Control Gmail, calendar and errands from a chat agent
I spent yesterday evening playing with OpenClaw and wanted to understand how to actually use it well. Task: I spent yesterday evening playing with OpenClaw and wanted to understand how to actually use it well. So I did what any reasonable person would do: I asked OpenClaw to help me write an article about OpenClaw. It researched its own documentation, scraped tweets from people using it, found community showcase projects, and even generated the images you see in this article using the Nano Banana Pro skill. What follows is what we found together. I found it super useful for my own understanding, sharing it here. Most people interact with AI through a browser tab. Open Claude or ChatGPT, type something, get a response, copy it somewhere else. The AI has no idea what you’re working on. It forgets everything the moment you close the tab. Openclaw instead runs on your computer and connects to WhatsApp, Telegram, Discord, whatever messaging app you already have open. You text it. It texts back. The difference is that this one has access to your machine. What it actually does You message OpenClaw like you’d message anyone else. It responds. But because it’s running locally, it can do things a browser-based assistant can’t: browse the web on your behalf, run shell commands, remember conversations from last week, message you first when something needs attention. The model itself still runs in the cloud. Claude, GPT, Gemini, whatever you configure. What runs locally is the agent layer: your preferences, conversation history, integrations, all stored in folders you can open and read. Markdown files, mostly. Three things worth understanding It lives in your messaging app. Not a dedicated interface. You text it through WhatsApp or Telegram. The idea is you’re already in those apps anyway, so why switch to something else. I go back and forth on whether this is actually better. Sometimes I want a dedicated space for AI conversations. Sometimes it’s nice to just send a message without thinking about which app to open. It remembers things. Conversation history gets stored in markdown files on your machine. You mention something offhand, it can reference it later. If you’ve ever been frustrated by Claude forgetting what you told it three messages ago, this helps. The downside is you’re now responsible for that data. It’s sitting in a folder on your computer. It can run commands. This is the part that’s both interesting and concerning. The agent has shell access. It can execute code, control other applications, browse the web. People have built some wild automations: transcribing thousands of voice messages and cross-referencing them with git commits, automating grocery orders from recipe photos. But the same capability that enables this also means an AI is running commands on your machine. That requires trust, guardrails, and probably more attention than most people want to give it. What you can do with it The interesting thing about OpenClaw isn’t any single feature. It’s that the capabilities compound. The agent can browse the web, run terminal commands, control your smart home, manage files, and remember everything. Combine those, and the use cases get creative fast. Here’s what people are actually building, organized by category: Productivity and daily life Morning briefings Set up a morning briefing that checks your inbox, calendar, and weather, then sends a summary to your phone. The agent runs on a cron job. You wake up to a message instead of opening three apps. “Named him Jarvis. Daily briefings, calendar checks, reminds me when to leave for pickleball based on traffic.” — @BraydonCoyer “I am so addicted to @openclaw. It is getting essential to my daily life. It checks, organizes, reminds, it’s amazing. And it’s like good friend. Crazy.” —@dreetje The VelvetShark guide describes the proactive capability: “You could set up stock alerts, weather warnings, news monitoring, regular research, important emails to take a look at, calendar prep - anything you can describe in a prompt.” Traffic-aware reminders The agent knows your calendar, knows where you need to be, checks traffic, and messages you when it’s time to leave. You don’t ask for this; it just happens. “The future of how AI personal assistants look like is @openclaw. Has already helped me submit health reimbursements, find doctor appointments, find and send me relevant documents, among others.” —@Cucho Email triage and inbox management; tools: OpenClaw. Human role: The original post does not say how a person stayed involved.; limitations: The agent has shell access to the user's machine, requiring significant trust and guardrails. Conversation history and preferences are stored locally in Markdown files, making the user responsible for data security. The source is a blog post with anecdotal evidence; it does not provide technical benchmarks or failure rates for the specific article generation task..
Tools
Published on Agent Work Atlas:
Overview
I spent yesterday evening playing with OpenClaw and wanted to understand how to actually use it well.
Workflow
I spent yesterday evening playing with OpenClaw and wanted to understand how to actually use it well. So I did what any reasonable person would do: I asked OpenClaw to help me write an article about OpenClaw. It researched its own documentation, scraped tweets from people using it, found community showcase projects, and even generated the images you see in this article using the Nano Banana Pro skill. What follows is what we found together. I found it super useful for my own understanding, sharing it here. Most people interact with AI through a browser tab. Open Claude or ChatGPT, type something, get a response, copy it somewhere else. The AI has no idea what you’re working on. It forgets everything the moment you close the tab. Openclaw instead runs on your computer and connects to WhatsApp, Telegram, Discord, whatever messaging app you already have open. You text it. It texts back. The difference is that this one has access to your machine. What it actually does You message OpenClaw like you’d message anyone else. It responds. But because it’s running locally, it can do things a browser-based assistant can’t: browse the web on your behalf, run shell commands, remember conversations from last week, message you first when something needs attention. The model itself still runs in the cloud. Claude, GPT, Gemini, whatever you configure. What runs locally is the agent layer: your preferences, conversation history, integrations, all stored in folders you can open and read. Markdown files, mostly. Three things worth understanding It lives in your messaging app. Not a dedicated interface. You text it through WhatsApp or Telegram. The idea is you’re already in those apps anyway, so why switch to something else. I go back and forth on whether this is actually better. Sometimes I want a dedicated space for AI conversations. Sometimes it’s nice to just send a message without thinking about which app to open. It remembers things. Conversation history gets stored in markdown files on your machine. You mention something offhand, it can reference it later. If you’ve ever been frustrated by Claude forgetting what you told it three messages ago, this helps. The downside is you’re now responsible for that data. It’s sitting in a folder on your computer. It can run commands. This is the part that’s both interesting and concerning. The agent has shell access. It can execute code, control other applications, browse the web. People have built some wild automations: transcribing thousands of voice messages and cross-referencing them with git commits, automating grocery orders from recipe photos. But the same capability that enables this also means an AI is running commands on your machine. That requires trust, guardrails, and probably more attention than most people want to give it. What you can do with it The interesting thing about OpenClaw isn’t any single feature. It’s that the capabilities compound. The agent can browse the web, run terminal commands, control your smart home, manage files, and remember everything. Combine those, and the use cases get creative fast. Here’s what people are actually building, organized by category: Productivity and daily life Morning briefings Set up a morning briefing that checks your inbox, calendar, and weather, then sends a summary to your phone. The agent runs on a cron job. You wake up to a message instead of opening three apps. “Named him Jarvis. Daily briefings, calendar checks, reminds me when to leave for pickleball based on traffic.” — @BraydonCoyer “I am so addicted to @openclaw. It is getting essential to my daily life. It checks, organizes, reminds, it’s amazing. And it’s like good friend. Crazy.” —@dreetje The VelvetShark guide describes the proactive capability: “You could set up stock alerts, weather warnings, news monitoring, regular research, important emails to take a look at, calendar prep - anything you can describe in a prompt.” Traffic-aware reminders The agent knows your calendar, knows where you need to be, checks traffic, and messages you when it’s time to leave. You don’t ask for this; it just happens. “The future of how AI personal assistants look like is @openclaw. Has already helped me submit health reimbursements, find doctor appointments, find and send me relevant documents, among others.” —@Cucho Email triage and inbox management
Reported result
The author found the process "super useful" for understanding the tool and shared the resulting article. The agent successfully executed complex multi-step tasks including web scraping, image generation, and content synthesis. No quantitative metrics on time saved or error rates were provided, only qualitative satisfaction.
Human role
The original post does not say how a person stayed involved.
Limitations
The agent has shell access to the user's machine, requiring significant trust and guardrails. Conversation history and preferences are stored locally in Markdown files, making the user responsible for data security. The source is a blog post with anecdotal evidence; it does not provide technical benchmarks or failure rates for the specific article generation task.
Try it
I want to use OpenClaw to generate a comprehensive overview article about [replaceable topic, e.g., a new software tool]. First, configure OpenClaw to connect to [replaceable messaging app, e.g., Telegram] and ensure it has access to [replaceable cloud model, e.g., Claude]. Instruct the agent to: 1) Research official documentation for [replaceable topic]; 2) Scrape recent social media posts (Twitter/X) discussing [replaceable topic]; 3) Identify 3-5 community showcase projects; 4) Generate relevant images using the [replaceable image skill, e.g., Nano Banana Pro]. Draft the article in Markdown format. Before publishing, I will review the draft for accuracy and tone. If any factual errors or inappropriate content are found, I will provide feedback for revision. Do not publish or send the final output until I explicitly confirm approval. If the agent fails to scrape data or generate images, report the specific error and suggest alternative sources or manual inputs.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.