OpenBot delivers what xAI's Grok Bot promises, but on your own machine and without a SuperGrok Plus subscription: AI coworkers you can hand real work to. Each one gets its own computer: a Chromium browser with its own logins, its own file space, and only the tools you grant it. You watch it work on its own screen, take the wheel when it reaches something it should not do alone, then hand it back. Every action goes through a gateway that approves it before it happens and writes it to an audit log after, something Grok Bot does not show. The right habit before you start: grant each agent only the tools it needs, since those are the only ones it can use.
Versus Grok Bot (xAI): the match
Price
OpenBot
€0, you pay your server and your tokens
Grok Bot (xAI)
€86/mo, and nothing below SuperGrok Plus
Access
OpenBot
You clone it, you run it
Grok Bot (xAI)
Locked behind xAI's most expensive tier
Where it runs
OpenBot
On your machine, your data stays home
Grok Bot (xAI)
On their cloud, with your logins inside
| VS | OpenBot | Grok Bot (xAI) ↗ |
|---|---|---|
| Price | €0, you pay your server and your tokens | €86/mo, and nothing below SuperGrok Plus |
| Access | You clone it, you run it | Locked behind xAI's most expensive tier |
| Where it runs | On your machine, your data stays home | On their cloud, with your logins inside |
| The model | Yours: OpenAI, Anthropic or Google | Grok, and nothing else |
| Action log | Every move allowed or refused, written at home | Whatever they choose to show you |
How to install it (even from zero)
This is the most technical build in the library, and the project is still young. Set aside a good half hour, and read my take below before you start.
- 1
Install Docker Desktop
Download it from docker.com, install, open it once. It will run the database and each agent's computer.
- 2
Get your two keys ready
OpenBot needs two: a CopilotKit Intelligence account (a free tier exists) and a model key from OpenAI or Anthropic. Without both, nothing starts. Create them first, you will save time.
- 3
Paste this prompt into Claude (Code or Desktop)
It fetches the project from the official repository, installs everything and starts the server:
Installe OpenBot sur ma machine (la plateforme open source d'agents IA auto-hébergée, repo GitHub CopilotKit/OpenBot). Clone le dépôt github.com/CopilotKit/OpenBot, installe Bun s'il n'est pas déjà là, copie le fichier .env.example vers .env, puis guide-moi pour remplir les clés demandées : INTELLIGENCE_API_KEY et COPILOTKIT_LICENSE_TOKEN via la commande copilotkit login, ma clé de modèle, et génère un KEY_ENCRYPTION_KEY avec openssl. Lance ensuite scripts/start.sh et vérifie que l'interface répond sur http://localhost:3010. Attention : ce projet est en version alpha, si une étape casse dis-le-moi clairement au lieu de contourner. Explique chaque étape simplement, je débute. Review the GitHub repository and check whether it is safe to install given my machine and my infrastructure. Ask me as many questions as possible, using widgets, to clarify my request and be as precise as you can.
- 4
Set the boundaries BEFORE letting it work
Open http://localhost:3010, go to the boundaries settings and set your deny rules. Then test on a harmless task: ask it to open a website and summarise the top story, then check the audit log to see exactly what it did.
Paste this prompt into Claude (Code or Desktop)What you actually do with it
The web task you do by hand
Checking competitor prices, filling a long form, gathering information scattered across ten pages: the agent does it in its browser while you watch its screen.
One coworker per job
Three profiles ship with the project, and you create others without writing code: one for monitoring, one for internal questions, one for checking risks. Each keeps its own memory and access.
See exactly what it did
Every attempted action is written to a log, including the refused ones and why. When you hand real access to an AI, that page is what lets you sleep.
Project health, at a glance
Before installing an open-source tool, one question matters: is the project alive? These scores are computed from public GitHub data. They are not made up.
70/100
GoodHealthy, followed project: a good everyday choice.
- Activity
- 100
Is the code moving? (commits in the last 30 days)
- Maintenance
- 100
How fresh is the latest update? Fresher is better.
- Community
- 74
Stars and contributors: people around means help and fixes.
- Maturity
- 40
Project age. Old enough to be solid, still alive.
- Momentum
- 36
Current pace versus its usual pace: speeding up or dozing off?
The repo in numbers
- GitHub stars
- 9k
- Forks
- 1.1k
- Contributors
- 47
- Commits
- 593
- Project age
- 0.2 yrs
- Last commit
- today
- Size
- 21.6 MB
- License
- MIT
Built with
- TypeScript 89.4%
- Rust 8.3%
- Python 1.3%
Scores computed automatically from the GitHub API (last-30-days activity, update freshness, community size, project age). Refreshed daily.
Loucash’s take
7/10
Careful, this one is not like the other cards: it came out only days ago, and its authors state plainly that it is alpha, with bugs to expect. I am putting it in the library anyway for one simple reason: xAI shipped Grok Bot six days earlier and it only opens up in SuperGrok Plus, the €86 a month tier. Here you get the same promise by cloning a repository. And OpenBot adds what xAI does not show: a gateway that approves every move before it happens and keeps a record of it. Two things to know before you start, which the landing page does not say loudly enough. You need a CopilotKit Intelligence account on top of the software: the free tier is enough to try, and you can host it yourself, but you depend on an outside piece. And above all, free does not mean costless: the tool charges nothing, but every action burns tokens at OpenAI or Anthropic. An agent working in a browser burns a lot of them. Watch your bill the first few days. My advice: install it to understand where the craft is heading, play with it on tasks without consequences, but put nothing serious on it until it leaves alpha.
It’s for you if: You saw xAI's Grok Bot go by and want the same thing without paying €86 a month, accepting you will tinker with a young tool.
Also worth a look in the library
n8n
Agents & orchestration€27/moWire your tools together and let it run: when this happens, do that. Self-hosted, no task counter stopping you mid-month.
View details →
Paperclip
Agents & orchestration€26/moThe open source dashboard to run a team of AI agents: you hand them tasks like you would to colleagues, they work on their own on a schedule, and you approve what matters. The software is free; the AI is your Claude or ChatGPT subscription, or your API key.
View details →
Hermes Agent
Agents & orchestration€23/moAn agent that learns from you and grows with use. Wire it to your second brain, run it on a local model, and you have a personal assistant that talks to nobody but you.
View details →
One email a week that saves you money
Every Tuesday: the open source alternatives that just entered the library, what they replace and what they save you. One email, nothing more.
Free. One email on Tuesdays. Unsubscribe in one click. Your data
Join the Tuesday readers
