Skip to main content
LLM Checker supports a JSON configuration file at ~/.llm-checker.json that overrides built-in defaults for every invocation. No flags required — set your preferences once and they apply globally.

​
File location

Create or edit the file at: 
~/.llm-checker.json
The file is read on startup. If it is absent or contains invalid JSON, LLM Checker falls back to built-in defaults and logs a warning.

​
Example configuration

The following example enables coding-optimized analysis with compact output and filters out models that are too large for typical developer machines: 
{
  "analysis": {
    "defaultUseCase": "code",
    "performanceTesting": true
  },
  "display": {
    "maxModelsPerTable": 15,
    "compactMode": true
  },
  "filters": {
    "minCompatibilityScore": 75,
    "excludeModels": ["very-large-model"]
  }
}

​
Full schema

​
analysis

Controls how LLM Checker scores and benchmarks models.
​
analysis.defaultUseCase
string
default:"general"
The use case applied when no --use-case flag is passed. Affects scoring weights across the Quality, Speed, Fit, and Context dimensions.Accepted values: general, coding, chat, reasoning, creative, fast.
​
analysis.performanceTesting
boolean
default:"false"
When true, LLM Checker runs an active benchmark against installed Ollama models instead of relying solely on hardware-estimated speed. Equivalent to passing --performance-test on every check call.Enable this only if you want real tok/s measurements rather than theoretical estimates.

​
display

Controls the appearance of the CLI output tables.
​
display.maxModelsPerTable
number
default:"10"
Maximum number of model rows shown in compatibility and recommendation tables. Increase this if you want to see a longer ranked list without passing --limit each time.
​
display.compactMode
boolean
default:"false"
When true, reduces table padding and omits secondary detail rows. Useful on narrow terminals or when piping output to a log file.

​
filters

Pre-filters the model pool before scoring and ranking.
​
filters.minCompatibilityScore
number
default:"0"
Minimum compatibility score (0–100) a model must achieve to appear in results. Models that fall below this threshold are silently dropped before the ranked list is printed.Setting this to 75 hides marginal models and surfaces only well-matched candidates.
​
filters.excludeModels
string[]
default:"[]"
An array of model name substrings to exclude from all results. Any model whose identifier contains one of these strings (case-insensitive) is dropped before scoring.Example: ["very-large-model", "uncensored"]

​
Minimal configuration

You do not need to include every key. LLM Checker merges your file with built-in defaults — only the keys you specify are overridden: 
{
  "display": {
    "compactMode": true
  }
}

​
Validation

LLM Checker does not currently expose a config validate command. To verify your file parses correctly, run: 
node -e "JSON.parse(require('fs').readFileSync(process.env.HOME + '/.llm-checker.json', 'utf8'))" && echo "Valid JSON"
On macOS and Linux, ~ resolves to $HOME. On Windows with WSL, use the WSL home directory (/home/<user>/.llm-checker.json) rather than the Windows user profile.

Build docs developers (and LLMs) love

Get started for freeTalk to us