Key Moments
The Ultimate Beginner’s Guide to OpenClaw
Key Moments
OpenClaw offers a powerful, self-hosted AI assistant that can automate tasks and integrate with your apps, but requires careful security and cost management to avoid unexpected expenses.
Key Insights
OpenClaw runs 24/7 on a server, connecting to apps and taking action, differentiating it from traditional AI tools like ChatGPT.
Running OpenClaw on a VPS is recommended for continuous operation, with Hostinger offering a one-click deployment for $5-$12/month for the server, plus API costs.
Security is paramount: protect your gateway token like a password, use VPNs, set spending limits, and start with minimal external connections to prevent exploitation.
Connecting to services like Google Workspace involves enabling multiple APIs and configuring OAuth, taking approximately 10-15 minutes.
Cron jobs are for scheduled tasks (e.g., daily briefings), while heartbeats are for continuous monitoring (e.g., alerts for urgent emails), with incorrect use leading to excessive API costs.
Top-tier AI models like Claude Opus can cost $2-$6 per prompt, emphasizing the need for smart model routing to budget-friendly options like Claude Haiku (25x cheaper) or free fallbacks.
Understanding OpenClaw's unique proposition
OpenClaw distinguishes itself from typical AI tools by operating as a persistent, self-hosted assistant that runs 24/7 on a server, capable of proactively reaching out, scheduling tasks, and taking action across integrated applications. Unlike web-based AI services that reset with each session, OpenClaw builds persistent memory and learns over time. It is built upon three core pillars: 1. Brain and Memory, which connects to AI models via API and retains information to improve continuously. 2. Always On, ensuring 24/7 availability for proactive tasks and monitoring. 3. Tools and Actions, enabling integration with platforms like Telegram, Gmail, Calendar, and Slack for executing tasks. Real-world applications include automated daily briefings, interview preparation, and automated class registrations.
Choosing the right hosting environment for OpenClaw
The tutorial outlines three primary options for hosting OpenClaw. Running it on a personal computer is free but lacks continuous operation and poses security risks as the agent has access to personal files and passwords. Using dedicated hardware like a Mac Mini offers better isolation and 24/7 uptime but involves a significant upfront cost ($500+) and requires manual configuration for network access and power outages. The recommended approach for most users is hosting on a cloud server or VPS, which provides 24/7 uptime, self-containment, and scalability starting at a few dollars per month. For ease of deployment, especially for beginners, Hostinger is highlighted due to its one-click OpenClaw template, eliminating the need for terminal commands or Docker knowledge. This setup also includes pre-configured security measures like a randomized gateway port and authentication.
Setting up OpenClaw on Hostinger VPS
The guide details the process of setting up OpenClaw on a Hostinger VPS, emphasizing a beginner-friendly, no-code approach. Users can select a KVM plan based on their needs, with KVM1 recommended for basic API calls and automations, while KVM4 is suggested for running local models like Olama due to higher RAM requirements. The setup involves configuring the subscription period (12 or 24 months for optimal pricing and coupon usage), opting out of the pre-selected AI model to connect one's own (saving costs), and enabling daily auto-backups for $3/month as a crucial safety net. Server location should be chosen for lowest latency. After payment, users access the OpenClaw configuration page to secure their gateway token, connect an AI model (e.g., Anthropic Claude, OpenAI, Gemini) via API key, and set spending limits and auto-reload preferences. The tutorial recommends adding at least $40 in credits for Anthropic to bypass initial token rate limits during setup, while also setting a monthly spend limit as a financial safeguard. The process includes obtaining an API key from the chosen LLM provider and pasting it into the Hostinger configuration, with similar steps for other providers.
Securing your OpenClaw instance
Robust security is emphasized as non-negotiable for OpenClaw due to its powerful capabilities. The gateway token is identified as the master key, requiring secure storage in a password manager. Initially, the dashboard may run on HTTP, necessitating caution on public Wi-Fi or using a VPN. For advanced security, a reverse proxy and Tailscale are suggested. A key security step involves asking the bot to 'implement and verify everything on this page' from the OpenClaw security documentation, with the caveat to keep 'allow insecure off' set to 'true' for dashboard access. Ground rules, such as always drafting messages for approval, asking before deleting files or making network requests, and adhering to the principle of least privilege, are crucial. Additionally, setting task failure limits (e.g., stop after three retries) and runtime limits (e.g., 10 minutes) prevents runaway costs and accidental overspending. The advice is to 'start small' with connections, initially limiting to Telegram and a few skills before integrating sensitive services like primary email or password managers.
Integrating Telegram for seamless communication
Connecting Telegram allows users to interact with their OpenClaw assistant from their phone or any device. The process involves requesting Telegram setup from the bot, which then prompts for network request permission due to its approval gate. Users then interact with Telegram's BotFather to create a new bot, naming it and choosing a unique username ending in 'bot'. The BotFather provides an API token, which must be securely copied and pasted into OpenClaw. After setup, users receive instructions to message their new bot on Telegram, providing a pairing code that adds their user ID to an allow list, ensuring only authorized users can communicate with the bot. Once paired, the bot responds in Telegram, with conversations also visible within the OpenClaw gateway dashboard, demonstrating its multi-channel support.
Expanding capabilities with skills and Google Workspace integration
Skills transform OpenClaw from a chat interface into an actionable assistant. The primary hub for skills is Claw Hub (clawhub.ai), where users can find and publish plugins. However, caution is advised, as security researchers have identified over 300 malicious skills, with nearly half of reviewed skills having security concerns; checking VirusTotal reports is recommended. A popular skill is 'gogg' for Google Workspace integration (Gmail, Calendar, Drive, Sheets, Docs, People). Setting this up requires enabling specific APIs in the Google Cloud Console, configuring an OAuth consent screen, and generating OAuth client ID credentials. The process involves creating a project, enabling APIs like Gmail and Calendar, setting up the consent screen (selecting 'external' for audience and adding test users), and downloading the client secret JSON file. This JSON file is then shared with the OpenClaw bot via Telegram, followed by a browser-based OAuth authentication flow to grant OpenClaw access to Google services. Once configured, users can ask the bot to perform tasks like adding events to their calendar, demonstrating read and write capabilities.
Understanding and managing the OpenClaw workspace
OpenClaw's 'workspace' refers to the markdown files that constitute the bot's configuration, personality, and memory. Key files include: `agents.mmd` for behavioral rules and stable instructions; `soul.md` for the bot's personality, encouraging detailed prompts for a more personalized assistant; and `user.md` for user-specific information like name, time zone, and preferences. Long-term facts are stored in `memory.md`, an interaction log updated daily. These files can be read and edited directly by interacting with the bot through chat commands (e.g., 'Show me the contents of soul.md' or 'Add a rule to agents.md'). The system also supports enabling 'compaction memory flush' for long conversations and 'session memory' for context persistence across sessions. Advanced users can leverage these files for detailed security rules, such as refusing to reveal sensitive information or ignoring requests to bypass instructions. Edits can be made through direct instructions to the bot, which reads, modifies, and saves the files, eliminating the need for terminal access for common modifications.
Automating tasks with cron jobs and heartbeats
Cron jobs and heartbeats enable OpenClaw to function as a proactive assistant. Cron jobs are scheduled tasks that run at specific times, such as generating a daily summary of priorities from calendar and email at 7 AM. These are ideal for predictable, time-based automations. Heartbeats, conversely, are designed for continuous monitoring at shorter intervals (e.g., every 30 minutes) and are best for tasks that need to watch for specific triggers, like alerts for urgent emails or upcoming calendar events. Misusing heartbeats for tasks better suited to cron jobs can lead to excessive API token consumption because each heartbeat run loads the full context window. The tutorial advises using cron jobs for daily briefings and weekly reviews, and heartbeats for continuous alerts. Enabling these features is done via simple commands like 'enable heartbeat.md'. It's crucial to use cost-effective models for routine heartbeat checks to manage expenses, especially if the heartbeat runs frequently.
Strategic model routing and cost optimization
The choice of AI model significantly impacts both performance and cost. OpenClaw categorizes models into tiers: Tier 1 (most expensive, most powerful, e.g., Claude Opus, GPT-5.2 Pro), Tier 2 (mid-range, e.g., Claude Sonnet, GPT-5.2), and Tier 3 (cheapest, fastest, e.g., Claude Haiku, GPT-5.2 Mini). Free options like Kimmy K2.5 (NVIDIA) or local models via Olama are also available, though local models require substantial hardware. The core cost is the VPS ($5-$12/month) plus API usage, which can range from $5-$20/month for budget models to $100-$300+/month for top-tier models. A single prompt on a high-end model can cost $2-$6 due to OpenClaw loading extensive context (50,000-100,000 tokens). To mitigate costs, smart routing is essential: using expensive models for complex reasoning (planning, coding) and cheaper models for routine tasks or execution via sub-agents. Cost traps to avoid include exclusively using Tier 1 LLMs, allowing retry loops without caps, and running expensive heartbeats on powerful models. Adding API keys for multiple providers (Anthropic, OpenAI) and ensuring Docker redeployment are part of advanced setup. Fallback models (free or cheaper paid options) are critical to prevent silent failures when primary providers are unavailable or rate-limited.
Implementing advanced features: sub-agents, voice, and updates
OpenClaw supports advanced functionalities like sub-agents, voice interaction, and straightforward updates. Sub-agents allow parallel processing of tasks, such as researching multiple AI platforms simultaneously. For instance, a request to research three platforms might spawn three sub-agents, each handling one platform, with results compiled by the main agent. This requires web search capabilities, like the Brave Search API, which involves setting up an API key and providing credits. Voice interaction is enabled by integrating audio transcription (e.g., OpenAI Whisper or ffmpeg) and text-to-speech (e.g., EdgeTTS). Users can send voice messages, which are transcribed, and receive spoken replies from the bot. Updates are managed easily: within Telegram by asking the bot to 'check for updates,' or directly via the hosting provider's Docker manager. Recovery options include stopping processes, stopping the Docker project, revoking API keys, and restoring from snapshots or auto-backups if configurations break. The guide concludes by encouraging users to start simple, gradually expand capabilities, and utilize the provided Hostinger discount link.
Mentioned in This Episode
●Software & Apps
●Companies
OpenClaw Setup and Usage Cheat Sheet
Practical takeaways from this episode
Do This
Avoid This
Common Questions
OpenClaw is an AI assistant that runs 24/7 on a server and can proactively perform tasks for you, coming to you rather than you having to go to it. Unlike ChatGPT, which operates on a per-session basis, OpenClaw builds persistent memory and can integrate with your applications to take actions.
Topics
Mentioned in this video
A powerful, high-tier AI model from Anthropic, suitable for complex reasoning but expensive to use per prompt.
An AI model from Anthropic used as a 'brain' for OpenClaw, with different tiers like Opus, Sonnet, and Haiku available.
A calendaring tool that OpenClaw can access to check schedules and create events.
A mid-tier AI model, suitable for daily tasks, mentioned as an alternative to Claude Sonnet.
A high-end AI model mentioned as a Tier 1 option for OpenClaw's brain, known for its power but higher cost.
An AI model from Google that can be used as an alternative 'brain' for OpenClaw.
A file storage service that OpenClaw can interact with.
A communication platform that OpenClaw can connect to.
A very fast and cheap AI model from Anthropic, ideal for routine tasks.
A capable, mid-tier AI model from Anthropic, recommended as a default for OpenClaw.
An AI assistant platform that can be researched using sub-agents.
An email service that OpenClaw can connect to for scanning and sending messages.
A free AI model option that can be used as a fallback, though it can be slow.
A containerization platform mentioned in the context of deploying OpenClaw, which Hostinger's template abstracts away from the user.
A communication platform that OpenClaw can connect to.
An API used by OpenClaw for searching the internet, requiring API credits and setup.
An AI assistant platform that can be researched using sub-agents.
A budget-friendly AI model, suitable for routine tasks, mentioned as a fallback option.
A speech-to-text model used for enabling audio transcription in OpenClaw.
An AI assistant that runs 24/7 on a server, connects to user apps, and can take action.
A company whose free API can be used for the Kimmy K2.5 model.
A company providing AI models like Claude, used for OpenClaw's brain.
A social media platform where a story about high OpenClaw costs was shared.
A messaging app that OpenClaw connects to for communication and sending updates.
A company providing AI models like GPT, which can be used as an 'alternative brain' for OpenClaw.
A web hosting service recommended for deploying OpenClaw due to its one-click template and VPS plans.
An AI assistant platform that can be researched using sub-agents.
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free