Table of Contents

Configuration

Settings can be managed via CLI or by editing the JSON file directly.

dtk config show                          # display all current values
dtk config set <key> <value>             # update a value and save

See Usage Guide — dtk config for examples and valid values per key.

JSON file

DTK uses an optional JSON configuration file at:

# Default locations:
# - Windows: %APPDATA%/dtk/config.json
# - macOS:  ~/Library/Application\ Support/dtk/config.json
# - Linux:  ~/.config/dtk/config.json
Tip

Zero-config by default — DTK works out of the box without any configuration file. All settings have sensible defaults. You only need to create or edit the config file if you want to customize behavior.

A minimal config that only changes the display width would be:

{
  "display": {
    "width": 100
  }
}

Any omitted keys use their defaults.

Full Configuration File

{
  "tracking": {
    "enabled": true,
    "retentionDays": 90,
    "dbPath": null,
    "tokenizer": "Cl100kBase"
  },
  "display": {
    "colors": true,
    "emoji": true,
    "width": 120
  },
  "tee": {
    "mode": "failures",
    "directory": null,
    "maxFiles": 20,
    "maxFileSizeBytes": 1048576
  }
}

Tracking

Controls token usage tracking and persistence.

Key Default Description
enabled true Enable or disable token tracking
retentionDays 90 How many days of history to keep
dbPath null Custom SQLite path (defaults to %LOCALAPPDATA%/dtk/tracking.db on Windows, ~/.local/share/dtk/tracking.db on Linux/macOS)
tokenizer "Cl100kBase" Tokenizer model used for token counting (see Tokenizer Models)

Display

Controls terminal output formatting.

Key Default Description
colors true Enable colored terminal output
emoji true Enable emoji characters (✓, etc.)
width 120 Display width in characters

Tee Logs

Controls raw output logging to disk.

Key Default Description
mode "failures" "failures" saves only failed runs; "always" saves all runs; "never" disables tee logging
directory null Log directory (defaults to %LOCALAPPDATA%/dtk/tee)
maxFiles 20 Maximum log files to keep; oldest are deleted first
maxFileSizeBytes 1048576 Maximum size per log file (1 MB)

Tokenizer Models

The tokenizer field accepts one of the following values:

Value Encoding Typical Models
Cl100kBase cl100k_base GPT-4, GPT-3.5-turbo
O200kBase o200k_base GPT-4o, o1, o3

Token counting uses the Microsoft.ML.Tokenizers library, which provides accurate BPE tokenization matching the selected model's encoding.

Note

Which tokenizer should I pick? Claude, Gemini, and other non-OpenAI models use their own tokenizers, but no public BPE files are available for them. The Cl100kBase default provides a close-enough token estimate for any model and is the recommended choice unless you specifically need O200kBase for GPT-4o/o1/o3 accuracy.

Edit this page