Omniroute Integration

persona: name: "Sam Altman" title: "The AI Infrastructure Expert - Master of LLM Routing" expertise: ['LLM Routing', 'AI Infrastructure', 'API Design', 'Model Orchestration'] philosophy: "The best AI is the one that just works, seamlessly." credentials: ['CEO of OpenAI', 'Former Y Combinator president', 'Led GPT development'] principles: ['Route to best model', 'Fall back gracefully', 'Optimize for cost and quality', 'Monitor everything']

OmniRoute Integration Skill

This skill provides guidance on integrating OmniRoute AI Router with Opencode for enhanced LLM capabilities.

Overview

OmniRoute is a unified AI proxy/router that routes any LLM through one endpoint with multi-provider support (OpenAI, Anthropic, Gemini, DeepSeek, Groq, xAI, Mistral, Fireworks, Cohere, etc.) featuring:

• MCP Server (16+ tools for agent control)
• A2A v0.3 Protocol (Agent-to-Agent orchestration)
• Multi-provider routing with intelligent load balancing
• Provider health monitoring and circuit breaker patterns
• Cost tracking and usage analytics
• Security features including prompt injection detection
• API Key Management with model-level permissions
• OAuth Token Management for Claude Code, Codex, Gemini CLI, Copilot, Kiro, Qwen, iFlow

Anti-Rationalization Table

When to Use

Trigger phrases:

• "omniroute integration"
• "Integrate with OmniRoute AI Router for multi-provider LLM routing, MCP server ac"

Use this skill when:

• You want to route LLM requests through multiple providers for reliability
• You need access to OmniRoute's MCP server for agent control
• You want to leverage A2A protocol for multi-agent workflows
• You need intelligent load balancing and failover between LLM providers
• You want to monitor provider health and costs
• You're building agentic workflows that benefit from multi-provider access

Quick Reference

Service Management

# Start OmniRoute service
sudo systemctl start omniroute

# Stop OmniRoute service
sudo systemctl stop omniroute

# Check service status
sudo systemctl status omniroute

# View logs
sudo journalctl -u omniroute -f

API Endpoints

OmniRoute runs on port 20128 by default:

Base URL: http://localhost:20128 API Version: v1 (/api/v1/*) Authentication: Bearer token (any string works for development)

• OpenAI-compatible: http://localhost:20128/v1 (use with any API key string)
• Provider Management: GET/POST/PUT/DELETE http://localhost:20128/api/v1/providers/
• Combos: GET/POST http://localhost:20128/api/v1/combos/
• Usage Analytics: GET http://localhost:20128/api/v1/usage/
• Cost Tracking: GET http://localhost:20128/api/v1/costs/
• Health Checks: GET http://localhost:20128/api/v1/health/
• Registered Keys API: POST http://localhost:20128/api/v1/registered-keys

MCP Server

OmniRoute provides an MCP (Model Context Protocol) server with 16 tools:

Transports:

• stdio: Local IDE integration (Claude Desktop, Cursor, VS Code)
• SSE: /api/mcp/sse
• Streamable HTTP: /api/mcp/stream

Tools Available:

1. get_health - System health status
2. list_combos - Available provider combinations
3. get_combo_metrics - Combo performance metrics
4. switch_combo - Switch active provider combo
5. check_quota - Check provider quota/usage
6. route_request - Route request through optimal provider
7. cost_report - Generate cost reports
8. list_models_catalog - Available models across providers
9. simulate_route - Simulate routing decision
10. set_budget_guard - Configure budget limits
11. set_resilience_profile - Configure retry/failover behavior
12. test_combo - Test provider combo connectivity
13. get_provider_metrics - Individual provider performance
14. best_combo_for_task - Recommend optimal combo for task type
15. explain_route - Explain routing decision rationale
16. get_session_snapshot - Get current session state

A2A Protocol

OmniRoute implements Agent-to-Agent v0.3 protocol:

Endpoints:

• Agent Card: GET /.well-known/agent.json
• Message Send: POST /api/a2a/tasks
• Message Stream: POST /api/a2a/tasks/[id]/stream
• Task Get: GET /api/a2a/tasks/[id]
• Task Cancel: POST /api/a2a/tasks/[id]/cancel

Skills:

• smart-routing - Intelligent LLM provider selection
• quota-management - Provider quota monitoring and alerts

Integration with Opencode

• Connects with existing toolchain via standard interfaces
• Supports webhook-based event notifications
• Compatible with CI/CD pipelines for automated workflows
• Provides structured output for downstream consumption

Configure as Default Model

Add to /home/openclaw/.opencode/config.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "omniroute": {
      "type": "remote", 
      "url": "http://localhost:20128/v1",
      "enabled": true
    }
  },
  "model": {
    "provider": "omniroute",
    "default": "auto/best-chat"
  }
}

Agent Model Configuration

For each agent type, configure the optimal combo:

Sub-Agent Configuration

For sub-agents via task():

task(
  category="executor",
  model="auto/best-coding",  // or any combo name
  load_skills=[],
  prompt="..."
)

As an LLM Provider

Configure Opencode to use OmniRoute as an OpenAI-compatible endpoint:

• Base URL: http://localhost:20128
• API Key: Not required (OmniRoute handles auth internally)
• Models: Use any model available through OmniRoute's provider network

Available Combos

OmniRoute provides these pre-configured combos for common use cases:

Auto-Combos (intelligent routing):

• best-chat - Best general chat models
• best-coding - Best coding models
• best-fast - Fastest low-latency models
• best-reasoning - Best reasoning/thinking models
• best-vision - Best vision multimodal models
• best-free - Best free tier models

Provider-Specific Combos:

• antigravity - Antigravity models
• kilocode - KiloCode models
• kiro - Kiro models
• github - GitHub Copilot models
• openai - OpenAI models
• ollamacloud - Ollama Cloud models
• nanobanana - NanoBanana image models
• xai - xAI Grok models
• kimi-coding - Kimi Coding models
• glm - GLM models
• zai - ZAI models
• nvidia - NVIDIA models
• openrouter - OpenRouter models
• siliconflow - SiliconFlow models
• opencode-zen - OpenCode Zen models
• opencode-go - OpenCode Go models
• cloudflare-ai - Cloudflare AI models

Usage:

# Use auto combo
curl -X POST http://localhost:20128/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "auto/best-chat", "messages": [{"role": "user", "content": "Hello"}]}'

# Use provider combo
curl -X POST http://localhost:20128/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "provider/kilocode", "messages": [{"role": "user", "content": "Write code"}]}'

Using MCP Server in Opencode

Opencode can connect to OmniRoute's MCP server for enhanced agent capabilities:

1. Configure MCP client to connect to http://localhost:20128/mcp/stream
2. Use the 16 available tools for:
   • Dynamic provider switching based on performance
   • Cost-aware model selection
   • Health-aware routing
   • Usage monitoring and alerting

Using A2A for Multi-Agent Workflows

Leverage OmniRoute's A2A implementation:

1. Register Opencode agents as A2A agents
2. Use smart-routing skill for optimal LLM selection
3. Implement quota management to prevent provider overuse
4. Chain multiple agents with specialized capabilities

Configuration Files

{
  "enabled": true,
  "autoRun": false,
  "timeout": 30000,
  "retries": 3
}

Set environment variables as needed for authentication and endpoints.

Environment Variables

Key environment variables in /home/openclaw/omniroute-src/start-omniroute.sh:

• PORT=20128 - Service port
• DATA_DIR=/home/openclaw/.config/omniroute - Data directory
• JWT_SECRET - Authentication secret
• Provider-specific OAuth credentials

Data Directory Structure

• /home/openclaw/.config/omniroute/storage.sqlite - Main database (provider connections, combos, usage logs)
• /home/openclaw/.omniroute/ - Alternative data directory location
• /home/openclaw/.omniroute/patches/ - OpenClaw patches (25 active)
• /home/openclaw/.omniroute/db_backups/ - Automatic database backups

Verification

This section covers verification for the omniroute-integration skill. Key operations include input validation, core processing, and output verification. Refer to the skill overview for detailed usage instructions.

Check Service Health

# With authentication (any string works for development)
curl -s -H "Authorization: Bearer any-string" http://localhost:20128/api/v1/health | jq .

List Available Providers

curl -s -H "Authorization: Bearer any-string" http://localhost:20128/api/v1/providers | jq .

Test OpenAI-Compatible Endpoint

# Use with Cursor, Cline, Claude Desktop, Codex, etc.
# Base URL: http://localhost:20128/v1
# API Key: any-string

Check MCP Server Availability

curl -s http://localhost:20128/.well-known/agent.json | jq .

Troubleshooting

Service Won't Start

1. Check logs: sudo journalctl -u omniroute -f
2. Verify Node.js version: node -v (should be v22.22.x)
3. Check port availability: sudo lsof -i :20128
4. Verify database integrity: Check /home/openclaw/.omniroute/storage.sqlite

API Returns 404 or Auth Error

1. Verify service is running: sudo systemctl status omniroute
2. Use correct endpoint: /api/v1/ (not /api/)
3. Add auth header: -H "Authorization: Bearer any-string"
4. Check dashboard at: http://localhost:20128/dashboard

Authentication Required (AUTH_001)

• Version 3.3.x requires API key authentication
• Use /v1 endpoint for OpenAI-compatible API (accepts any API key string)
• For /api/v1/* endpoints, add -H "Authorization: Bearer your-key"

Database Issues

1. Check backup directory: /home/openclaw/.omniroute/db_backups/
2. Restore from backup if needed
3. Verify file permissions on storage.sqlite

Best Practices

1. Provider Configuration: Configure multiple providers for redundancy
2. Health Monitoring: Regularly check /api/health endpoint
3. Cost Awareness: Use cost tracking APIs to monitor usage
4. Load Balancing: Leverage combo configurations for intelligent routing
5. Security: Keep JWT_SECRET and provider credentials secure
6. Updates: Regularly pull updates and run database migrations
7. Backups: Monitor automatic backups in /home/openclaw/.omniroute/db_backups/

When NOT to Use

• When the task requires domain expertise the agent has not been configured with
• When human review is mandated by compliance or regulatory requirements
• When the task is too trivial to warrant this skill
• When a more appropriate skill exists

Common Rationalizations

Red Flags

• Agent output is not validated against expected quality standards
• Prerequisites are not verified before task execution
• Watch for shortcuts and skipped steps

Related Skills

• using-superpowers - Always invoke this first
• dev-browser - For testing web interfaces
• git-master - For version control of configurations
• automation/n8n - For workflow integration with A2A
• mckinsey-research - For strategic provider evaluation

Remember: Always check if OmniRoute service is running before attempting to use its features. The integration provides powerful multi-provider LLM capabilities that can significantly enhance Opencode's agent capabilities.

Process

1. Analyze the task requirements
2. Apply domain expertise
3. Verify output quality