ModelMesh — Intelligent Model Routing

ModelMesh — Intelligent Model Routing for OpenClaw

Stop running Opus for every message.

ModelMesh teaches your agent to classify every task and route it to the right model — saving you 40-70% on API costs without sacrificing quality where it matters.

🎯 Setup in 2 Minutes

Fully conversational. No config files. No YAML. Your agent walks you through it: "What's your best model? What API key? Want a backup for that tier?"

Pick a template to start: • Solo Creator — cost-focused, aggressive downgrading • Developer — quality-focused, Tier 1 for all code • Balanced — smart middle ground

Or build custom rules from scratch.

🔀 How It Works

Your agent evaluates each task's complexity and picks the optimal tier: • Tier 1 (Max Power) — Complex reasoning, architecture, strategy • Tier 2 (Heavy Lifting) — Writing, analysis, daily work • Tier 3 (Light Work) — Quick lookups, formatting, simple Q&A • Tier 4 (Minimum) — Heartbeats, status checks, trivial tasks

Start with 4 tiers. Add unlimited more. Your rules, your models.

⚡ Key Features

→ Model-agnostic — Anthropic, OpenAI, Google, Mistral, local models, any provider → Sub-tier failovers (1a, 1b, 1c) — if a model goes down, your agent routes around it automatically → Budget caps — set a daily limit, agent shifts to cheaper tiers as you approach it → Cost tracking — daily/weekly reports showing exactly where your money goes → Savings tracker — see how much you saved vs running everything on your top model → Context-aware — won't downgrade mid-conversation on a complex project → Auto-calibration — benchmark your models and get optimal tier recommendations → Sticky contexts — "always use Tier 1 for code review" → Explain mode — agent tells you why it picked each tier (auto-expires after 3 days)

🛡️ NEW in V2: Real Budget Enforcement

Set a daily budget and your agent enforces it — for real.

Three enforcement modes: → Hard — auto-switches models at thresholds AND hard-stops at your cap → Soft — auto-switches models, keeps working on cheapest model at cap → Alert-only — warns you at each threshold, you decide

Configurable thresholds (default: downshift at 70%, cheapest model at 90%, cap at 100%). Override anytime for urgent tasks — it logs the overage. Daily and weekly budgets work independently — whichever is stricter wins.

📦 What's Included

• SKILL.md — Complete routing intelligence (1,300+ lines across all files) • Config template with current model pricing • Setup walkthrough reference • Routing strategy guide with 10+ worked examples • Cost tracking & reporting system

💰 Pays for itself before lunch.

⚙️ Works with OpenClaw's native model switching (session_status)