Every AI coding tool we tried had the same shape. It sat inside your editor, waited for you to type, and handed the work back to you one suggestion at a time. Helpful, in the way a very fast intern is helpful. Also exhausting, because you were still the one holding everything together.
We wanted the other shape. An agent that has its own place to work, its own tools, and enough autonomy to take a task and finish it while you do something else. Today that agent is available to everyone.
The agents dashboard. Every card is a real cloud machine with live CPU, memory, and disk.
What an agent gets
When you create a JackHamr agent, we provision a real Linux machine in the cloud for it. Not a sandboxed function that dies after thirty seconds. A machine, with a file system, a shell, network access, and the ability to install whatever the job needs.
- A streaming chat you open in the browser, with full history and file handling
- The ability to run commands, read and write files, and browse the web
- Platform tools for memory, images, and skills it can load on demand
- A generated portrait, name, and title so your roster reads like a team, not a list of IDs
- Live CPU, memory, and disk on the dashboard so you can see it working
Sign in with GitHub or Google. Choose a model: Claude Sonnet and Opus, GLM-5, or Kimi K2.5 at launch. Bring an Anthropic subscription if you have one rather than an API key.
Why a machine matters
An assistant can suggest a fix. An agent with a machine can apply it, run the tests, see them fail, read the log, fix the real problem, and run them again. The difference between those two is the difference between a tool and a teammate.
An agent building a pricing page in chat, with the tool call it ran and the artifact it saved.
You direct. The agent ships.
This first release is deliberately focused on that loop. Everything we ship after today builds on the same foundation: one agent, one machine, and enough trust to leave it alone. Very soon, many agents and many machines.
Welcome to JackHamr.
Originally published on the JackHamr blog. JackHamr is an AI agent platform where agents plan, build, test, and review software end to end. Free to start.
Top comments (1)
Per-agent machines make a lot of sense once you've watched two agents fight over the same working tree. How do you handle shared state when two of them need the same repo?
Testing is my day job (Staff SDET), so I ran jackhamr.ai through an automated pass while reading. The report card came back 99/100 (A), which is very clean. The one thing I'd fix: page titles on the compare and blog pages run long. /compare is 93 characters ("JackHamr vs Claude Code, Cursor, Codex, Devin, Factory, Replit, Lovable, Bolt and More (2026)") and most /compare/* pages sit between 66 and 84, so Google cuts them off right where the differentiator is. Dropping the " — JackHamr" suffix and the "(2026)", or capping the title template near 60 characters, handles it. The scan turned up 1 other issue, also minor.
Since those compare pages change every time a competitor does, a recurring check on them is probably worth having.