You’ve heard the buzz. OpenClaw is running agents that write code, triage inboxes, send morning briefings to Telegram, and monitor GitHub CI/CD pipelines — all while you sleep. You installed it. Now you’re staring at a config file wondering what to actually turn on.
That’s the problem this guide solves.
At Sphere Inc., we help businesses and power users design, deploy, and maintain AI systems that genuinely work — not just ones that look impressive in a demo. We’ve guided dozens of teams through OpenClaw setups across macOS, Linux servers, and cloud VMs, and we’ve documented every configuration trap, security pitfall, and skill conflict we’ve encountered along the way.
This guide is the result of that accumulated expertise. We’ll walk you through every one of OpenClaw’s 26 Tools and 53 bundled Skills — what each does, what risks it carries, and how to configure a setup that’s both powerful and defensible. By the end, you’ll know exactly which switches to flip and why.
|
1. What Is OpenClaw, and Why Does Setup Matter So Much?
OpenClaw is not a chatbot. It’s a self-hosted agent gateway – a persistent process that runs on your own hardware, connects to large language models (Claude, GPT, Gemini, local Ollama), and bridges those models to your real-world systems: your files, your terminal, your browser, your calendar, your messaging apps.
The difference between OpenClaw and talking to Claude in a browser comes down to three things: persistence, tools, and execution. A browser session disappears the moment you close the tab. OpenClaw keeps running. It doesn’t wait to be asked – with the right automation config, it monitors, acts, and pushes results to your phone.
That power is exactly why setup matters so much. Every tool you enable is a capability you’re granting to an AI system. Enable the right ones with the right guardrails, and you have infrastructure. Enable everything carelessly, and you’ve handed root access to a process that can be influenced by the content it reads. The difference is config, and that’s what this guide is about.
2. System Requirements & Installation
2.1 Prerequisites
Before running the installer, verify your environment meets the following minimums:
| Requirement | Minimum | Recommended |
|---|---|---|
| Node.js | v22+ | v24+ (use nvm to manage) |
| RAM | 4 GB | 8–16 GB for multi-agent workflows |
| Disk Space | 1 GB free | 5 GB+ for model caches and logs |
| Operating System | macOS 12+, Ubuntu 20.04+, Windows WSL2 | Ubuntu 22.04 LTS on VPS |
| API Key | Anthropic or OpenAI API key | Anthropic (Claude) for best agent quality |
A quick note on API keys: older tutorials recommended using Claude OAuth subscriber tokens. As of early 2026, Anthropic restricted those tokens to Claude Code and claude.ai. For OpenClaw, use a standard Anthropic API key from the official console. Plan for API usage costs – every background task, heartbeat, and reasoning loop consumes tokens, and the bill adds up on a 24/7 always-on agent.
2.2 Installation Method 1: One-Line Script (Recommended for Beginners)
The simplest path for macOS and Linux. The script installs Node.js if missing, then installs and launches OpenClaw’s gateway:
| curl -fsSL https://openclaw.ai/install.sh | bash |
This drops you directly into the onboarding wizard, which walks you through connecting your first AI model and messaging channel. Your workspace is stored at ~/.openclaw on macOS/Linux, or %userprofile%\.openclaw on Windows. The control dashboard runs on http://localhost:18789 by default.
| ⚠️ Sphere Inc. Warning The one-liner script works well on clean machines, but on production Linux servers we’ve seen it fail on permission issues roughly 30% of the time. If you’re deploying on a VPS, use the npm method or Docker instead. |
2.3 Installation Method 2: npm Global Install (Recommended for Developers)
For users comfortable with Node.js tooling, this gives you more control:
| npm install -g openclaw@latest openclaw onboard |
If you hit EACCES permission errors on macOS (this is common), your npm global directory isn’t configured correctly. Fix it once with:
| mkdir ~/.npm-global npm config set prefix ‘~/.npm-global’ echo ‘export PATH=~/.npm-global/bin:$PATH’ >> ~/.bashrc && source ~/.bashrc npm install -g openclaw@latest |
2.4 Installation Method 3: Docker (Recommended for VPS / Production)
If you’re running OpenClaw on a remote server, NAS, or cloud VM – Docker is the right choice. Container isolation prevents agent mishaps from affecting your host system, and it makes updates trivial.
| docker pull openclaw/openclaw:latest docker run -d –name openclaw \ -p 18789:18789 \ -v ~/.openclaw:/root/.openclaw \ openclaw/openclaw:latest |
Mount your ~/.openclaw directory so your config, memory, and session history survive container restarts. This is the setup Sphere Inc. uses for client deployments on Azure and AWS.
2.5 Windows WSL2
Native Windows support exists, but for serious use the better path is WSL2 with Ubuntu. Node.js processes and agent tooling behave more predictably in a Linux environment. Install WSL2 with Ubuntu 22.04, then follow the Linux npm or Docker instructions above.
⚡ How Sphere Inc. Can Help
Sphere Inc. provides managed OpenClaw deployment as part of our AI Infrastructure service. We handle environment provisioning, security hardening, API key management, and ongoing monitoring – so your team gets a production-ready agent without the setup friction. Contact us at sphereinc.ai/deploy.
3. The Core Architecture: Tools vs. Skills
The single most important concept in OpenClaw’s design is the distinction between Tools and Skills. Confuse the two and your configuration will be either dangerously over-permissioned or frustratingly limited. Get it right, and you have a precise, auditable setup.
3.1 Tools: The Capability Layer
Tools are organs. They determine what OpenClaw can physically do. The exec tool lets it run shell commands. The web_search tool lets it query the internet. The browser tool lets it control Chrome. The message tool lets it send messages to Telegram, Slack, Discord, and more.
Without a Tool enabled, the corresponding capability simply doesn’t exist – regardless of what Skills are installed or what the AI reasons about. Tools are the real permission system. This is the most critical thing to understand about OpenClaw security.
3.2 Skills: The Knowledge Layer
Skills are textbooks. They teach OpenClaw how to combine its Tools to accomplish specific tasks. The gog Skill teaches it how to interact with Google Workspace – Gmail, Calendar, Drive, Tasks. The github Skill teaches it how to use the gh CLI to manage repositories. The obsidian Skill teaches it how to organize notes.
Does installing a Skill give OpenClaw new permissions? No. Installing the obsidian Skill doesn’t mean OpenClaw can suddenly write to your vault if the write Tool is disabled. Skills are instructions. Tools are what make those instructions executable.
| The Three Conditions for a Skill to Actually Work 1. Tool enabled: OpenClaw must have the underlying capability (e.g., exec to run CLI tools). 2. Bridge installed: The companion CLI tool must be present on the machine (e.g., the gh CLI for the github Skill). 3. Authorized: The relevant service must have granted access (e.g., OAuth for Google, a bot token for Telegram). All three are required. Miss one and nothing works. |
3.3 The Concentric Circle Model
We find it useful to think of OpenClaw’s architecture in three layers:
- Layer 1 – Core Capabilities (8 Tools): File access, command execution, web access. Nearly everyone needs all eight of these.
- Layer 2 – Advanced Capabilities (18 Tools): Browser control, memory, multi-session orchestration, messaging, automation. Enable selectively based on actual use cases.
- Layer 3 – Knowledge Layer (53 Skills): Instructions for working with Google, GitHub, Slack, Obsidian, and more. Install only what you actively use.
OpenClaw also connects to ClawHub, a skill registry with over 13,700 third-party Skills. Since February 2026, ClawHub integrates VirusTotal scanning to flag malicious downloads – but you should still review any third-party Skill’s GitHub repo before installing, especially if it requests exec or message access.
4. Layer 1: The 8 Core Tools (Enable All of These)
These eight Tools are the foundation. Without them, OpenClaw can't do much of anything useful. Below we cover each one, its purpose, its risk level, and our recommended configuration.
4.1 File Operations: read, write, edit, apply_patch
- read – Read-only access to the filesystem. Low risk. Enable without hesitation.
- write – Creates new files and overwrites existing ones. Medium risk. Enable, but be aware it can overwrite files without confirmation.
- edit – Performs structured, diff-style edits to existing files. Medium risk. Safer than write for modifying existing content.
- apply_patch – Applies code patches in unified diff format. Medium risk. Essential for development workflows.
All four file tools work together and are typically enabled as a group using the group:fs shorthand in your config.
4.2 Execution: exec, process
exec is the most powerful and most dangerous Tool in OpenClaw. It lets the agent run any shell command – install packages, execute scripts, delete files, manage services. Without exec, most real automation workflows fail immediately. With exec and no guardrails, you've granted the AI root access to your machine.
This is why Sphere Inc. always configures exec with approval mode enabled in every deployment we manage:
| "approvals": |
With approval enabled, every command is shown to you before it runs. You confirm or reject it. Yes, it adds friction. That friction is the entire point – it's your last line of defense against a misjudged action or a Prompt Injection attack from malicious content the agent reads.
process manages background processes – list running tasks, check output, kill stuck processes. Enable alongside exec.
4.3 Web Access: web_search, web_fetch
web_search performs keyword searches (like Google). web_fetch retrieves full web page content. Together they give OpenClaw the ability to research the internet – essential for morning briefings, competitor monitoring, content research, and staying current on any domain.
Risk is medium for web_fetch, primarily because fetching arbitrary URLs is an attack vector for Prompt Injection – malicious instructions embedded in web content. OpenClaw has mitigations, but the risk doesn't disappear. Be selective about what URLs you ask the agent to fetch from untrusted sources.
| ⚡ How Sphere Inc. Can Help Sphere Inc.’s AI consulting team configures approval workflows and secure exec policies as part of every OpenClaw deployment. We’ve developed a hardened baseline config that enables all 8 core tools with sensible defaults – available to clients as part of our AI Security Audit service. |
5. Layer 2: The 18 Advanced Tools (Enable Selectively)
Layer 2 is where OpenClaw transforms from a command executor into a genuine AI assistant. These tools add memory, browser control, scheduled automation, and cross-device capability. Each one extends what OpenClaw can do – and each one expands the attack surface. Apply the rule: if you can’t articulate a use case, leave it off.
5.1 Browser: browser, canvas, image
browser gives OpenClaw control over Chrome via Chrome DevTools Protocol – click buttons, fill forms, navigate pages, take screenshots. The v2026.3.13 release added a CDP attach mode that connects to your already-signed-in Chrome session without needing extensions.
Recommended use cases: price research, spec comparison, reading paywalled content, form filling. Critical caveat: never let the agent reach the checkout or payment stage of any transaction. Keep the irreversible ‘last mile’ actions manual. Configure it to fill carts, not empty wallets.
canvas provides a visual whiteboard workspace for diagrams and structured visual output. image enables vision capabilities – the agent can analyze and reason about image content. Both are low risk.
5.2 Memory: memory_search, memory_get
These tools enable cross-session memory – OpenClaw remembers information from previous conversations. After consistent use, it learns your tech stack, preferences, recurring tasks, and communication style. You stop repeating context every session. The agent becomes progressively more useful the longer you use it.
Risk is medium. Memory content persists on your machine and informs future behavior. Make sure sensitive information you wouldn’t want cached – passwords, personal health details, private financial specifics – is explicitly excluded from memory via your config.
5.3 Multi-Session Orchestration: sessions_list, sessions_history, sessions_send, sessions_spawn, session_status, sessions_yield, subagents
This tool group enables parallel agent sessions and multi-agent coordination. You can run one session discussing a product strategy while another researches competitors – without cross-contamination.
sessions_send and sessions_spawn enable inter-session communication and the spawning of sub-agents for delegated tasks. sessions_yield pauses a parent agent to wait for sub-agent results. subagents monitors and manages running sub-agents.
For single-instance personal setups, the basic sessions tools (list, history, status) are useful. The orchestration tools (spawn, yield, subagents) are for multi-agent workflows – enable them when you need them.
5.4 Messaging: message
The message Tool lets OpenClaw send messages to Discord, Slack, Telegram, WhatsApp, and iMessage. This is what turns cron automation into a push-notification system – rather than polling for results, the agent finds you.
| Sphere Inc. Security Rule Configure message to send only to yourself. Never configure it to send messages to other people on your behalf without a human review step. Messages sent in your name by AI can’t be unsent – and if the agent misreads context, uses the wrong tone, or gets manipulated by a Prompt Injection attack, you bear the reputational consequences. Keep the human-in-the-loop for external communication. |
5.5 Automation: cron, gateway
cron is the scheduling engine. It runs any prompt on a time-based schedule – every morning, every hour, every Monday. gateway lets OpenClaw restart itself and manage process lifecycle.
Together, cron and message are what makes OpenClaw feel like infrastructure rather than a chatbot. Schedule a task. Define what to do. Tell it where to send results. The agent runs while you sleep.
5.6 Voice: tts
text-to-speech converts OpenClaw’s replies into voice messages. On Telegram, they appear as round voice note bubbles. Supports ElevenLabs (highest quality), OpenAI TTS, and Edge TTS (free). Low risk. Enable if voice output adds value to your workflow.
5.7 Hardware: nodes
Cross-device hardware control – remote screenshots, GPS location, camera access on paired iOS and Android devices. High risk. Most users don’t need the agent independently accessing their camera. Disabled in all Sphere Inc. baseline configs.
5.8 Workflow Engine: llm_task, lobster
lobster is a multi-step workflow definition engine. llm_task inserts AI reasoning steps into those workflows. Only relevant if you’re building complex automated pipelines. If you don’t have a specific workflow engine use case, disable both.
6. Layer 3: The 53 Official Skills – What to Install
53 bundled Skills sounds overwhelming. In practice, most people need 5–12. The rest exist for use cases that simply don’t apply to you.
Critical default behavior: bundled Skills auto-load if the corresponding CLI tool is installed on the system. This means Skills you haven’t consciously activated may already be running. Use the allowBundled whitelist in your config to explicitly control which Skills are active.
6.1 Notes & Knowledge Management
| Skill | Platform | Works On VM? | Our Take |
|---|---|---|---|
| obsidian | Obsidian (local files) | Only if vault is local to agent | Best if OpenClaw is on same machine as vault |
| notion | Notion (cloud API) | Yes – cloud-based | Best choice for VM deployments |
| apple-notes | Apple Notes | No – macOS only | Local only; not for server setups |
| bear-notes | Bear | No – macOS only | Local only; not for server setups |
6.2 Productivity & Email
gog is the most comprehensive productivity Skill in the bundle. It integrates the full Google Workspace suite – Gmail, Google Calendar, Google Drive, Google Docs, Sheets, and Tasks – through the official OAuth flow. You can revoke access any time through your Google Account security settings.
himalaya offers basic IMAP/SMTP email access (send and receive), but lacks calendar and Drive integration. If you’re on Google, gog is strictly better. himalaya is the right choice if you need to connect a non-Google email account.
- things-mac – Things 3 task management. macOS only.
- apple-reminders – Apple Reminders. macOS only.
- trello – Trello board management. Cloud-based, works on any OS.
6.3 Messaging & Social Media (Handle With Caution)
wacli (WhatsApp), imsg (iMessage), bird (X/Twitter), slack, and discord provide deep platform integration – not just sending messages, but searching message history, syncing conversations, and managing channels. Risk rating on WhatsApp and iMessage is Very High, because full account access is a significant trust grant.
| Sphere Inc. Recommendation Install messaging Skills only if you have an explicit automation use case (e.g., a support triage bot, a status notification system). Don’t install them speculatively. And again: keep external communication behind a human review step, even when the integration is configured. |
6.4 Developer Tools
| Skill | Function | Risk | Sphere Recommendation |
|---|---|---|---|
| github | GitHub via gh CLI – PRs, issues, Actions, repos | Medium | Install for any dev workflow |
| tmux | Multi-pane terminal session management | Low | Install for server workflows |
| session-logs | Search and analyze past OpenClaw conversation logs | Low | Useful for debugging and audit |
| coding-agent | Dispatch coding tasks to Claude Code, Cursor, etc. | Medium | Advanced – see note below |
The coding-agent Skill opens the door to AI-orchestrating-AI workflows. Install Claude Code on your agent’s machine, connect it via coding-agent, and you can dispatch entire coding tasks via Telegram: ‘Clone this repo, study the architecture, and build a demo page.’ OpenClaw passes the task to Claude Code, which executes autonomously and pushes a notification when done. This is one of the most powerful patterns we’ve seen at Sphere Inc.
6.5 Security: 1password
The 1Password Skill gives OpenClaw access to your password vault – look up credentials, auto-login, fill forms. The critical limitation: the permission model is all-or-nothing. Once authorized, the agent has access to your entire vault. There’s no way to limit it to specific entries.
Our recommendation: don’t install this Skill for your primary vault. If you need AI-accessible credentials, create a separate, minimal vault containing only passwords you’re explicitly comfortable sharing with an AI agent.
6.6 All 53 Skills – Quick Reference
| Category | Skill | Risk Level |
|---|---|---|
| Notes | obsidian, notion, apple-notes, bear-notes | Low |
| Tasks | things-mac, apple-reminders, trello | Low–Medium |
| Work/Email | gog (Google Workspace), himalaya (IMAP) | Medium–High |
| Chat | slack, discord, wacli, imsg, bluebubbles | Medium–Very High |
| Social | bird (X/Twitter) | Very High |
| Developer | github, coding-agent, tmux, session-logs | Low–Medium |
| Music | spotify-player, sonoscli, blucli | Low |
| Smart Home | openhue, eightctl | Low |
| Food | food-order, ordercli | High |
| Creative | openai-image-gen, nano-banana-pro, video-frames, gifgrep | Low |
| Voice/Audio | sag, openai-whisper, openai-whisper-api, sherpa-onnx-tts | Low |
| Security | 1password | Very High |
| AI Integration | gemini, oracle, mcporter | Low–Medium |
| System | clawhub, skill-creator, healthcheck, summarize, weather | Low |
| Places | goplaces, local-places | Low |
| Media | camsnap, nano-pdf, blogwatcher, model-usage | Low–Medium |
| Comms | voice-call, peekaboo, canvas, songsee | Low–High |
7. The Production Config: A Complete openclaw.json Template
Below is Sphere Inc.'s recommended starting configuration – the baseline we use for individual power users and small team deployments. Modify it to fit your use case, but use it as your starting point rather than the defaults.
| "tools": { "allow": [ "read", "write", "edit", "apply_patch", "exec", "process", "web_search", "web_fetch", "browser", "image", "memory_search", "memory_get", "sessions_list", "sessions_history", "session_status", "tts", "message", "cron", "gateway" ], "deny": ["nodes", "canvas", "llm_task", "lobster", "sessions_send", "sessions_spawn", "sessions_yield", "subagents"] , "approvals": "exec": , "skills": "allowBundled": [ "gog", "github", "tmux", "session-logs", "weather", "summarize", "clawhub", "healthcheck" ] , "channels": "telegram": "allowFrom": ["+YOUR_PHONE_NUMBER"], "groups": |
Key decisions in this config:
- 22 tools enabled, 6 disabled – nodes (no use case), canvas (not needed), llm_task/lobster (no workflow engine), sessions_spawn/yield/subagents (not yet using multi-agent orchestration).
- exec approval enabled – every command requires confirmation before running.
- allowBundled whitelist – only 8 Skills active, preventing auto-loaded Skills from running silently.
- Telegram locked to your number – prevents unauthorized users from sending commands to your agent.
8. Automating Your Workflows With cron + message
This is where OpenClaw stops being a chatbot and starts being infrastructure. The pattern is always the same: trigger → action → deliver. Define when it runs, what it does, and where results go.
Here are the automations Sphere Inc. clients use most frequently:
8.1 Daily Brief
Every morning at a scheduled time, OpenClaw checks your calendar for the day’s events, scans your inbox for unread messages that need replies, fetches the weather, and checks overnight CI/CD status. The compiled briefing arrives as a Telegram message before you’ve had your first coffee.
This single automation reliably replaces the habit of checking five different apps in the first 20 minutes of the day. Sphere Inc. clients report saving 20–30 minutes of morning fragmentation daily.
8.2 Email Triage
Twice a day, OpenClaw scans the inbox, categorizes messages by urgency, archives newsletters, and sends a summary of action-required items with one-line context for each. What was 30 minutes of inbox management becomes 5 minutes of reviewing a digest.
8.3 CI/CD Monitoring
When a GitHub Actions workflow fails, OpenClaw reads the error log, identifies the probable cause, and pushes a Telegram message with the diagnosis. Production issues can be triaged from a phone while away from a desk.
8.4 Content & Research Digest
OpenClaw monitors specific RSS feeds, Hacker News threads, and subreddits for topics you care about. It compiles a daily digest of signal worth paying attention to – not writing for you, but surfacing what’s worth writing about.
8.5 Meeting Prep
Before a calendar event, OpenClaw pulls relevant emails, documents, and notes related to the meeting participants or topic, and sends you a briefing 15 minutes before the event starts.
| The Implementation Pattern Every automation is a cron entry that triggers a prompt. The prompt tells OpenClaw which tools to use and where to send results. The hard part isn’t the config syntax – it’s identifying which daily friction points are worth automating. Start with one that saves you the most time, get it working reliably, then add more. Don’t try to automate everything at once. |
9. Security: The Decisions You Can’t Skip
OpenClaw is powerful because it can execute actions on your machine, on your behalf. That power is also the attack surface. Here are the five security decisions you need to make before going live.
9.1 Always Enable exec Approval
Already covered above – but it bears repeating. No production deployment of OpenClaw should have exec running without approval mode. This is non-negotiable in every Sphere Inc. configuration.
9.2 Restrict Messaging Channels
Lock your Telegram (or WhatsApp, or iMessage) integration to your own phone number using allowFrom. Without this, anyone who knows your bot token can send commands to your agent. In group channels, require explicit @mention to trigger the agent.
9.3 Use allowBundled Whitelist
Bundled Skills auto-activate when the corresponding CLI tool is installed. Without an explicit whitelist, Skills you haven’t thought about may be running quietly in the background. Use allowBundled to enumerate exactly which Skills are active.
9.4 Keep the Last Mile Manual
For any irreversible action – sending a message to someone external, making a purchase, posting publicly, deleting files you can’t recover – keep a human in the loop. OpenClaw can draft, prepare, and stage these actions. The confirmation step stays with you.
9.5 Give the Agent Its Own Accounts
The most thoughtful OpenClaw operators treat their agent like a new employee: they set up separate email addresses, storage buckets, and service accounts rather than handing over their personal credentials. Security risk is proportional to access granted. Limit what the agent can reach, and you limit the blast radius of any mistake.
10. Troubleshooting Common Installation Issues
| Problem | Likely Cause | Fix |
|---|---|---|
| EACCES permission error on npm install | npm global directory permissions | Reconfigure npm prefix (see Section 2.3) |
| One-liner script fails on Ubuntu VPS | Permission/environment issues | Use npm or Docker method instead |
| ‘Model ID not allowed’ error | OAuth tokens deprecated | Switch to standard Anthropic API key |
| Agent not responding on Telegram | Bot token not configured or gateway not running | Check openclaw logs –follow for gateway status |
| Skills auto-loading you didn’t expect | Bundled Skills default-on behavior | Add allowBundled whitelist to config |
| exec commands running without approval | Approval mode not configured | Add approvals.exec block to openclaw.json |
| Agent loses memory between sessions | memory_search/memory_get not enabled | Add both to tools.allow |
| Gateway not accessible on remote server | Port 18789 not exposed or firewall blocking | Open port or use Tailscale for secure remote access |
For any issue not covered here, run openclaw doctor – it surfaces misconfigured settings and risky policies automatically.
11. Getting Started: The Sphere Inc. Recommended Path
You don’t need all 26 Tools on day one. You don’t need all 53 Skills. Start minimal, prove value, then expand. Here’s the path we walk clients through:
- Install via your preferred method and complete the onboarding wizard.
- Connect a single messaging channel – Telegram is fastest to set up.
- Apply the baseline config from Section 7. Lock down exec and restrict Telegram to your number.
- Enable only the Skills you actively use today. Use allowBundled.
- Build one automation – a daily brief is a great first one. Get it running reliably.
- Run openclaw doctor and review any warnings before adding more capabilities.
- Expand deliberately – add tools and Skills as you identify real use cases, not speculatively.
The most common mistake new OpenClaw users make is enabling everything at once to see what it can do. The second most common mistake is never enabling approval on exec. Don’t make either mistake.
How Sphere Inc. Supports Your OpenClaw Journey
Sphere Inc. is an AI implementation consultancy specializing in agent infrastructure, workflow automation, and AI security. We work with individual power users, development teams, and enterprise organizations to design, deploy, and maintain AI systems that deliver real operational value.
| Service | What’s Included | Best For |
|---|---|---|
| Managed Deployment | Environment provisioning, hardened config, API setup, monitoring | Teams without dedicated DevOps |
| Workflow Automation Design | Custom cron+message automations mapped to your workflows | Operations and productivity use cases |
| AI Security Audit | Config review, approval policy testing, remediation report | Regulated industries and sensitive data environments |
| AI Coding Integration | coding-agent setup with Claude Code, CI/CD monitoring, GitHub automation | Engineering teams |
| OpenClaw Training | Half-day workshops for teams: setup, config, security, automation patterns | Organizations rolling out OpenClaw at scale |
If you’re evaluating OpenClaw for enterprise deployment, we offer a complimentary 30-minute architecture review to assess fit and surface any configuration risks specific to your environment.
