
</CardGroup>