Anton, the open-source AI coworker, now in the cloud
It’s Friday afternoon, and you’ve got an AI agent running on your laptop — one of the capable ones that writes SQL, runs a little Python, and builds a chart. You ask it to audit last quarter’s refunds and have a dashboard ready for Monday. It starts working. Then you close the lid to catch your train.
Monday morning: nothing. It stopped the moment the screen went dark. That isn’t a flaw in the agent — it’s a flaw in where it lives. Anything on your laptop only runs while the laptop is awake, and a coworker who only works while you’re watching isn’t much of a coworker.
So we gave one a better place to work. Today, Anton runs on MindsHub — a one-click, always-on cloud deployment, live in about three minutes. Anton is the open-source AI coworker; you work with it through CoWork, the UI for MindsHub’s agents. Together, in the cloud, they keep going long after you’ve closed your laptop.
A coworker, not a chatbot
Hand Anton an outcome — a report, a dashboard, a quarterly audit — and it hands back the finished artifact, not a suggestion. That’s what makes it feel less like a chatbot and more like a teammate:
- It does the work, not just the talking. Anton writes the SQL, runs Python in a sandbox, reads its own results, and builds the visualization. You get a deliverable, not a draft.
- It remembers. Anton keeps an episodic memory of your schemas, your past instructions, and the way you like things done — so you’re not re-explaining your data model every Monday.
- It gets better at your work. Every task it finishes feeds a growing library of what worked. The Anton you run in March is sharper on your data than the one you met in January.
You see all of it through CoWork: one clean workspace where you brief the agent and watch every step it takes — the same UI that will drive Hermes and more agents over time.
Why a laptop holds any agent back
A teammate you have to babysit isn’t much of a teammate. Run an agent like this on your own machine and three things get in the way:
- It sleeps when you sleep. Close the lid or drop off Wi-Fi, and a long migration or an overnight audit dies mid-loop.
- It needs the keys — and your laptop is the wrong keyring. To query your databases, an agent needs credentials. A plaintext config file on your local drive is the last place production secrets should live.
- It competes with everything else on your machine. Big queries and iterative runs want steady compute, not whatever’s left over from your forty browser tabs.
None of this is the agent’s fault. It’s where you’re running it.
Give Anton a desk in the cloud
Doing this by hand means building containers, wiring reverse proxies, juggling environment variables, and opening database tunnels. On MindsHub it’s one click and about three minutes. Here’s what your coworker gets:
- A desk that’s always on. Anton runs in an isolated cloud container, 24/7. The overnight audit runs overnight. The Monday report is finished before you are.
- The keys, without handing them out. Connection strings and API tokens stay in MindsHub’s credentials vault. Anton uses them at runtime but never sees them in the clear — and they never land in a config file or a chat log.
- The right model for each step, pre-wired. Anton routes through MindsHub’s Model Router out of the box. No keys to paste, no gateway to stand up.
- Nothing happening behind your back. Every planning loop, query, and line of terminal output streams into the scratchpad in real time, right inside CoWork — and doubles as an audit trail.
Hand off the work and close your laptop
The line between a tool and a teammate is simple: a teammate keeps working when you step away. Hand off the quarterly audit on Friday, shut your laptop, and find the dashboard waiting on Monday.
Anton is live on MindsHub now, on the $9.95/month tier, which includes 5 million router tokens.
MindsHub by MindsDB is the open platform for running AI agents — OpenClaw, NanoClaw, Anton, and Hermes — with the infrastructure they need to do real work: a Model Router across providers, a credentials vault, tools and data access, persistent execution, scratchpads, and memory. Founded 2018 in Berkeley. Backed by Benchmark, Mayfield, Y Combinator, and NVIDIA.