Skip to main content
Environment variables let you adjust LLM Checker behavior for the current shell session or a single command. They are useful for CI/CD pipelines, Docker containers, and one-off overrides without touching ~/.llm-checker.json.

​
Quick reference

VariableTypeDefaultDescription
LLM_CHECKER_NO_GPUbooleanfalseDisable GPU detection entirely
LLM_CHECKER_VRAM_GBnumberauto-detectedOverride detected VRAM in GB
LLM_CHECKER_RAM_GBnumberauto-detectedOverride detected system RAM in GB
OLLAMA_BASE_URLstringhttp://localhost:11434Ollama API base URL
LLM_CHECKER_LOG_LEVELstringinfoLog verbosity level
LLM_CHECKER_CACHE_DIRstring~/.llm-checkerDirectory for cache and benchmark data
NO_COLORanyunsetDisable all ANSI color output

​
Variable reference

​
LLM_CHECKER_NO_GPU
boolean
default:"false"
When set to true (or any truthy string), LLM Checker skips GPU enumeration and treats the system as CPU-only. Hardware scoring, VRAM budgets, and backend selection all fall back to CPU paths.Use this in unit tests and headless CI runners where GPU drivers are not installed.
export LLM_CHECKER_NO_GPU=true
​
LLM_CHECKER_VRAM_GB
number
Forces LLM Checker to use the specified VRAM amount (in GB) instead of the auto-detected value. Useful when the system GPU driver reports incorrect memory, or when you want to simulate a different GPU tier.
export LLM_CHECKER_VRAM_GB=8
Setting LLM_CHECKER_NO_GPU=true takes precedence — VRAM is ignored when GPU detection is disabled.
​
LLM_CHECKER_RAM_GB
number
Forces LLM Checker to use the specified system RAM amount (in GB) instead of the auto-detected value. Useful when auto-detection returns incorrect values or for testing different memory budget scenarios.
export LLM_CHECKER_RAM_GB=32
​
OLLAMA_BASE_URL
string
default:"http://localhost:11434"
Base URL for the Ollama HTTP API. Override this to point at a remote Ollama daemon, a different port, or an SSH tunnel.
export OLLAMA_BASE_URL=http://remote-server:11434
This affects all commands that query Ollama: check, recommend, ollama-plan, installed, and MCP tools.
​
LLM_CHECKER_LOG_LEVEL
string
default:"info"
Controls internal log verbosity. Accepted values in ascending order of detail:
  • error — only fatal errors
  • warn — warnings and errors
  • info — standard operational messages (default)
  • debug — verbose trace output including hardware detection steps, scoring internals, and policy evaluation details
export LLM_CHECKER_LOG_LEVEL=debug
​
LLM_CHECKER_CACHE_DIR
string
default:"~/.llm-checker"
Directory where LLM Checker stores the scraped Ollama model cache, benchmark results, and calibration artifacts. Override this if you want a shared cache across users, a tmpfs location, or a project-local directory.
export LLM_CHECKER_CACHE_DIR=/custom/cache/path
​
NO_COLOR
any
When set to any non-empty value, LLM Checker disables all ANSI escape codes and renders plain monochrome text. This is the standard NO_COLOR convention respected by most CLI tools.
export NO_COLOR=1
Set NO_COLOR=1 in CI pipelines where color codes break log parsers or produce garbled artifacts.

​
CI/CD usage

​
GitHub Actions

jobs:
  llm-check:
    runs-on: ubuntu-latest
    env:
      LLM_CHECKER_NO_GPU: "true"
      NO_COLOR: "1"
      LLM_CHECKER_LOG_LEVEL: "warn"
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20
      - run: npm install -g llm-checker
      - run: llm-checker check --no-verbose

​
Shell script with inline overrides

You can scope overrides to a single command without exporting to the shell session: 
# Override VRAM for this command only
LLM_CHECKER_VRAM_GB=16 llm-checker check

# Simulate CPU-only on a GPU machine
LLM_CHECKER_NO_GPU=true llm-checker check --use-case coding

# Point at a remote Ollama instance
OLLAMA_BASE_URL=http://192.168.1.100:11434 llm-checker recommend --category coding

​
Full debug session

Capture all diagnostic output to a file for support or troubleshooting: 
export LLM_CHECKER_LOG_LEVEL=debug
export DEBUG=1
llm-checker check --detailed 2>&1 | tee debug.log

​
RAM sensitivity test in CI

for ram in 8 16 32 64; do
  echo "=== ${ram} GB RAM ==="
  LLM_CHECKER_RAM_GB=$ram llm-checker check --no-verbose
done
LLM_CHECKER_NO_GPU and LLM_CHECKER_VRAM_GB override real hardware. Avoid committing these as permanent CI environment variables — use them only for specific test scenarios where you need deterministic hardware simulation.

Build docs developers (and LLMs) love

Get started for freeTalk to us