Experimental Features

Preemptive Compaction Threshold

Option: preemptive_compaction_threshold

Default: 0.85

Type: number (0.5-0.95)

Threshold percentage to trigger preemptive compaction. The preemptive-compaction hook is enabled by default; this option customizes the threshold at which compaction is triggered.

How It Works

When context window usage reaches the specified threshold (default 85%), preemptive compaction automatically runs to reduce context size before hitting hard limits. This prevents errors and ensures smooth operation.

Configuration

{
  "experimental": {
    "preemptive_compaction_threshold": 0.85
  }
}

Recommended Values

• 0.75-0.80 - More aggressive, compacts earlier
• 0.85 - Default, balanced approach
• 0.90-0.95 - Less aggressive, compacts later

Truncate All Tool Outputs

Option: truncate_all_tool_outputs

Default: false

Type: boolean

Truncates ALL tool outputs instead of just whitelisted tools (Grep, Glob, LSP, AST-grep). The tool output truncator is enabled by default - disable via disabled_hooks.

How It Works

By default, only outputs from specific tools (Grep, Glob, LSP, AST-grep) are truncated. When this option is enabled, outputs from all tools are truncated when they exceed size limits.

Configuration

{
  "experimental": {
    "truncate_all_tool_outputs": true
  }
}

Aggressive Truncation

Option: aggressive_truncation

Default: false

Type: boolean

When token limit is exceeded, aggressively truncates tool outputs to fit within limits. More aggressive than the default truncation behavior. Falls back to summarize/revert if insufficient.

How It Works

When enabled, this feature uses more aggressive truncation strategies when the token limit is exceeded. If truncation alone isn't sufficient, it will attempt to summarize content or revert changes.

Configuration

{
  "experimental": {
    "aggressive_truncation": true
  }
}

Auto Resume

Option: auto_resume

Default: false

Type: boolean

Automatically resumes session after successful recovery from thinking block errors or thinking disabled violations. Extracts the last user message and continues.

How It Works

When a session encounters a thinking block error or thinking disabled violation, and recovery is successful, this feature automatically extracts the last user message and continues the session without manual intervention.

Configuration

{
  "experimental": {
    "auto_resume": true
  }
}

Use Cases

• Long-running sessions that may encounter thinking block errors
• Automated workflows where manual intervention isn't possible
• Scenarios where session continuity is critical

DCP for Compaction

Option: dcp_for_compaction

Default: false

Type: boolean

Enable DCP (Dynamic Context Pruning) for compaction - runs first when token limit exceeded. Prunes duplicate tool calls and old tool outputs before running compaction.

How It Works

DCP (Dynamic Context Pruning) analyzes the context and removes duplicate tool calls and old tool outputs before running standard compaction. This can be more efficient than compaction alone.

Configuration

{
  "experimental": {
    "dcp_for_compaction": true
  }
}

Benefits

• More efficient context reduction
• Preserves important information better
• Faster than standard compaction in some cases

Complete Configuration Example

Here's an example configuration with all experimental features:

{
  "experimental": {
    "preemptive_compaction_threshold": 0.85,
    "truncate_all_tool_outputs": false,
    "aggressive_truncation": false,
    "auto_resume": false,
    "dcp_for_compaction": false
  }
}

Configuration Locations

• ~/.config/opencode/oh-my-opencode.json - User-wide
• .opencode/oh-my-opencode.json - Project-specific

Learn More About Configuration →

Best Practices

• Start with defaults - Most users don't need experimental features. Only enable if you have specific requirements.
• Enable one at a time - Test each feature individually to understand its impact.
• Monitor behavior - Watch for unexpected behavior when using experimental features.
• Report issues - If you encounter problems, report them on GitHub.
• Stay updated - Experimental features may change between versions.

Feature Status

Experimental features may:

• Be promoted to stable features in future versions
• Be modified or removed based on feedback
• Have breaking changes without notice
• Have limited documentation

Check the GitHub repository for the latest status of experimental features.