OpenClaw Complete Guide 2026: The Open-Source Personal AI That Actually Does Things

1. What Is OpenClaw?

OpenClaw is a free, open-source personal AI assistant framework built by Peter Steinberger and a growing community of contributors. It was first released in late 2025 under the names Clawd and Moltbot before being rebranded OpenClaw in January 2026. The GitHub repository lives at github.com/openclaw/openclaw and is published under an open-source license.

The project's tagline — "THE AI THAT ACTUALLY DOES THINGS" — captures what sets it apart from conventional AI chat tools. OpenClaw is not a wrapper around a chatbot API. It is a persistent, always-on agent that:

• Runs as a background process (the Gateway) on your own hardware
• Accepts messages from chat apps you already use — WhatsApp, Telegram, Discord, Signal, iMessage, Slack
• Maintains a persistent memory across all conversations, so it never forgets context
• Can take actions: read and write files, run shell commands, control browsers, send emails, manage calendars, trigger webhooks, and more
• Can create and install its own new skills by writing code during a conversation
• Runs scheduled background tasks (heartbeats) without waiting for you to initiate a chat

Users describe the first experience of setting it up as a category-defining moment — comparable to the first time they used ChatGPT or saw a smartphone. Andrej Karpathy (formerly of OpenAI/Tesla) tweeted: "Excellent reading thank you. Love oracle and Claw." Dave Morin wrote: "At this point I don't even know what to call OpenClaw. It is something new. After a few weeks in with it, this is the first time I have felt like I am living in the future since the launch of ChatGPT."

Why it matters in 2026

The year 2026 is seeing a bifurcation in how people use AI. On one side are cloud products (ChatGPT, Gemini, Claude chat interfaces) where your data is processed on someone else's servers and context resets with every session. On the other side is a growing movement of self-hosted, privacy-first AI tools that run locally, respect data ownership, and can be infinitely extended. OpenClaw is the most accessible and fully-featured example of this second category — it is deliberately designed so that a non-engineer can install it in five minutes and start using it from their phone without writing a single line of code.

2. Architecture: How the Gateway Works

Understanding OpenClaw's architecture helps explain both its power and its requirements.

The Gateway

The central component is the Gateway: a Node.js process that runs persistently on your machine (Mac, Windows, or Linux) and listens on port 18789. It performs three jobs simultaneously:

1. Model routing — It receives messages, forwards them to your configured LLM provider (Anthropic, OpenAI, Google Gemini, local model, etc.) with the full system context and memory, and returns the response.
2. Channel management — It bridges your chat channels (Telegram bot, WhatsApp connection, Discord bot, etc.) to the AI, routing messages in and responses out.
3. Tool execution — When the AI decides to use a tool (read a file, run a shell command, browse a URL, call an API), the Gateway executes the tool on the local machine and feeds the result back to the model.

The system prompt and memory

OpenClaw maintains a persona file — a structured system prompt that describes who the AI is, what it knows about you, and what skills it has access to. This file lives on your machine and evolves over time as you interact. The AI reads it at the start of every conversation, which is what gives the persistent-memory experience: the assistant doesn't "forget" between sessions because your context is always present in the prompt.

This is distinct from how chat sessions work in cloud AI products, where each conversation typically starts from scratch. With OpenClaw, the AI literally reads a growing file about you before responding — it's closer to how a human employee builds institutional knowledge over time.

Skills as modules

Capabilities are implemented as skills: TypeScript/JavaScript modules that expose tools the AI can call. Each skill is a self-contained file that describes its name, what it does, and what API it exposes. The AI can list available skills, call them, and — remarkably — write new skills during a conversation. If you say "create a skill that checks my Todoist tasks and messages me every morning," OpenClaw will write the code, save it to disk, and hot-reload it without any restart required.

Platform requirements

3. Installation: Up and Running in 5 Minutes

OpenClaw installs with a single command. The installer handles Node.js, dependencies, and basic configuration automatically.

Step 1 — Run the installer

macOS / Linux:

curl -fsSL https://openclaw.ai/install.sh | bash

Windows (PowerShell):

powershell -c "irm https://openclaw.ai/install.ps1 | iex"

The one-liner downloads the installer, installs Node.js if not already present, installs the OpenClaw npm package globally, and sets up the configuration directory.

Step 2 — Run the onboarding wizard

openclaw onboard --install-daemon

The wizard walks you through:

• Choosing an AI model provider (Anthropic Claude, OpenAI, Google Gemini, or a local model)
• Entering your API key — stored locally on your machine, never sent to OpenClaw servers
• Naming your AI assistant and setting its persona
• Installing the Gateway as a background daemon so it starts automatically at system boot

Step 3 — Verify the Gateway is running

openclaw gateway status

You should see the Gateway listening on port 18789. The onboarding takes about two minutes from start to finish.

Step 4 — Open the control dashboard

openclaw dashboard

This opens the control UI in your default browser. You can immediately type in the chat panel and get an AI response — your assistant is working. From here, you can install skills, view memory, configure channels, and monitor background tasks.

Alternative install methods

For power users, OpenClaw also supports installation via:

• npm: npm install -g openclaw — useful for CI/CD pipelines and server deployments
• Docker: official Docker images are available for containerized deployments on VMs or home servers
• Nix: a Nix flake is available for reproducible environments
• "Hackable" install: clones the repository directly so you can modify the source code — recommended for developers who want to contribute or deeply customize behavior

4. Connecting Chat Channels

OpenClaw's most immediately useful feature is that you can reach your AI assistant from the messaging apps already on your phone — no new app to install, no new interface to learn.

Telegram (recommended for beginners)

Telegram is the easiest channel to set up and the most popular among new OpenClaw users. You create a Telegram bot via BotFather (a few minutes), paste the bot token into OpenClaw's config, and immediately start messaging your assistant via the Telegram app. The assistant responds in the same chat thread it receives messages in — DMs, group chats, or channels all work.

WhatsApp

WhatsApp integration is provided via the Baileys library (QR-code pairing). You scan a QR code with your WhatsApp app once, and OpenClaw connects as a linked device. Many users report that WhatsApp feels the most natural because it's the messaging app they're already in all day.

Discord

The Discord skill connects OpenClaw to a server you control. This is particularly popular for teams and multi-user setups: you can invite several people to a private Discord server and let them interact with the same OpenClaw instance.

All supported channels at a glance

You can run multiple channels simultaneously on the same OpenClaw instance. Some users have OpenClaw accessible via WhatsApp for personal use and Discord for team use, both routing to the same memory and skills.

5. Supported AI Models

One of OpenClaw's design principles is model-agnosticism: you bring your own API key, and you can switch models at any time. "Your keys, your choice."

Most active users recommend starting with Anthropic Claude (Sonnet or Haiku tier) for the best balance of capability and cost, but MiniMax M2.5 has become popular for always-on deployments because of its favorable pricing at high request volumes.

6. Core Capabilities Deep Dive

6.1 Persistent Memory

Memory is the feature users most consistently highlight as transformative. Every interaction updates the persona file and memory stores. The assistant knows your timezone, your preferred tone, your ongoing projects, your relationships, and the ongoing state of long-running tasks. Memory is stored locally on your machine in plain files — not in any cloud database — so you own it completely and can read, edit, or backup it manually.

The memory system includes:

• Long-term memory: facts about you that persist indefinitely (your name, preferences, regular commitments)
• Session memory: the state of the current conversation
• Task memory: ongoing tasks and their status, so the AI can pick up mid-task after an interruption
• Cross-channel memory: insights from a WhatsApp conversation are available in the Telegram conversation — it's all one agent

6.2 Browser Control

OpenClaw can drive a real Chromium browser using Playwright. This means it can:

• Navigate to any URL and extract information
• Fill in forms and submit them (travel check-ins, online orders, booking systems)
• Log in to websites using credentials from your local password manager (1Password skill)
• Take screenshots and send them back to you in chat
• Automatically provision API keys — one user reported that their OpenClaw "opened the Google Cloud Console and provisioned a new OAuth token" autonomously when it realized it needed one

The browser runs on your machine, which means it can access sites that require local cookies, corporate VPN authentication, or session state from your regular browser profile.

6.3 Full System Access

OpenClaw has optional full shell access to the machine it runs on. You can scope it to sandboxed directories or grant full root/admin access — the choice is yours. With system access enabled, you can:

• Read and write any file (useful for editing documents, configs, codebases)
• Run any shell command or script
• Trigger other CLI tools (git, docker, ffmpeg, etc.)
• Restart services, modify system settings, or execute backups

One power user told their OpenClaw to "turn off the PC" via Telegram — it executed a clean shutdown and turned itself off in the process.

6.4 Canvas: Visual Workspace

The Canvas feature, available through companion apps on macOS and iOS, provides a visual workspace where OpenClaw can interact with your screen using an A2UI (AI-to-UI) interface. It can see what's on your screen, interpret UI elements, and interact with applications in ways that go beyond command-line access.

6.5 Voice: Talk Mode

The Voice skill adds wake-word detection and audio input. You can speak to your OpenClaw assistant by name, and it will respond — either in text (in your chat app) or in synthesized speech via ElevenLabs integration. One user reported their OpenClaw "called my phone and spoke to me with an Aussie accent from ElevenLabs."

7. Skills System and ClawHub

Skills are the extensibility layer that transforms OpenClaw from a capable base into a practically unlimited automation platform.

What a skill is

A skill is a JavaScript/TypeScript module that exposes one or more tool functions to the AI. Each function has a name, description, input schema, and implementation. When the AI receives a message and decides it needs to use a skill, it calls the function, receives the result, and incorporates it into its response.

Example skills and what they do:

ClawHub: the community skill registry

ClawHub is the public skill registry for OpenClaw — similar to npm for Node.js packages, but specifically for AI agent skills. As of March 2026, ClawHub hosts over 100 community-built skills ranging from grocery shopping automation (Tesco Autopilot) to 3D printer control (Bambu), health data integration (Oura Ring, WHOOP), and food ordering (Foodora).

Installing a skill from ClawHub is as simple as telling your AI: "Install the Obsidian skill from ClawHub." OpenClaw will download, verify, and hot-reload the skill — no terminal required.

Security: VirusTotal partnership

In February 2026, OpenClaw announced a partnership with VirusTotal to scan all skills published on ClawHub using VirusTotal's threat intelligence platform. Every skill submission is automatically scanned against 70+ antivirus engines before appearing in the public registry. This addresses a real concern with community-written code that executes with shell access on your machine.

Self-writing skills

One of the most remarkable capabilities is that OpenClaw can write its own skills mid-conversation. Tell it: "I need a skill that monitors my AWS costs and messages me if they exceed $50 in a day." It will write the TypeScript code, create the skill file, and activate it — all within the chat thread. This self-extensibility is what separates it from rigid assistant platforms and is a key reason the community describes it as feeling like "early AGI."

8. Smart Home and IoT Integration

OpenClaw's system and browser access makes it a natural hub for smart home automation that understands natural language and context — not just keyword triggers.

Philips Hue

The OpenHue skill connects to your Hue Bridge on the local network. You can say "turn on the living room lights at 50% warmth" or "set a movie scene" — the assistant translates natural language into Hue API calls. Because OpenClaw has memory, you can also say "do the same thing you did last Friday evening" and it will recall the scene it set.

Home Assistant

The Home Assistant skill gives OpenClaw access to every device managed by your HA installation — thermostats, locks, garage doors, sensors, vacuum robots, and anything connected via HA's ecosystem. Combined with cron scheduling and memory, you can build routines like: "every weekday morning at 7am, turn on the kitchen lights, start the coffee maker, and tell me today's weather and calendar."

8Sleep, Sonos, and health devices

Community skills cover an expanding range of IoT devices: the 8Sleep smart mattress (tracking sleep and adjusting temperature), Sonos multi-room audio, WHOOP health tracker, and Oura Ring. One user set their OpenClaw to fetch WHOOP biomarker data automatically and send a morning readiness report before their first meeting.

9. Proactive Behavior: Heartbeats & Cron

Most AI tools are purely reactive — they wait for you to say something before doing anything. OpenClaw can be proactive: it reaches out to you, performs background tasks, and acts on time-based or event-based triggers without you initiating a conversation.

Heartbeats

A heartbeat is a periodic check-in that OpenClaw performs autonomously. You configure the frequency (daily, hourly, etc.) and what the AI should do during each heartbeat. Common heartbeat workflows:

• Morning briefing: At 7am, summarize today's calendar, the weather forecast, unread emails, and any news the agent has been tracking — sent as a single message to your phone
• Travel reminders: Check traffic 45 minutes before a calendar event that requires driving, and alert you if you need to leave early
• Weekly digest: Every Sunday, compile a summary of the week's completed tasks and a preview of the next week
• Background context updates: One user's OpenClaw "wrote a document connecting two completely unrelated conversations from different comms channels" — synthesizing insights without being asked

Cron scheduling

The Cron skill exposes standard cron-syntax scheduling to the AI and to you via chat. You can say "remind me to take my medication at 9pm every evening" or "run the nightly backup script at 3am on weekdays," and OpenClaw will configure the cron job without you touching a terminal. Cron jobs are stored locally and survive reboots because the Gateway runs as a system daemon.

Webhook triggers

External services can trigger OpenClaw via webhooks. A practical example: connect your CI/CD pipeline to send a webhook when tests fail, and OpenClaw will inspect the logs, diagnose the failure, and message you with a root-cause summary and proposed fix.

10. Privacy and Security

Privacy is one of OpenClaw's strongest value propositions — especially compared to cloud-based AI services.

What stays on your machine

• All memory and context: The persona file and memory stores are plain files in a local directory. Nothing is stored in any cloud database.
• All API keys: Your AI provider keys, OAuth tokens, and credentials are stored in a local config file. They're never sent to openclaw.ai or any third party.
• All skill code: Skills run locally. ClawHub hosts the code for distribution, but execution always happens on your machine.
• Conversation logs: If logging is enabled, logs are stored locally. You can disable logging entirely.

What leaves your machine

• Messages to AI providers: Your prompts and memory context are sent to whichever AI API you've configured (Anthropic, OpenAI, etc.) under their privacy policies. This is unavoidable when using cloud models. To avoid this entirely, use a local model via Ollama.
• Chat channel messages: Messages travel through Telegram, WhatsApp, or whatever channel you use — subject to those platforms' privacy policies. Signal is the highest-privacy option.
• Skill API calls: When a skill calls an external API (Spotify, GitHub, etc.), that traffic goes to the respective service as normal.

Open-source auditability

Because the entire codebase is on GitHub, anyone can audit what the Gateway does. There's no black box. Security researchers can — and do — review the code and submit issues. The VirusTotal partnership for ClawHub skills adds an additional layer of malware scanning for community-contributed code.

Running fully offline

For maximum privacy, OpenClaw can be configured to use a local model (Llama 3, DeepSeek, Gemma, Mistral, etc.) via an Ollama-compatible endpoint. In this configuration, no data ever leaves your network. Performance depends on your hardware — a Mac Mini M4 Pro or a Linux box with a mid-range GPU can run Llama 3.3-70B at useful speeds for most daily assistant tasks.

11. Real-World Use Cases

11.1 Personal Life Management

Users consistently report using OpenClaw to offload the cognitive overhead of life administration. Common patterns: managing email subscriptions ("get it to unsubscribe from a whole bunch of emails I don't want"), health insurance claims, doctor appointment searches, daily habit tracking via WHOOP or Oura, morning briefings, note-taking in Obsidian, and shopping lists. One user used their OpenClaw to file a health insurance reimbursement dispute — the AI handled the entire email thread and prompted the insurer to reopen an investigation.

11.2 Software Development Assistance

Developers are major OpenClaw users. Common patterns: monitoring CI/CD pipelines via webhook triggers and auto-diagnosing failures; running Claude Code or OpenAI Codex sessions from a phone while away from the desk; autonomously running test suites and opening PRs when tests pass ("running tests on my app and capturing errors through a Sentry webhook then resolving them and opening PRs"); reviewing code at a high level and summarizing what needs attention; and managing GitHub issues and pull requests from a chat message.

11.3 Small Business Operations

Several users describe using OpenClaw effectively as a business operations layer. It can monitor business email, manage contractor communications, track project statuses across tools, generate weekly reports, and manage social media scheduling. One user added: "It's running my company."

11.4 Research and Information Management

OpenClaw works well as a research assistant that builds an evolving knowledge base. Upload YouTube videos via a skill and turn them into reusable agent workflows; feed it documents and ask it to synthesize insights across sources; combine it with Obsidian for a "second brain" that updates as you chat. This is complementary to (but different from) tools like Google NotebookLM — OpenClaw acts on the knowledge, not just summarizes it.

11.5 Creative Projects

Creative users have used OpenClaw to generate custom AI meditations with TTS and ambient audio, build personal content pipelines for newsletters and social media, automatically generate Sora videos with custom workflows, and create personalized Stumbleupon-style article discovery tools. One user built a flight search tool from scratch by asking their OpenClaw to "write a terminal CLI with multi-provider flight search" — it did so in a single session.

11.6 Family and Household Use

Because OpenClaw runs over standard messaging apps, it's accessible to non-technical household members. One user set their family up on a shared Discord where everyone can ask the same OpenClaw instance questions, request reminders, or control smart home devices — creating what amounts to a family AI concierge.

12. OpenClaw vs. Commercial AI Assistants

The trade-offs are real: OpenClaw requires technical setup (about 5 minutes, but still more than saying "Hey Siri"), it requires a machine to be running, and the quality of responses depends on which LLM you configure. But for users who value privacy, extensibility, and actual task execution — rather than just answers — there is no commercial equivalent.

13. Limitations and Who It Is For

Current limitations

Who OpenClaw is ideal for

• Developers and technical users who want an always-on AI that can interact with their development environment, codebase, and cloud infrastructure from anywhere
• Privacy-conscious professionals who need AI capabilities but don't want sensitive data processed on third-party servers
• Self-hosters and tinkerers who enjoy building and extending tools and want full control over their AI assistant's behavior
• Productivity enthusiasts who want to automate repetitive life administration across multiple apps and devices
• Small teams that want a shared AI assistant accessible via their existing team chat without paying per-seat SaaS pricing

Who should probably wait

• Users who need a polished, zero-setup consumer experience — commercial assistants like Siri or Google Assistant are still simpler to get started
• Users who need enterprise-grade governance, audit trails, and SSO out of the box — those features are early or absent
• Users without a machine to dedicate to running the Gateway (a Raspberry Pi is sufficient, but requires setup)

14. Frequently Asked Questions

Is OpenClaw free?

The OpenClaw software itself is free and open source. You pay for the AI model API you connect to (e.g., Anthropic, OpenAI). MiniMax M2.5 and DeepSeek are among the most cost-effective options; local models via Ollama cost nothing per query. Typical monthly API spend for moderate daily use is $5–30.

Do I need to be a developer to use OpenClaw?

The one-liner installer and onboarding wizard are designed for non-developers. You need to be comfortable opening a terminal and pasting a command. Installing skills and configuring channels is done through chat conversations, not code. That said, the experience is significantly better if you're comfortable with basic command-line tools.

Does OpenClaw require a GPU?

No. The Gateway is a Node.js process that makes API calls to AI providers — it runs on any CPU. A GPU is only needed if you want to run a local LLM model via Ollama as the AI backend, which is optional.

Can I run OpenClaw on a Raspberry Pi?

Yes. Several users run OpenClaw 24/7 on a Raspberry Pi 4 or 5. It works well when connected to cloud AI models (the Pi doesn't need to run the model itself). A Pi 5 + 8GB RAM + SSD is recommended for reliability.

How is OpenClaw different from running Claude Code or Cursor?

Claude Code and Cursor are developer tools for coding assistance within a terminal or IDE session. OpenClaw is a persistent personal agent you interact with via everyday messaging apps — it's always running, remembers everything, and handles non-coding tasks (email, calendar, smart home) in addition to development tasks. Users often combine both: "managing Claude Code sessions I can kick off anywhere."

Can multiple people share one OpenClaw instance?

Yes. You can configure multiple users to message the same Gateway via Discord, Slack, or other multi-user channels. Memory can be scoped per-user or shared, depending on configuration. This is how small teams use it as a shared operations tool.

How does OpenClaw handle sensitive credentials like bank logins?

Credentials are stored locally via 1Password or your system keychain — never in plaintext in the config file. When the browser skill needs to log in, it retrieves the credential from the local credential store. Credentials are never sent to OpenClaw servers (which don't exist — there are no OpenClaw servers involved in your agent's operation).

What happens when the machine running OpenClaw is off?

The Gateway needs the machine to be on. Messages sent while the machine is off will typically be delivered when it comes back online (depending on the channel's buffering behavior). For 24/7 availability, deploy on a low-power always-on device (Raspberry Pi, NUC, home server) or a cloud VM.

How does OpenClaw relate to projects like AutoGen or LangGraph?

AutoGen, LangGraph, and similar are frameworks for developers to build multi-agent systems from code. OpenClaw is a complete, pre-built product you can install and use immediately — no coding required. Technically, both categories are "agentic AI," but OpenClaw targets end users, while LangGraph targets application developers. Many developers use both: OpenClaw for personal use, custom frameworks for the products they build.

Is the project actively maintained?

Yes — as of March 2026, OpenClaw is one of the most actively developed open-source AI projects on GitHub, with near-daily releases and a large community on Discord. Several users note that it's "the first 'software' in ages for which I constantly check for new releases on GitHub."

15. References & Further Reading

• OpenClaw — Official Website
• OpenClaw — GitHub Repository (open-source code)
• OpenClaw — Official Documentation
• ClawHub — Community Skill Registry
• OpenClaw Blog — VirusTotal Partnership for Skill Security (February 2026)
• OpenClaw Blog — Introducing OpenClaw (January 2026)
• MacStories — "Clawdbot showed me what the future of personal AI assistants looks like" (Federico Viticci)
• OpenClaw — Full Integrations List (50+ integrations)
• OpenClaw — Trust and Security Page

OpenClaw is one of those rare tools where the gap between what you can imagine and what actually works has essentially closed. If you've ever wished you had a personal assistant who lives in your phone, remembers everything, and can actually do the work — not just answer questions about it — OpenClaw is the closest thing that exists today. The five-minute install is the lowest-risk way to experience it firsthand.