Detailed Description of Everclaw: AI Inference You Own, Forever Powering Your OpenClaw Agents

Introduction to Everclaw

Everclaw is a groundbreaking open-source solution designed to empower users running an OpenClaw agent by providing decentralized AI inference capabilities. Unlike traditional centralized AI services that rely on proprietary APIs with credit limits and billing cycles, Everclaw enables users to leverage their own staked Morpheus (MOR) tokens for persistent, censorship-resistant AI access. This system ensures that the AI models—such as GLM-5, GLM 4.7 Flash, Kimi K2.5, and others—are powered by decentralized infrastructure, reducing dependency on single-point failures.

The project is built around two core pillars:

Decentralized Inference via Morpheus Network – Users connect their OpenClaw agent to the Morpheus P2P network, allowing it to access open-source AI models without relying on third-party APIs.
Self-Owned Access with Staked MOR Tokens – Instead of paying for API credits, users stake MOR tokens as a refundable deposit, ensuring they retain control over their inference sessions indefinitely.

This guide explores Everclaw’s architecture, installation process, key features, and how it transforms AI workflows by eliminating downtime, reducing costs, and enhancing security.

Core Concepts Behind Everclaw

1. Morpheus Network: The Backbone of Decentralized Inference

Morpheus is a decentralized AI inference network that enables peer-to-peer (P2P) model hosting and request routing. Unlike centralized APIs, Morpheus operates on a blockchain-based system where users can:

Host models (e.g., GLM-5, Kimi K2.5) on their own nodes.
Stake MOR tokens to secure inference sessions.
Route requests through a decentralized network, ensuring reliability and censorship resistance.

Everclaw acts as a bridge between OpenClaw agents and Morpheus, allowing seamless integration without requiring users to manage blockchain nodes themselves.

2. Staking MOR Tokens for Inference Access

Unlike traditional AI APIs that charge per request or credit limit users after usage, Everclaw operates on an economic model where:

Users stake MOR tokens as a refundable deposit.
Each staked session lasts 7 days by default, at which point the MOR is returned to the user.
The same tokens can be reused for new sessions indefinitely.

This ensures that users never run out of inference credits and retain full control over their AI access, even if an API key expires or costs fluctuate.

3. Open-Source First Model Router

Everclaw prioritizes open-source models while providing a fallback to premium services like Claude. The model router dynamically assigns tasks based on complexity:

Simple tasks (e.g., cron jobs, quick lookups) → GLM 4.7 Flash (fast, lightweight).
Standard tasks (writing, research, coding) → GLM-5 (Opus 4.5-level reasoning).
Complex or unhandled tasks → Claude (as a last resort).

This approach ensures that users always get the best possible model for their needs without unnecessary costs.

Installation Process: Setting Up Everclaw

Prerequisites

Before installing Everclaw, ensure your system meets these requirements:

macOS (Intel or Apple Silicon).
OpenClaw agent installed and running (npm install -g openclaw@latest).
Node.js v18+ (auto-installed via Homebrew or nvm for non-admin users).
ETH/USDC on Base blockchain (to swap for MOR tokens).

One-Liner Installer (Recommended)

The easiest way to install Everclaw is by running the following command:

curl -fsSL https://get.everclaw.xyz | bash

This script:

Checks for dependencies (curl, git, Node.js).
Auto-installs missing tools with verification.
Works on both admin and non-admin macOS accounts.
Supports Apple Silicon and Intel Macs.
Clones Everclaw to the correct location (~/.openclaw/workspace/skills/everclaw).
Bootstraps a free GLM-5 starter key (1,000 requests/day).

Alternative Installation Methods

If you prefer manual installation, you can use one of these methods:

From ClawHub

clawhub install everclaw-inference

⚠️ Note: Avoid using the bare everclaw slug on ClawHub—it belongs to a different product (Everclaw Vault). Refer to CLAWHUB_WARNING.md for details.

Manual Clone

git clone https://github.com/EverClaw/Everclaw.git ~/.openclaw/workspace/skills/everclaw

⚠️ Warning: Use everclaw-inference (not just everclaw) to avoid confusion.

Post-Installation Configuration

1. Basic Setup (Gateway Only)

To configure your OpenClaw agent with Everclaw’s decentralized inference:

node ~/.openclaw/workspace/skills/everclaw/scripts/setup.mjs \
  --template gateway-only \
  --key YOUR_KEY \
  --apply \
  --test \
  --restart

Replace YOUR_KEY with an API key from app.mor.org.
This sets up a gateway-only configuration, meaning your agent will route requests through Morpheus without requiring local P2P inference.

2. Full Setup (Local P2P + Gateway)

For users who want both decentralized and local inference, follow these steps:

Step 1: Install the Morpheus Proxy-Router

~/.openclaw/workspace/skills/everclaw/scripts/install.sh

This installs the proxy router that connects your agent to the Morpheus P2P network.

Step 2: Create a Wallet (Stored in macOS Keychain)

node ~/.openclaw/workspace/skills/everclaw/scripts/everclaw-wallet.mjs setup

This creates an encrypted wallet stored in your macOS Keychain.
No external accounts are required.

Step 3: Swap ETH or USDC for MOR Tokens

node ~/.openclaw/workspace/skills/everclaw/scripts/everclaw-wallet.mjs swap eth 0.05

Send 0.05 ETH to the wallet address provided.
The script automatically swaps it for MOR tokens on Base.

Step 4: Approve MOR Staking

node ~/.openclaw/workspace/skills/everclaw/scripts/everclaw-wallet.mjs approve

This allows your wallet to stake MOR tokens for inference sessions.

Step 5: Install Proxy and Guardian (Auto-Starts on Boot)

~/.openclaw/workspace/skills/everclaw/scripts/install-proxy.sh

This sets up a local proxy that handles all blockchain interactions.
The Guardian component ensures self-healing in case of failures.

Key Features of Everclaw

1. Decentralized Inference with Morpheus Proxy Router

Everclaw’s Morpheus Proxy Router connects your OpenClaw agent to the decentralized network, allowing it to:

Route requests through a P2P infrastructure.
Automatically handle session management, including renewals and fallbacks.
Support multiple models (GLM-5, GLM 4.7 Flash, Kimi K2.5) with dynamic routing.

2. OpenAI-Compatible Proxy

Everclaw’s proxy translates standard OpenAI API calls into Morpheus-compatible requests. This means:

Any OpenClaw client can interact with Everclaw without modifications.
Users can switch between models by changing the model parameter in their requests.

3. Auto-Session Manager & Fallback Mechanisms

Everclaw’s system ensures seamless AI access even if a session expires mid-request:

Auto-Retry on Session Expiry – If a 7-day session ends, Everclaw opens a new one automatically.
Cooldown-Safe Errors – Returns proper OpenAI error types to prevent misclassification as billing issues.
Three-Tier Model Router –
Light tasks (e.g., cron jobs) → GLM 4.7 Flash.
Standard tasks (writing, coding) → GLM-5.
Complex/fallback tasks → Claude.

4. x402 Payment Client & ERC-8004 Agent Registry

Everclaw enables on-chain payments and agent discovery:

x402 Payment Handling – Automatically detects payment-required responses (e.g., HTTP 402) and signs USDC transactions via EIP-712.
ERC-8004 Agent Registry – Allows your agent to discover, verify, and pay other agents on-chain without intermediaries.

5. Gateway Guardian v5: Self-Healing & Billing Awareness

Everclaw’s Gateway Guardian ensures reliability by:

Direct Curl Probes – Uses tiny prompts (~50 chars) to check gateway health without bloating logs.
No Signal Spam – Prevents accidental message delivery during failures.
Billing-Aware Escalation – Classifies errors as billing, transient, or timeout and takes appropriate actions (e.g., backoff, restart).
DIEM Reset Awareness – Adjusts probe intervals based on Base’s daily DIEM reset cycle.
Proactive Credit Monitoring – Warns when the Venice DIEM balance drops below a threshold.

6. Three-Shift Task Planning

Everclaw introduces a structured work management system to optimize productivity: | Shift | Time Range | Purpose | |-------|------------------|----------------------------------| | Morning (6 AM) | Meetings, comms | User-interactive tasks | | Afternoon (2 PM) | Deep work | Coding, writing, building | | Night (10 PM) | Autonomous | Research, maintenance, prep |

Priority Tiers – Tasks are categorized as P1 (must-do), P2 (should-do), or P3 (could-do).
Approval Workflow – Nothing executes without user approval.
Shift Handoff – Each shift logs completion summaries for the next.

7. Always-On Power Configuration

Since Everclaw agents run 24/7, macOS power settings must be configured to prevent interruptions:

sudo bash ~/.openclaw/workspace/skills/everclaw/scripts/always-on.sh

This ensures:

No sleep or standby during operation.
Caffeinate LaunchAgent prevents system sleep.
Power Nap + Wake on LAN allows remote wakeup.
Auto-restart after power failures.

8. SkillGuard & ClawdStrike: Security Auditing

Everclaw includes security tools to protect against:

SkillGuard – Detects credential theft, code injection, and data exfiltration in installed skills.

  node ~/.openclaw/workspace/skills/security/skillguard/src/cli.js scan

ClawdStrike – Audits OpenClaw configurations for security gaps (e.g., open ports, weak auth).

  bash ~/.openclaw/workspace/skills/security/clawdstrike/scripts/collect_verified.sh

9. PromptGuard: Injection Defense

To prevent AI agents from being manipulated:

Detects injection attempts in English, Korean, Japanese, and Chinese.
Rates threats by severity.
Shares threat intelligence via the HiveFence network for collective defense.

10. Bagman: Secure Key Management

Everclaw ensures that sensitive keys (e.g., MOR wallet credentials) are stored securely:

Uses 1Password runtime injection to avoid disk storage.
Supports ephemeral session keys with limited permissions.
Implements EIP-7710 delegation for scoped agent authority.

Available AI Models & Model Selection

Everclaw supports a variety of open-source and premium models, dynamically routed based on task complexity:

| Model | Type | Router Tier | Notes | |---------------------|---------------|-------------------|----------------------------------------| | GLM-5 | General | ⭐ STANDARD + HEAVY | Default model (Opus 4.5-level reasoning). | | GLM 4.7 Flash | Fast | LIGHT | Quick responses for trivial tasks. | | Kimi K2.5 | General | — | Solid general-purpose model. | | Kimi K2 Thinking| Reasoning | — | Extended thinking for complex problems. | | GLM-5:web | General + Web | — | GLM-5 with web search capability. | | GLM 4.7 | General | — | Full GLM model. | | Qwen3 235B | General | — | Large parameter count (frontier quality).| | GPT-OSS 120B | General | — | OpenAI’s open-source alternative. |

Why GLM-5?

Zhipu’s GLM-5 (744B MoE model) is benchmarked against Claude Opus 4.5:

92.7% AIME 2026 (advanced reasoning).
86% GPQA-Diamond (structured output).
50.4% Humanity’s Last Exam (with tools).

This model excels in hallucination resistance, making it ideal for reasoning, coding, and agentic tasks.

MOR Token Economics: Staking & Rewards

How MOR Tokens Work

Staked, not spent: Unlike traditional API keys, MOR tokens are refundable deposits.
Session Duration:
1 hour → ~11 MOR returned.
1 day → ~274 MOR returned.
Default: 7 days → ~1,915 MOR returned.

Getting Started with MOR

To begin using Everclaw:

Swap ETH or USDC on Base for MOR tokens using the included scripts:

   node ~/.openclaw/workspace/skills/everclaw/scripts/everclaw-wallet.mjs swap eth 0.05

Stake the acquired MOR tokens to open an inference session.
Repeat the process—your MOR is recycled indefinitely.

Cost Efficiency

No API bills or credit limits: Everclaw ensures you always have access to AI models.
Staked MOR = Refundable deposit: You only pay for compute, not consumables.

Quick Reference: Command Cheat Sheet

| Action | Command | |----------------------------|------------------------------------------------------------------------| | Setup (Gateway Only) | node scripts/setup.mjs --key YOUR_KEY --apply --test --restart | | Dry-Run Setup | node scripts/setup.mjs | | Install from ClawHub | clawhub install everclaw-inference | | Update (ClawHub) | clawhub update everclaw-inference | | Check Version | bash skills/everclaw/scripts/install-everclaw.sh --check | | Install Router | bash scripts/install.sh | | Create Wallet | node scripts/everclaw-wallet.mjs setup | | Swap ETH→MOR | node scripts/everclaw-wallet.mjs swap eth 0.05 | | Swap USDC→MOR | node scripts/everclaw-wallet.mjs swap usdc 50 | | Approve MOR Staking | node scripts/everclaw-wallet.mjs approve | | Install Proxy + Guardian | bash scripts/install-proxy.sh | | Start Router | bash scripts/start.sh | | Check Proxy Health | curl http://127.0.0.1:8083/health | | Route a Prompt | node scripts/router.mjs "your prompt here" | | x402 Request | node scripts/x402-client.mjs GET ... | | Scan a Skill | node security/skillguard/src/cli.js scan |

Who Is Everclaw For?

Everclaw is designed for users who: ✅ Run an OpenClaw agent (Mac mini, laptop, or VPS) and want 24/7 uptime. ✅ Use AI daily (writing, research, coding, communication) without worrying about API downtime. ✅ Own or want to own MOR tokens and want to earn inference instead of sitting idle. ✅ Believe in decentralization—they don’t trust a single company’s API for AI infrastructure. ✅ Need secure crypto handling (DeFi, token management, wallet operations). ✅ Have experienced API key expirations or credit limits and want a permanent solution.

How Everclaw Solves Common Problems

Problem 1: API Key Expiry & Credit Limits

Everclaw’s Solution: Users stake MOR tokens as a refundable deposit, ensuring they never run out of inference credits. The same tokens can be reused indefinitely.

Problem 2: Downtime Due to Billing Failures

Everclaw’s Solution: The Gateway Guardian v5 classifies errors as billing, transient, or timeout and takes appropriate actions (e.g., backoff, restart). The Auto-Session Manager ensures sessions renew automatically.

Problem 3: Dependency on Single-Point Failures

Everclaw’s Solution: By routing requests through the Morpheus P2P network, Everclaw provides a censorship-resistant fallback to open-source models (e.g., GLM-5) and premium services (Claude).

Problem 4: High API Costs & Unexpected Charges

Everclaw’s Solution: Users pay only for compute, not consumables. The x402 Payment Client handles USDC transactions securely, with budget controls to prevent runaway spending.

Security & Privacy Considerations

1. Wallet Security

Everclaw wallets are stored in the macOS Keychain, ensuring secure storage without external accounts.
Users can swap ETH/USDC for MOR tokens directly from their terminal.

2. Data Protection

PromptGuard detects and blocks injection attempts (e.g., credential theft, code manipulation).
Bagman ensures keys are never stored on disk but use secure runtime patterns.

3. Decentralization & Censorship Resistance

Everclaw’s reliance on the Morpheus P2P network means it is not dependent on a single API provider.
Users can discover and pay other agents on-chain via the ERC-8004 registry, reducing trust in intermediaries.

Performance & Scalability

1. Local vs. Decentralized Inference

| Feature | Gateway (Decentralized) | Local P2P (Staked MOR) | |-----------------------|-------------------------------|--------------------------------| | Speed | Fast (via Morpheus network) | Slower (local latency) | | Cost | Free (stake MOR once) | Higher (but owned compute) | | Uptime Guarantee | High (decentralized) | Depends on local node health |

2. Model Performance Benchmarks

Everclaw’s models are benchmarked against industry standards:

GLM-5: Matches Claude Opus 4.5 in reasoning, coding, and analysis.
GLM 4.7 Flash: Optimized for fast, lightweight tasks.
Kimi K2.5: Solid general-purpose model.

3. Scaling with MOR Staking

Since MOR tokens are staked, not spent, users can:

Increase staking duration (e.g., 7 days → 1 month) for longer sessions.
Swap more ETH/USDC to earn more MOR and increase inference capacity.

Troubleshooting Common Issues

1. "Session Expired" Errors

Solution: Everclaw’s Auto-Session Manager automatically renews sessions before expiry. If a session fails, it opens a new one.
Check logs:

  tail -f ~/.openclaw/logs/guardian.log

2. Proxy or Network Connectivity Issues

Ensure the Morpheus proxy is running:

  bash scripts/install.sh

Verify connectivity with:

  curl http://127.0.0.1:8083/health

3. MOR Token Swap Failures

If swapping ETH/USDC fails, check wallet balance and ensure the correct chain (Base).
Use the included script:

  node scripts/everclaw-wallet.mjs swap eth 0.05

4. Always-On Power Issues

If your Mac sleeps unexpectedly:

sudo bash skills/everclaw/scripts/always-on.sh --restore

This restores default power settings.

Conclusion: Why Everclaw Is a Game-Changer

Everclaw transforms AI workflows by providing: ✔ Decentralized, self-owned inference – No more API key expirations or credit limits. ✔ Open-source first model routing – Users get access to top-tier models (GLM-5) without premium costs. ✔ Self-healing & billing awareness – The Gateway Guardian ensures reliability even in failure scenarios. ✔ Structured work management – Three-shift task planning optimizes productivity. ✔ Security & privacy – Secure key storage, injection defense, and on-chain agent discovery.

For users who rely on AI for work, research, or automation, Everclaw eliminates downtime, reduces costs, and enhances security—all while empowering them to control their own inference infrastructure.

Next Steps

Install Everclaw using the one-liner:

   curl -fsSL https://get.everclaw.xyz | bash

Get your MOR tokens by swapping ETH/USDC on Base.
Configure your OpenClaw agent:

   node ~/.openclaw/workspace/skills/everclaw/scripts/setup.mjs --key YOUR_KEY --apply --restart

Explore advanced features like three-shift task planning or on-chain payments.

With Everclaw, AI inference becomes your own, forever.

[Image Suggestions for This Description]

Diagram of Morpheus P2P Network: Showing how OpenClaw agents connect to the decentralized network.
Everclaw Installation Flowchart: Visualizing the one-liner installer and manual steps.
Model Router Comparison: A table highlighting GLM-5 vs. GLM 4.7 Flash vs. Claude.
Always-On Power Configuration: Screenshot of macOS power settings (disable sleep, enable Wake on LAN).
Wallet Setup in Keychain: Illustration of how MOR tokens are stored securely.
Three-Shift Task Planning Example: Sample schedule for morning/afternoon/night shifts.

Final Note: Everclaw is still evolving—check the GitHub repository for updates and community-driven improvements! 🚀

Everclaw: Open-Source AI Inference on Staked MOR