Keep Claude Code. Give it a schedule, a chain and a run history.
Even n8n tells people working alone to just use Claude Code. Cronable is built on the same idea. It does not replace Claude Code — it runs it, headlessly, on your own machine, and adds the operation around it: when it runs, what fires it, what runs next, and what happened last night.
The short answer
Claude Code is a superb agent, and it can already schedule itself. Desktop scheduled tasks run on your machine while the app is open and the computer is awake. Cloud scheduled tasks run on a fresh clone of your GitHub repositories — in Anthropic’s cloud, or for Team and Enterprise on a self-hosted environment inside your own network (public beta) — and can start from a schedule, an API call or GitHub events. And /loop repeats a prompt inside an open session. Each of those runs one saved prompt as its own agent session.
Cronable is what you reach for when the agent becomes part of a process. A Claude step fires on a schedule, a webhook, a new file or a new email. It runs only if the step before it succeeded, and its answer flows into the next step. A failure retries with backoff, then goes to a handler job and an alert. Every run is recorded with live, secret-masked logs, next to the shell, HTTP and integration steps that make up the rest of the workflow.
It all runs on a machine you own — a Mac or Windows app, or a server that keeps going when your laptop is closed — using the Claude login already on that machine. And you do not wire it by hand: describe the workflow and the built-in assistant builds it.
Cronable vs Claude Code
Where Cronable and Claude Code actually differ, capability by capability.
| Capability | Cronable | Claude Code on its own |
|---|---|---|
| Runs Claude Code unattended | Yes — a Claude step runs claude -p in the folder you choose | Yes — Desktop or cloud scheduled tasks, /loop, or claude -p from cron |
| Works on your local files while your laptop is closed | Yes, when you install it on a server or an always-on machine | Desktop tasks need the app open and the computer awake; cloud tasks run with the lid closed, on a fresh clone of your GitHub repos, in Anthropic’s cloud or your own self-hosted environment |
| Run a step only if the previous one succeeded | Yes — chain on parent success and pass its output in | Not documented — each task runs on its own trigger |
| Steps that need no model | Shell, SSH, HTTP, SQL, email and 123 integrations, next to the agent | Every task is an agent session that does the work through its tools |
| What can start a run | Schedules, parent success, SaaS webhooks, new files, new mail, sheet rows and more | A schedule or Run now; API calls and GitHub events for cloud tasks; channels can push webhooks into a running session |
| When a run fails | Retry with backoff, then a handler job; a timeout stops the whole process tree | API calls retry automatically; the run shows as Failed in the task’s history |
| Failure alerts | De-duplicated alert emails, or any message you like from a handler job | No alert setting on a task; Projects (beta) send a desktop notification when a thread errors, and hooks can script one |
| Run history | Every run, with status, timing, output and live, secret-masked logs | Scheduled tasks keep a history of runs; claude -p leaves session transcripts on disk |
| See every automation at once | A live canvas of every workflow, coloured by state | A list of scheduled tasks; each run opens as its own session |
| Set it up by describing it | The assistant builds the job, its schedule and its failure handling | Ask Claude to create the task, in a Desktop session or with /schedule |
| Where the definitions live | One YAML file per job, in your git repo | Desktop: a prompt file per task, schedule kept in the app. Cloud: your claude.ai account |
| Share with a team | More users from Plus; Business adds SSO, role-based access, an audit trail and four-eyes change approval | Cloud tasks belong to your individual account and are not shared |
| Secrets | AES-256-GCM encrypted on your disk, masked from logs | Cloud tasks: network secrets the session never sees; locally, your own environment |
| What it costs | A Cronable plan sized by jobs, plus Claude usage on your own account | Included with your Claude plan; runs draw on your usage |
Runs Claude Code unattended
Cronable
Yes — a Claude step runs claude -p in the folder you chooseClaude Code on its own
Yes — Desktop or cloud scheduled tasks, /loop, or claude -p from cronWorks on your local files while your laptop is closed
Cronable
Yes, when you install it on a server or an always-on machineClaude Code on its own
Desktop tasks need the app open and the computer awake; cloud tasks run with the lid closed, on a fresh clone of your GitHub repos, in Anthropic’s cloud or your own self-hosted environmentRun a step only if the previous one succeeded
Cronable
Yes — chain on parent success and pass its output inClaude Code on its own
Not documented — each task runs on its own triggerSteps that need no model
Cronable
Shell, SSH, HTTP, SQL, email and 123 integrations, next to the agentClaude Code on its own
Every task is an agent session that does the work through its toolsWhat can start a run
Cronable
Schedules, parent success, SaaS webhooks, new files, new mail, sheet rows and moreClaude Code on its own
A schedule or Run now; API calls and GitHub events for cloud tasks; channels can push webhooks into a running sessionWhen a run fails
Cronable
Retry with backoff, then a handler job; a timeout stops the whole process treeClaude Code on its own
API calls retry automatically; the run shows as Failed in the task’s historyFailure alerts
Cronable
De-duplicated alert emails, or any message you like from a handler jobClaude Code on its own
No alert setting on a task; Projects (beta) send a desktop notification when a thread errors, and hooks can script oneRun history
Cronable
Every run, with status, timing, output and live, secret-masked logsClaude Code on its own
Scheduled tasks keep a history of runs; claude -p leaves session transcripts on diskSee every automation at once
Cronable
A live canvas of every workflow, coloured by stateClaude Code on its own
A list of scheduled tasks; each run opens as its own sessionSet it up by describing it
Cronable
The assistant builds the job, its schedule and its failure handlingClaude Code on its own
Ask Claude to create the task, in a Desktop session or with /scheduleWhere the definitions live
Cronable
One YAML file per job, in your git repoClaude Code on its own
Desktop: a prompt file per task, schedule kept in the app. Cloud: your claude.ai accountShare with a team
Cronable
More users from Plus; Business adds SSO, role-based access, an audit trail and four-eyes change approvalClaude Code on its own
Cloud tasks belong to your individual account and are not sharedSecrets
Cronable
AES-256-GCM encrypted on your disk, masked from logsClaude Code on its own
Cloud tasks: network secrets the session never sees; locally, your own environmentWhat it costs
Cronable
A Cronable plan sized by jobs, plus Claude usage on your own accountClaude Code on its own
Included with your Claude plan; runs draw on your usageWhat you get by running it on your own machine
Your AI agents, on a schedule
Run Claude Code, Codex or Gemini CLI headlessly in a folder on your machine, under the AI account you already use, with a timeout that stops the agent and everything it started. Its final answer becomes the run’s output and flows into the next step.
Chains that wait for success
A step fires only when its parent succeeded, receives the parent’s output, and a failure branches to a handler job instead of running the rest on broken state. Agent steps sit next to shell, HTTP, database and integration steps in one workflow, on one canvas.
Every run recorded, with live logs
Each run keeps its status, timing, output and secret-masked logs, streamed live while it happens. Chain a job to fire only when its parent succeeded, retry with backoff, and route a failure to a handler job instead of down the pipe.
Say what you want; it builds the pipeline
Type “every morning, check yesterday’s Stripe refunds and post the total to Slack” and the built-in assistant builds the whole job — the steps, the schedule, the credentials it needs and what happens when one fails — wired and ready to run. You review it on the canvas and change anything. On every plan, including Free.
SaaS webhooks reach your machine
Give Stripe, GitHub or a form tool a URL and the hosted relay forwards the event to your install — again with no port opened and nothing exposed to the internet. Each install gets its own partition, and jobs fire the moment the event lands. Basic and up.
Reach it from anywhere — no open ports
Remote control lets you drive your install from a browser wherever you are. The machine dials out to the hosted relay; you never expose a port, forward anything on your router, or run your own tunnel. It is opt-in, off until you switch it on, and included from Basic up.
Cronable and Claude Code
Does Cronable replace Claude Code?
No — it runs it. A Claude job starts the Claude Code CLI headlessly (claude -p) in the folder you choose, with the model and effort you pick, using the Claude login or API key already on that machine. Usage stays on your own Claude account. Cronable adds what sits around the agent: the schedule or trigger, the steps before and after it, retries, alerts and the record of every run.
Claude Code already has scheduled tasks. What does Cronable add?
Each of Claude Code’s scheduled tasks runs one saved prompt as its own agent session — locally while the Desktop app is open, or as a cloud session on a fresh clone of your repos. Cronable makes the agent one step in a workflow: it fires on a schedule, a SaaS webhook, a new file or a new email, waits for the step before it to succeed, hands its answer to the next step, and routes a failure to a handler. All of it runs on your own machine, with your local files, around the clock if that machine is a server.
n8n says “If it’s just you, use Claude Code or Codex.” Is Cronable for one person or a team?
Both. One person can run Cronable on Free with 6 jobs, or on Basic with 35, and keep using Claude Code for the thinking. When a team needs to share the same workflows, Plus adds up to 5 users, and Business up to 20 with single sign-on, role-based access, an audit trail and four-eyes approval of job changes.
Is it safe to let an agent run unattended?
Treat it like a shell. A Claude job runs with full permissions as the user Cronable runs as — it is not sandboxed and not confined to its folder — so give it instructions and a folder you would trust a script with, and never run Cronable as root. A timeout stops the agent and every process it started. Keep secrets out of the instructions: they are passed on the command line, and log masking is best-effort.
Can I set up the jobs by describing them?
Yes. The built-in assistant builds a job from a plain-English description on every plan, including Free. On Business you can also turn on Cronable’s MCP server and create and manage jobs from Claude or another MCP client — it is off until you enable it, and every change goes through the same roles and audit log.
What if my laptop is asleep when a job is due?
On a laptop, a job that fell due while it slept fires once when the machine wakes. For work that must run on the hour, every hour, install Cronable on a server or an always-on machine — the jobs, secrets and run history work the same.
Automate your operations.
Cronable is launching soon. Join the waitlist and we’ll email you the moment self-serve accounts — and the apps for Mac, Windows and Linux — go live.