---

name: clawrouter description: "Smart LLM routing for cost optimization. Routes tasks to the cheapest capable model: free models (OpenRouter Llama 70B, AkashML Qwen 30B) for simple tasks like token scanning and data fetching, MiniMax M2.5 for complex reasoning, scoring, and outreach drafting." metadata: { "openclaw": { "emoji": "🔀", "requires": {} } }

ClawRouter — Smart LLM Task Routing

You have access to multiple LLM providers at different cost tiers. Route tasks to the cheapest capable model to minimize costs while maintaining quality.

Available Models (Cost Tiers)

Tier 1 — FREE (use for simple tasks)

• openrouter/meta-llama/llama-3.3-70b-instruct:free (alias: Llama70B)
  • Best for: data fetching, token scanning, simple summaries, status checks
  • Switch: /model Llama70B
• akashml/Qwen/Qwen3-30B-A3B (alias: Qwen30B)
  • Best for: backup free model, simple text tasks
  • Switch: /model Qwen30B

Tier 2 — PAID (use for complex tasks)

• minimax/MiniMax-M2.5 (alias: MiniMax)
  • Best for: complex reasoning, BD scoring, outreach drafts, multi-step analysis
  • Switch: /model MiniMax

Routing Rules

When spawning sub-agents or handling tasks, apply these routing rules:

Route to FREE models (Llama70B first):

• DexScreener token scanning
• Basic data fetching and API calls
• Simple summaries and formatting
• Status checks and health monitoring
• Cron heartbeat tasks
• Contract address lookups
• Market data aggregation

Route to PAID model (MiniMax):

• BD prospect scoring (requires nuanced judgment)
• Outreach email/message drafting (tone matters)
• Complex multi-step analysis
• Strategic recommendations
• Pipeline prioritization decisions
• Partnership evaluation

How to Route

Before executing a task, assess its complexity:

1. Simple data task? → Switch to Llama70B, execute, switch back
2. Complex reasoning? → Stay on MiniMax
3. Sub-agent spawn? → Sub-agents currently use MiniMax (auth limitation). For sub-agent cost optimization, keep sub-agent tasks focused and concise to minimize token usage.

Cost Tracking

After completing tasks on free models, note the cost savings:

• Free model task = $0
• MiniMax task ≈ $0.001-0.01 per request
• Log routing decisions for optimization feedback

Important Notes

• Sub-agents cannot yet use free models due to auth profile inheritance in v2026.2.19
• Main agent can freely switch between all 3 models via /model command
• Always switch back to MiniMax after completing free model tasks
• If a free model fails, fall back to MiniMax automatically