frouter（frouter 搜索）

frouter — Free AI Model Router CLI

frouter is a terminal UI that discovers, pings, and ranks free AI models from NVIDIA NIM and OpenRouter in real time — then writes the best model directly into your OpenCode or OpenClaw config.

When to use this skill

• Select the fastest responding free LLM for your OpenCode or OpenClaw session
• Check live availability, latency, and uptime of free models before starting work
• Configure NVIDIA NIM / OpenRouter API keys interactively with browser auto-open
• Non-interactively capture the best model ID in shell scripts (frouter --best)
• Benchmark free model tiers (S+ → C based on SWE-bench Verified scores)
• Switch models on the fly without manually editing JSON config files

Installation

# One-shot (no install)
npx frouter-cli

# Global install (recommended)
npm i -g frouter-cli
# or with bun
bun install -g frouter-cli

Instructions

Step 1: First run — API key setup wizard

frouter

On first run, the setup wizard opens provider key pages in your browser automatically. Press ESC to skip any provider you don't have a key for yet.

API key sources:

Keys are stored at ~/.frouter.json (permissions 0600).

You can also pass keys via environment variables:

NVIDIA_API_KEY=nvapi-xxx frouter
OPENROUTER_API_KEY=sk-or-xxx frouter

Step 2: Interactive TUI — select and apply a model

frouter

The TUI pings all models every 2 seconds. Use arrow keys to navigate, / to search, and Enter to apply a model to OpenCode or OpenClaw.

Key actions:

Target picker (after Enter):

Configs written:

• OpenCode CLI → ~/.config/opencode/opencode.json
• OpenClaw → ~/.openclaw/openclaw.json

Existing configs are automatically backed up before overwriting.

Step 3: Non-interactive best-model selection (scripts)

# Print best model ID after ~10 s analysis
frouter --best

# Capture in a variable
MODEL=$(frouter --best)
echo "Best model: $MODEL"

Selection tri-key sort: status=up → lowest avg latency → highest uptime.

Step 4: Tune auto-sort behavior (optional)

# Pause auto-resorting for 2.5 s after scroll navigation
FROUTER_SCROLL_SORT_PAUSE_MS=2500 frouter

# Disable rolling metrics cache (force legacy recompute)
FROUTER_METRICS_CACHE=0 frouter

Or persist in ~/.frouter.json:

{
  "ui": { "scrollSortPauseMs": 1500 }
}

TUI columns

Examples

Example 1: Start frouter and apply a model to OpenCode

frouter
# Navigate to desired model → Enter → select OpenCode CLI → Enter

Example 2: Get best model for a script

MODEL=$(frouter --best)
opencode --model "$MODEL" "refactor this function"

Example 3: Set API key without entering TUI

NVIDIA_API_KEY=nvapi-xxx frouter --best

Example 4: Test-onboard from a clean state

# Without touching your real config
npm run test:fresh-start

Model tier scale (SWE-bench Verified)

Best practices

1. Set both API keys — NIM and OpenRouter have overlapping models; having both maximises coverage
2. Use --best in scripts instead of hardcoding a model ID — model availability changes
3. Press S to write config only when you want to set the model without launching the tool
4. T to filter tier — if you only want S+ or S models, press T to cycle the tier filter
5. Keep frouter updated — model catalog syncs daily via GitHub Actions; accept in-app update prompts

References

• GitHub: jyoung105/frouter
• npm: frouter-cli
• Homepage
• TUI guide — full keyboard shortcuts and column reference
• Providers guide — NVIDIA NIM and OpenRouter setup