n8n-nodes-openrouter-selector

n8n community node for intelligent OpenRouter model selection based on task, budget, and benchmarks.

Features

• Task-Based Selection: Optimized model recommendations for:

  • Translation (Chinese ↔ English optimized)
  • Coding & Development
  • Data Analysis & Reasoning
  • Vision & Image Analysis
  • Conversational Chat
  • Text Embedding
  • Summarization
  • Mathematical Reasoning

• Budget Awareness: Three budget tiers:

  • Cheap: Lowest cost, quality secondary
  • Balanced: Good price-performance ratio
  • Premium: Best quality, cost no concern

• Benchmark-Based Scoring: Uses external benchmark data:

  • Artificial Analysis (Intelligence, Coding, Math indices)
  • LMSYS Chatbot Arena (Elo ratings)
  • LiveBench (Coding, Math, Reasoning scores)

• Dynamic Model Override: Select specific models with real-time scoring preview

• Flexible Filtering:

  • Minimum context length
  • JSON mode requirement
  • Vision/multimodal requirement
  • Cost limits
  • Provider whitelist/blacklist

Installation

In n8n

1. Go to Settings → Community Nodes
2. Click Install a community node
3. Enter: n8n-nodes-openrouter-selector
4. Click Install

Manual Installation

# In your n8n custom nodes directory
cd ~/.n8n/custom
npm install n8n-nodes-openrouter-selector

Development Installation

git clone https://github.com/ecolights/n8n-nodes-openrouter-selector.git
cd n8n-nodes-openrouter-selector
pnpm install
pnpm build

# Link to n8n
cd ~/.n8n/custom
npm link /path/to/n8n-nodes-openrouter-selector

Prerequisites

This node requires:

1. Supabase Database with the benchmark schema (see docs/BENCHMARK_SYSTEM.md):

   • models_catalog - OpenRouter model data (synced via separate workflow)
   • model_name_mappings - Master mapping table (manually maintained)
   • model_benchmarks - Benchmark scores (auto-synced weekly)
   • task_profiles - Task-specific scoring weights
   • unmatched_models - Review queue for new models

2. n8n Workflow: TN_benchmark_sync_artificial_analysis for weekly benchmark sync

3. Credentials: Supabase URL and API key (service role for write access)

Quick Schema Setup

The full schema with triggers, functions, and RLS policies is documented in docs/BENCHMARK_SYSTEM.md.

Core Tables Overview:

-- Model name mappings (Source of Truth - manually maintained)
CREATE TABLE model_name_mappings (
  openrouter_id TEXT UNIQUE NOT NULL,      -- e.g. "anthropic/claude-sonnet-4"
  canonical_name TEXT NOT NULL,             -- Display name
  aa_name TEXT,                             -- Artificial Analysis name
  aa_slug TEXT,                             -- AA URL slug
  provider TEXT,                            -- anthropic, openai, google, etc.
  verified BOOLEAN DEFAULT false
);

-- Benchmark scores (auto-filled by sync workflow)
CREATE TABLE model_benchmarks (
  openrouter_id TEXT REFERENCES model_name_mappings(openrouter_id),
  -- Artificial Analysis
  aa_intelligence_index DECIMAL(5,2),
  aa_coding_index DECIMAL(5,2),
  aa_math_index DECIMAL(5,2),
  -- LMSYS Arena
  lmsys_elo INTEGER,
  -- LiveBench
  livebench_overall DECIMAL(5,2),
  livebench_coding DECIMAL(5,2),
  -- Computed composites (via trigger)
  composite_general DECIMAL(5,2),
  composite_code DECIMAL(5,2),
  composite_math DECIMAL(5,2)
);

-- Task-specific scoring weights
CREATE TABLE task_profiles (
  task_name TEXT UNIQUE NOT NULL,           -- general, code, translation, etc.
  weight_aa_intelligence DECIMAL(3,2),
  weight_aa_coding DECIMAL(3,2),
  weight_lmsys_elo DECIMAL(3,2),
  weight_livebench DECIMAL(3,2),
  boost_anthropic DECIMAL(3,2),
  boost_openai DECIMAL(3,2),
  boost_deepseek DECIMAL(3,2)
);

Usage

Basic Usage

1. Add the OpenRouter Model Selector node to your workflow
2. Configure credentials (Supabase URL + API Key)
3. Select a Task Category (e.g., "coding")
4. Select a Budget (e.g., "balanced")
5. Execute to get the recommended model

Output Format

Full Output (default):

{
  "recommended": {
    "modelId": "anthropic/claude-sonnet-4",
    "provider": "anthropic",
    "displayName": "Claude Sonnet 4",
    "contextLength": 200000,
    "supportsJson": true,
    "modality": "text+image->text",
    "pricing": {
      "promptPer1kUsd": 0.003,
      "completionPer1kUsd": 0.015,
      "combinedPer1kUsd": 0.009
    },
    "score": 87.5,
    "scoreBreakdown": {
      "benchmarkFit": 38,
      "taskFit": 28,
      "budgetFit": 18,
      "capabilityFit": 9,
      "providerBonus": 1.5
    },
    "reasoning": "Excellent benchmark performance for Coding & Development, ideal balanced pricing, anthropic provider bonus (+15%)."
  },
  "alternatives": [...],
  "queryMetadata": {
    "task": "coding",
    "budget": "balanced",
    "totalModelsEvaluated": 313,
    "modelsPassingFilters": 187,
    "executionTimeMs": 245
  }
}

Using the Selected Model

Connect the output to an HTTP Request node or OpenRouter integration:

[OpenRouter Model Selector] → [HTTP Request to OpenRouter API]
                                 URL: https://openrouter.ai/api/v1/chat/completions
                                 Body: { "model": "={{$json.recommended.modelId}}", ... }

Scoring Algorithm

The scoring formula is deterministic and based on external benchmarks:

score = (benchmark_fit × 0.4) + (task_fit × 0.3) + (budget_fit × 0.2) + (capability_fit × 0.1) × provider_boost

Components

| Component | Weight | Description | |-----------|--------|-------------| | Benchmark Fit | 40% | Score from external benchmarks (AA, LMSYS, LiveBench) | | Task Fit | 30% | How well the model matches task requirements | | Budget Fit | 20% | Cost alignment with budget preference | | Capability Fit | 10% | Context length, JSON support, verification status | | Provider Boost | ×1.0-1.2 | Task-specific provider bonuses |

Provider Boosts by Task

| Task | Provider Boosts | |------|-----------------| | Translation | DeepSeek +20%, Qwen +15%, Anthropic +10% | | Coding | Anthropic +15%, OpenAI +10%, DeepSeek +8% | | Analysis | Anthropic +15%, OpenAI +10%, Google +5% | | Vision | OpenAI +15%, Google +12%, Anthropic +8% | | Math | DeepSeek +15%, Qwen +12%, OpenAI +10% |

Configuration

Node Parameters

| Parameter | Type | Description | |-----------|------|-------------| | Task Category | Dropdown | Type of task (coding, translation, etc.) | | Budget | Dropdown | Cost preference (cheap, balanced, premium) | | Model Override | Dynamic Dropdown | Override with specific model | | Filters | Collection | Advanced filtering options | | Options | Collection | Output configuration |

Filter Options

| Filter | Type | Default | Description | |--------|------|---------|-------------| | Min Context Length | Number | 8000 | Minimum tokens | | Require JSON Mode | Boolean | false | Only JSON-capable models | | Require Vision | Boolean | false | Only multimodal models | | Max Cost per 1K | Number | 0 | Cost limit (0 = no limit) | | Provider Whitelist | Multi-select | [] | Include only these providers | | Provider Blacklist | Multi-select | [] | Exclude these providers |

Benchmark Sync Workflow

The node requires benchmark data to be synced weekly via n8n workflow.

n8n Workflow: TN_benchmark_sync_artificial_analysis

Trigger: Weekly (Sunday 03:00 UTC) + Manual + Webhook

Data Flow:

[1. Fetch Artificial Analysis] ──────┐
                                      │
[2. Fetch OpenRouter Catalog] ────────┼──► [4. Merge All Data]
                                      │           │
[3. Fetch Existing Mappings] ────────┘            ▼
                                      [5. Process & Match Models]
                                              │
                               ┌──────────────┴──────────────┐
                               ▼                             ▼
                    [6. Upsert Benchmarks]      [7. Store Unmatched]
                               │                             │
                               └──────────────┬──────────────┘
                                              ▼
                                   [8. Merge Results]
                                              │
                                              ▼
                                   [9. Telegram Notification]

Detailed Documentation: See docs/BENCHMARK_SYSTEM.md

Manual Sync Trigger

# Via n8n Webhook
curl -X POST https://n8n.dev.ecolights.de/webhook/benchmark-sync

Development

# Install dependencies
pnpm install

# Build
pnpm build

# Watch mode
pnpm dev

# Lint
pnpm lint

# Format
pnpm format

License

MIT

Author

EcoLights ([email protected])

Links

• OpenRouter
• n8n Community Nodes
• Supabase