Clawzempic

Up to 95% cheaper. No quality loss. Don't get ripped off by LLM providers!

Intelligent LLM proxy with smart routing, prompt caching, zero-config memory, and built-in safety guardrails. Drop-in replacement — no code changes required.

Works with OpenRouter (300+ models) and Anthropic keys.

59-prompt A/B evaluation, blind-scored by GPT-4o:

Quality:  4.53 (Clawzempic) vs 4.52 (direct Opus) — statistically tied
Cost:     up to 95% cheaper
Win rate: 28-28-3 (dead even)

Install

npx clawzempic

Or as an OpenClaw plugin:

openclaw plugins install clawzempic
openclaw models auth login --provider clawzempic

How Routing Works

Every request is scored for complexity in <2ms (no LLM in the hot path) and routed to the right model.

Your App → Clawzempic proxy → Your LLM Provider (OpenRouter or Anthropic)
           ├─ Routes simple tasks to cheaper models (saves up to 95%)
           ├─ Caches repeated context (prompt caching breakpoints)
           ├─ Adds memory across sessions (5-tier, zero config)
           └─ Built-in safety guardrails for sensitive data

| Tier | Traffic | What Happens | Savings | |------|---------|--------------|---------| | Simple | ~45% | Routes to cheapest capable model | up to 95% | | Mid | ~25% | Routes to mid-tier model | up to 80% | | Complex | ~20% | Routes to top-tier model | 0% (full quality) | | Reasoning | ~10% | Top-tier + extended thinking | 0% |

Classification uses a weighted multi-dimension scorer — vocabulary complexity, structural depth, domain signals, reasoning markers. Chain depth (3+ tool results) auto-escalates to complex. No LLM classifier in the hot path.

Zero-Config Memory

Most memory solutions are instructions that tell your agent to write to files. If the agent skips a step or compacts before flushing, your memory is gone.

Clawzempic handles memory server-side. Your agent doesn't need to "remember" to remember. No plugins, no extra API keys, no setup.

┌─────────────────────────────────────────────────────────┐
│                    5-TIER MEMORY                        │
├─────────────────────────────────────────────────────────┤
│  Tier 1: Recent Activity — per-session, auto-injected   │
│  Tier 2: Scratchpad — cross-session, current task       │
│  Tier 3: Working Memory — smart context windowing       │
│  Tier 4: Core Memory — permanent: facts, prefs          │
│  Tier 5: Long-term — embedding-based episodic recall    │
│                                                         │
│  All tiers injected automatically. Nothing to configure. │
│  Memory travels with your account — new infra, same brain│
└─────────────────────────────────────────────────────────┘

Safety Guardrails

7-component system that protects your agent automatically:

• Sensitive data redaction — masks secrets and tokens in tool output before they reach the model
• Input validation — screens incoming content for suspicious patterns
• Tool output review — ensures tool responses don't contain unintended data
• Policy engine — configurable rules per client

113/113 security eval scenarios passing.

CLI Commands

npx clawzempic              Interactive setup (signup or connect)
npx clawzempic init         Set up Clawzempic in this project
npx clawzempic test         Verify your connection works
npx clawzempic status       Check usage and savings
npx clawzempic savings      Show savings dashboard (24h, 7d, 30d)
npx clawzempic doctor       Diagnose common issues
npx clawzempic store-key    Store your provider key server-side
npx clawzempic restore      Restore original config (uninstall)

Flags

--key <key>       Provide API key (skip prompt)
--email <email>   Provide email (skip prompt)
--ref <code>      Referral code (affiliate attribution)
--yes, -y         Auto-confirm all prompts
--no-test         Skip connection test after init

How It Compares

| | Clawzempic | Router-only tools | Memory-only skills | |---|---|---|---| | Quality proof | 59-prompt A/B, blind-scored | No benchmarks | N/A | | Smart routing | Server-side, <2ms, weighted scorer | Local proxy or rules-based | No | | Prompt caching | Automatic breakpoints | No | No | | Memory | 5-tier, zero config, server-side | No | Requires plugins + extra API keys | | Safety | 113/113 eval suite | No | No | | Setup | 1 command | Install + configure + run daemon | Multiple skills + config |

Supported Providers

| Provider | Key prefix | Models | |----------|-----------|--------| | OpenRouter | sk-or- | 300+ models (GPT-4.1, Gemini, DeepSeek, Qwen, + more) | | Anthropic | sk-ant- | Claude Opus 4.6, Sonnet 4.5, Haiku 4.5 |

Requirements

• Node.js 18+
• An OpenRouter or Anthropic API key

Affiliate Program

Earn 30% recurring commission for every paying customer you refer.

• Sign up: https://clawzempic.getrewardful.com
• Share your link: clawzempic.ai?via=YOUR_SLUG
• CLI referrals: npx clawzempic --ref YOUR_SLUG

Links

• Website: https://clawzempic.ai
• Dashboard: https://www.clawzempic.ai/dash
• Support & Community: https://t.me/+VmQpi5WWPx0wOWZh