Close the laptop.Your agents keep working.
Cronspace runs your agents in the cloud — several at once, each on its own schedule. Start a run at your desk, watch it from your phone, and approve what needs you from wherever you are.
Cronspace is in private beta. Have an invite code? Create your account.
Featuring
Compatible with the models you already use.
Qwen
MiniMax
An AI chat is brilliant for thirty seconds, then it forgets.
You pay for an assistant that starts from zero every conversation, runs only while you watch, and loses everything when the tab closes. So the work that actually repeats — the Monday report, the client follow-up, watching the competition — is still yours to do by hand.
Cronspace inverts it. The chat is still there, but it’s a channel. The product is the work.
A blank prompt.
Every session starts from zero.
Agents that persist.
Their instructions, tools and history stay put.
Work while you watch.
Close the tab and it's gone.
Durable runs.
They survive closes, failures and deploys.
A transcript.
Value dies in the scroll.
An inbox of finished work.
Outputs, decisions and files you can act on.
A usage cap.
Hit it mid-task and you wait.
Pay as you go.
The only ceiling is the one you set.
How it works
Three steps, then it runs itself.
- 1
Set up an agent.
Pick a model, write what it should do, switch on the tools it needs. Every workspace starts with a general-purpose agent, so you can skip this on day one.
- 2
Give it work.
Ask in the chat and it does it right there. When the work should repeat, say so — the agent drafts the task with you and puts it on a schedule.
- 3
Review in the inbox.
Finished runs, files, and anything waiting on your approval. Approve from your phone; the agent picks up where it left off.
The inbox
Your home screen is the work, not the chat history.
Cronspace opens on an inbox with four answers: what needs your decision, what’s running right now, what finished and what failed. Every run carries its outputs and the full step-by-step of how it got there.
Nothing lives in the browser. Start a run at your desk, watch it from your phone on the train — the stream picks up mid-flight instead of starting over.
- Needs your decision
- In progress
- Completed
- Failed
Human in the loop
Nothing irreversible happens without you.
Set the policy per agent: approve everything, or only the calls that touch something sensitive. When an agent reaches one, it stops and asks — showing you the exact tool and the exact arguments, plus why it wants to run it.
A waiting agent costs nothing. It can sit suspended for three days at zero compute and resume the moment you tap approve, from whatever device you happen to have.
This isn’t blind autonomy. It’s delegation with supervision — how any team works.
Recurring work
The Monday report writes itself on Sunday night.
Put a task on a schedule and it runs without you: every morning, every Monday, the first of the month. It runs in the cloud, on time, whether or not your laptop is open — and the result is in your inbox before you are.
Runs are idempotent by design, so a retry never doubles the work.
The sandbox
Every agent gets a computer of its own.
Most assistants can only call the tools someone built for them. Cronspace hands the agent a throwaway Linux machine where it can safely write code in isolation, run it and look at the result — which is the difference between describing the work and doing it.
Runs real code. The agent writes Python and executes it, reads what came back, and decides what to do next. That is the difference between an assistant limited to the tools we shipped and one that can build the tool it needs — reshape a messy export, reconcile two spreadsheets, get the arithmetic exactly right instead of estimating it. pandas, numpy and matplotlib are already installed, and anything it writes to outputs/ comes back as a file you can open.
Capabilities
Give an agent the tools for the job.
Code
Runs Python in an isolated sandbox — clean data, crunch numbers, produce a file.
Browser
Opens the web from that same sandbox to read, check and gather what it needs.
Files
Attach images, PDFs, documents and datasets to a conversation; the agent works from them.
Artifacts
Whatever a run produces — a report, a CSV, a chart — is saved as a file you can open and share.
Shopify
Connect a store and ask about sales in plain language, without touching an analytics dashboard.
Meta Ads
Connect your ad accounts and ask what's working, what's burning budget, and what to do about it.
Your own tools
Connect any MCP server and its tools become the agent's. If one goes down, the run keeps going.
Pricing
No weekly limit. No “try again in four hours.”
Chat plans meter you. You hit the cap in the middle of the thing you were doing, and the answer is to wait for a window to reopen — usually on the afternoon you were getting somewhere.
Cronspace charges for what you actually run. A heavy week costs more than a quiet one, and neither one gets cut short.
It matters most for the runs you aren’t watching. A scheduled task at 06:30 doesn’t have to check whether you burned an allowance yesterday: if there are credits in your account and it’s inside the limits you set, it runs.
Full pricing goes up before we open the doors. Pay as you go is the model it will describe.
Control
Delegation you can audit.
Approval policies
Per agent, per tool. Nothing irreversible slips through.
Spend limits
Pay as you go doesn't mean unbounded. Every agent has a monthly ceiling you set and a live cost breakdown. A long run that crosses it stops mid-flight instead of quietly draining the budget.
Encrypted credentials
Store tokens are encrypted at rest and decrypted only inside the step that needs them — never in the agent's context, never in a log.
A full record
Every run keeps its steps, tools, arguments and outputs. You can always answer “what exactly did it do?”
Not locked to one model. Or to our screens.
Each agent runs on the model you choose, from any provider — switching is one field, not a migration. When the models get better, so does your agent, for free.
And everything the interface does, the REST API does too: create agents, launch runs, resolve approvals, read costs. Workspace API keys included.
Your data
Your work never trains a model.
We only work with companies that commit to that in writing. If one will not, we do not use it — not in chats, not in scheduled runs, not for generated images.
It cost us a model. One image model had no company willing to make that promise, so we removed it rather than write an exception into our policy.
How we handle your dataFAQ
Questions worth asking.
- How is this different from ChatGPT or Claude?
- Those are chats: you're present, you prompt, you get an answer. Cronspace is where work runs — on a schedule, in the cloud, without you — and lands in an inbox. The chat is one way in, not the product. And it isn't sold by the plan: you pay for what you run.
- Are there usage limits?
- No plan caps and no weekly windows. Cronspace charges for what you actually run, so a heavy week costs more than a quiet one and neither gets cut off mid-task. The only ceilings are the spend limits you set per agent and the credits in your account — a scheduled run inside both of those always runs.
- Can I trust an agent with real work?
- Only as far as you decide. Nothing irreversible runs without approval, every agent has a spend ceiling, and every step is recorded. You choose how long the leash is.
- What happens if it fails halfway?
- Runs are durable. A crash, a deploy or a closed laptop doesn't kill one — it retries and continues. Failures land in your inbox instead of disappearing.
- Which model does it use?
- Whichever you pick, per agent, from any provider.
- Do you train on my data?
- No. We only work with companies that commit to that in writing, and if one will not, we do not use it — not in chats, not in scheduled runs, not for generated images. It cost us a model: one image model had no company willing to make that promise, so we removed it rather than write an exception into our policy.
- How do I get in?
- Cronspace is in private beta and accounts need an invite code. If you have one, you can create your account now. If you don't, join the waitlist and we'll email you when your spot opens.
Get in line.
Cronspace is open to a small group while we build in the open. Join the waitlist and we'll email you when it's your turn.