Commands Requiring sql.js The following commands require the optional sql.js package for SQLite database access: sync Downloads and stores the full Ollama model catalog into a local SQLite database. Run this before using search or smart-recommend.
llm-checker sync llm-checker sync --force # Force full sync llm-checker sync --incremental # Only sync new/updated models llm-checker sync --quiet # Suppress progress output
Flag Description -f, --forceForce full resync even if recent data exists --incrementalOnly sync new and updated models -q, --quietSuppress all progress output
smart-recommend Advanced recommendations using the full scoring engine and the local SQLite model database. Provides best, fastest, and highest-quality picks with score breakdowns.
llm-checker smart-recommend llm-checker smart-recommend --use-case reasoning llm-checker smart-recommend --use-case coding --limit 5 --target-tps 30
Flag Description -u, --use-caseOptimize for: general, coding, chat, reasoning, creative, fast -l, --limitMaximum number of recommendations (default: 5) --target-tpsTarget tokens per second (default: 20) --target-contextTarget context length (default: 8192) --include-visionInclude multimodal models --include-embeddingsInclude embedding models -j, --jsonOutput as JSON
Hardware Simulation simulate Simulates hardware profiles to show compatible LLM models for a different system, without needing the actual hardware. Supports both preset profiles and fully custom hardware configurations.
# List all available preset profiles llm-checker simulate --list # Use a preset profile llm-checker simulate --profile rtx4090 llm-checker simulate --profile m4pro24 --use-case coding # Custom hardware configuration llm-checker simulate --gpu "RTX 5060" --ram 32 --cpu "AMD Ryzen 7 5700X" llm-checker simulate --gpu "RTX 4090" --ram 64 llm-checker simulate --ram 16
Flag Description -p, --profile <name>Preset hardware profile to simulate (e.g., rtx4090, m4pro24, h100) -l, --listList all available hardware profiles --gpu <model>Custom GPU model (e.g., "RTX 5060", "RX 7800 XT", "Apple M4 Pro") --ram <gb>Custom RAM in GB --cpu <model>Custom CPU model --vram <gb>Override GPU VRAM in GB (auto-detected from GPU model if omitted, requires --gpu) -u, --use-case <case>Use case for scoring (default: general) --optimize <profile>Optimization profile (default: balanced) --limit <number>Number of models to show (default: 1) --no-verboseDisable progress output
Run llm-checker simulate --list to see all built-in profiles including laptop, desktop, workstation, and data center tiers.
Catalog & Discovery list-models Lists all models from the Ollama model database with filtering options. Does not require sql.js — uses the built-in curated catalog.
llm-checker list-models llm-checker list-models --category coding llm-checker list-models --popular llm-checker list-models --size small llm-checker list-models --json
Flag Description -c, --category <category>Filter by category: coding, talking, reading, reasoning, multimodal, creative, general -s, --size <size>Filter by size: small, medium, large, or specific (e.g., "7b", "13b") -p, --popularShow only popular models (>100k pulls) -r, --recentShow only recently updated models (last 30 days) --limit <number>Limit number of results (default: 50) --fullShow full details including variants and tags --jsonOutput as JSON
ollama Manage Ollama integration and check its availability.
llm-checker ollama llm-checker ollama --list llm-checker ollama --running llm-checker ollama --compatible llm-checker ollama --recommendations
Flag Description -l, --listList installed models with compatibility scores -r, --runningShow running models with performance data -c, --compatibleShow only hardware-compatible installed models --recommendationsShow installation recommendations
gpu-plan Multi-GPU placement advisor. Computes single-GPU and pooled VRAM envelopes, strategy recommendation, and ready-to-paste environment variables.
llm-checker gpu-plan llm-checker gpu-plan --model-size 14 # Validate a 14GB model llm-checker gpu-plan --json
Flag Description --model-sizeTarget model size in GB to validate against the plan -j, --jsonOutput plan as JSON
Output includes:
Detected GPU count and total VRAM/unified memory
Single-GPU safe model size envelope
Pooled (multi-GPU) safe model size envelope
Placement strategy with rationale
Recommended env vars (CUDA_VISIBLE_DEVICES, etc.)
verify-context Verifies the practical context window limit for a local Ollama model by combining the declared context from model metadata with a hardware memory budget estimate.
llm-checker verify-context llm-checker verify-context --model qwen2.5-coder:14b llm-checker verify-context --model llama3.2:3b --target 32768
Flag Description -m, --modelModel to verify (default: first installed model) -t, --targetTarget context tokens to validate against (default: 8192) -j, --jsonOutput as JSON
Output includes:
Declared context window from model metadata
Estimated memory-safe context limit
Recommended runtime context value
Pass/warn/fail status with per-check breakdown
amd-guard AMD/Windows reliability guard with actionable mitigation hints. Checks ROCm availability, detects common driver issues, and provides a fix list.
llm-checker amd-guard llm-checker amd-guard --json
Flag Description -j, --jsonOutput report as JSON
Output includes:
Platform and primary backend detection
ROCm availability and detection method
Per-check status (pass/warn/fail)
Actionable recommendations for AMD/Windows setups
Tool-calling compatibility tester. Sends a standardized add_numbers tool-calling prompt to local Ollama models and scores the response.
llm-checker toolcheck llm-checker toolcheck --model qwen2.5-coder:14b llm-checker toolcheck --all
Flag Description -m, --modelTest a specific model by name --allTest all installed models (default: tests first installed model) --timeoutPer-model timeout in ms (default: 45000) -j, --jsonOutput as JSON
Status levels: SUPPORTED, PARTIAL, UNSUPPORTEDMCP Setup mcp-setup Prints or applies the Claude Code MCP configuration for LLM Checker.
llm-checker mcp-setup llm-checker mcp-setup --apply # Run claude mcp add automatically llm-checker mcp-setup --npx # Use npx instead of global binary llm-checker mcp-setup --json # Output setup details as JSON
Flag Description --nameMCP server name in Claude (default: llm-checker) --npxUse npx llm-checker-mcp instead of global llm-checker-mcp --applyExecute claude mcp add ... automatically -j, --jsonOutput setup details as JSON
Enterprise Policy Commands policy init Generates a policy.yaml template for enterprise governance.
llm-checker policy init llm-checker policy init --file ./my-policy.yaml llm-checker policy init --file ./my-policy.yaml --force
Flag Description -f, --fileOutput path for the policy file (default: policy.yaml) --forceOverwrite an existing file
policy validate Validates a policy file against the v1 schema. Exits non-zero on schema errors.
llm-checker policy validate llm-checker policy validate --file ./my-policy.yaml llm-checker policy validate --json
Flag Description -f, --filePolicy file to validate (default: policy.yaml) -j, --jsonOutput validation result as JSON
Audit Export audit export Evaluates policy compliance against model candidates from check or recommend, then exports machine-readable reports in JSON, CSV, or SARIF format.
# Single format JSON report llm-checker audit export \ --policy ./policy.yaml \ --command check \ --format json \ --out ./reports/check-policy.json # All configured formats llm-checker audit export \ --policy ./policy.yaml \ --command check \ --format all \ --out-dir ./reports
Flag Description --policyRequired. Policy file path--commandEvaluation source: check or recommend (default: check) --formatReport format: json, csv, sarif, or all (default: json) --outOutput file path (single-format only) --out-dirOutput directory when --out is omitted (default: audit-reports) -u, --use-caseUse case for check mode (default: general) -c, --categoryCategory hint for recommend mode --optimizeOptimization profile for recommend mode --runtimeRuntime for check mode --include-cloudInclude cloud models in check-mode analysis --max-sizeMaximum model size filter --min-sizeMinimum model size filter -l, --limitModel analysis limit for check mode (default: 25) --no-verboseDisable verbose progress
Integration Examples CI artifact (JSON)
SIEM ingestion (CSV)
Security scanning (SARIF)
llm-checker audit export \ --policy ./policy.yaml \ --command check \ --format json \ --out ./reports/policy-report.json
GitHub Actions Policy Gate name : Policy Gate on : [ pull_request ] jobs : policy-gate : runs-on : ubuntu-latest steps : - uses : actions/checkout@v4 - uses : actions/setup-node@v4 with : node-version : 20 - run : npm ci - run : node bin/enhanced_cli.js check --policy ./policy.yaml --runtime ollama --no-verbose - if : always() run : node bin/enhanced_cli.js audit export --policy ./policy.yaml --command check --format all --runtime ollama --no-verbose --out-dir ./policy-reports - if : always() uses : actions/upload-artifact@v4 with : name : policy-audit-reports path : ./policy-reports
When --format all is used, the export honors the reporting.formats list in your policy.yaml. If that list is empty, it defaults to json, csv, and sarif.
