Under "Did Claude Duet just kill Grok Bot?", a video comparing the new consumer agents, a commenter laid out the decision this post answers: "if I'm using Open Claw or Hermes, anything within that region, should I be stopping & utilizing those and moving to Dots, Duet, or Grok bots? Have these LLMs essentially just replaced Open Claw? One advantage I have is that my Open Claw and my Hermes know me. They know my style, clients -all of those things" (5 likes, @MattStringhamcom).
The comments under the agent videos split the same way. One reports their main agent is "still openclaw. Since Feb." (11 likes). Another says recent updates pushed them to make Grok Bot their "main driver." A third cancelled Grok Bot over its usage limits and went back to Hermes. The answer rests on three decisions: what your jobs need, where the memory should live, and what budget shape fits.
Two families of agent
A personal agent has two parts: the model that does the reasoning, and the harness around it, the program that holds memory, connects tools, runs steps, and asks for approval. The harness decides what the agent can reach, remember, and act on, and the two routes treat it differently.
The DIY route is a harness you install and run yourself, with the model as a setting you change. OpenClaw's homepage describes it as open source, running on your machine, with "No subscription. No hosted tier" and "Your memories. Your skills. Your models. Your machines. Your data." Nous Research's Hermes Agent is free and open source too, with desktop apps for Mac, Windows, and Linux, and docs that say it "creates skills from experience" and "builds a deepening model of who you are." Both connect to multiple model providers: an API account, an open model running locally, or a mix.
The all-in-one route is the four consumer agents from this fall: xAI's Grok Bot, Meta's Muse, OpenAI's dots, and Anthropic's Claude Duet. Each runs its work on a cloud computer the vendor operates, on a subscription, per the launch announcements (What Is a Personal AI Agent? Why the Demos Show Dinner Reservations compares the four). You configure access in their apps; the vendor maintains and updates the harness and sets what the agent can reach and when it asks for approval.
What the all-in-one buys you
Setup is measured in minutes, and the approval prompts come from a product team. The costs sit in three places. Usage resets on the vendor's schedule, and agent work can draw from an allowance faster than chat (Why Does AI Usage Run Out So Fast? How Subscription Limits Work explains the mechanics; the commenter who cancelled Grok Bot for Hermes cited exactly this). The tasks follow the vendor's roadmap: the four announcements describe errand work, email and messages, scheduling, purchases, and browser work, and the features you get are the ones the vendor ships. And the agent's memory is stored on the vendor's side, under the vendor's data policies (Do AI Agents Train on Your Data? What Agents See, Save, and Learn covers what the four vendors do with it).
What the DIY route buys you
Control of the three things the consumer vendor handles. Memory lives in files on the machine you run it on, so an agent's knowledge of your clients and your style is copyable, not account-bound. Models are swappable per job, including local ones that never leave the machine (What Is Local AI? Running Models on Your Own Computer). And no allowance resets on a schedule: a DIY agent draws from an API bill that tracks what you use (What Does It Cost to Run an AI Agent? Subscriptions vs API works through the numbers) or from a machine you already paid for.
Maintenance moves to you: the commenter whose main agent is OpenClaw calls upgrades "annoying" (11 likes), though their agent has since built its own upgrade process. Security does too: your exposure comes from the code you install and the permissions you grant (How to Tell If a GitHub Repo Is Safe to Run covers the checklist). And the setup time the consumer bot absorbs becomes your responsibility.
Why this Lab runs on a DIY harness
The publishing operation behind this site runs on a DIY agent: a scheduled scrape at 1:20 AM, a script that mines the comments into themes, a coverage check, the draft, a review pass, verification, the deploy. This morning that pipeline mined 5,468 comments into 13 themes before 3 AM. It has published a post a day since September.
None of the four consumer bots exposes the pieces this lane needs: custom scripts we wrote, files in our own workspace, three schedules we set ourselves. So we spend maintenance hours instead of an allowance, and the knowledge of how this operation works lives in our files. The maintenance cost is not hypothetical. Three pipeline failures made it into our own ledger in October, and each one cost us about an hour to find and fix.
A test before you switch
Three questions, in order:
Do your jobs fit the bot's lanes? If the work is errands, email, calendar, bookings, purchases, browsing, the four consumer agents are built for that, and the maintenance they absorb is worth more than the control you'd give up. If the jobs need custom tools, your own files, or models of your choosing, a consumer bot may not expose them.
Where should the memory live? The commenter's hesitation sits in the last line: "my Open Claw and my Hermes know me. They know my style, clients -all of those things." On the DIY route that knowledge lives in files, so switching harnesses later usually means copying files. On the consumer route it's stored in the vendor's cloud, so leaving usually means starting over. Portability is the deciding factor several commenters name on their own.
What budget shape fits? A subscription with a reset schedule, a usage-tracked API bill, or a machine you own. The consumer bots bundle the first; a DIY harness lets you pick either of the other two.
Then run the cheap experiment: the bot and your harness side by side for two weeks, one lane each. The comment threads show people switching in both directions: one reports hopping from OpenClaw to Hermes to dots to Grok Bot (8 likes), and another cancelled the bot and returned to their harness. Keep your files and the context is easier to carry over, though tools and permissions still need reconnecting.
FAQ
Have all-in-one bots replaced DIY agents like OpenClaw?
Not as a category. The bots and the harnesses share the model layer; the harness, the program holding memory, tools, permissions, and schedules, is what differs, and it decides who maintains the setup and how you pay. An all-in-one bot replaces the setup and maintenance work, but it doesn't run custom scripts or keep your context in files you own. Choose by job fit, memory ownership, and budget shape.
Can I run both a DIY agent and an all-in-one bot?
Yes, and it's a common pattern: a lane split. Errand-shaped jobs go to the bot, with approval prompts on. Pipeline-shaped jobs, anything with custom scripts, your own files, or heavy daily volume, stay on the harness. How to Keep Track of What Your AI Agents Are Doing covers watching both at once.
What happens to my agent's memory if I switch tools?
It depends on the route. Consumer bots store memory in the vendor's cloud, tied to your account, so moving vendors usually means rebuilding context. OpenClaw and Hermes keep memory and skills as files on the machine you run them on, so moving between them means copying files and reconnecting tools; formats differ, so expect some adjusting.
What does a DIY agent cost compared to a subscription?
The software is free; both OpenClaw and Hermes are open source. You pay for model access. An API account bills per million tokens and tracks what you use, which can land under or over a $20 subscription depending on volume (What Does It Cost to Run an AI Agent? Subscriptions vs API works an example). A local model moves the cost to hardware you already own.
Which DIY harness should I start with, OpenClaw or Hermes?
Both are free, open source, and run on Mac, Windows, or Linux. OpenClaw connects through chat apps you already use and installs from desktop apps. Hermes ships desktop apps with a skills system its docs say improves with use. For a first harness, the shared habit, memory in files you own, may matter more than which one you pick.
Want to go further?
Start with the lane split: pick the one job that annoys you most, and run it where it fits. For a comparison of the four bots, What Is a Personal AI Agent? Why the Demos Show Dinner Reservations covers the category. For the cost side of running agents, Why Does AI Usage Run Out So Fast? How Subscription Limits Work covers how subscription allowances reset.