cronable
AI agent scheduler

The reliable layer for your AI agents.

Coding agents like Claude Code, Codex and Gemini CLI can do real work on your machine. What they don’t come with is a schedule, a trigger, a next step or a record. Cronable runs them headlessly as jobs on your own machine or server, every night, on a GitHub event or when a file lands. It chains what happens next on success, retries and alerts on failure, and keeps a full history of every run. Describe the workflow in a sentence and the assistant builds it.

On the canvas

A nightly agent run, end to end

Pull the repo, let the agent work, push the branch. If the agent fails or times out, the push never happens and you get the error instead.

jobs/agents
Integrations

The tools your week is made of

The agents you already use, the models behind them and the places their work ends up. These are the integrations this page uses.

Claude

Claude (CLI)

Runs Claude Code headlessly (claude -p) in a working directory with your model and effort settings, and keeps its final answer as the run’s output.

Integration details

Codex (CLI)

Runs codex exec in a repo with the sandbox you choose: read-only, workspace-write (the default) or full access.

Google Gemini

Gemini (CLI)

Runs Gemini CLI headlessly with the approval mode you pick. Actions that would need approval are denied in headless mode.

Claude

Claude (API)

One call to the Anthropic API, text in and text out, for classifying, summarizing or drafting inside a chain.

Integration details

OpenAI

One Chat Completions call, optionally with tools you grant it: folders it may read, hosts it may fetch, jobs it may run.

Integration details
Ollama

Local Model (Ollama)

A model on your own hardware. The prompt and the reply never leave the machine.

OpenRouter

OpenAI-Compatible (Gateway)

Any OpenAI-compatible endpoint, from a gateway like OpenRouter to your own model server.

Git

Git

Pull, commit and push with the machine’s own git credentials. No tokens are stored in the job.

GitHub

GitHub

Start a chain on pushes, pull requests, issues, releases and workflow runs, and create issues, comment or dispatch a workflow.

Integration details

Folder watch

Starts a run when a file lands in a folder you choose and has finished writing.

Slack

Post the agent’s answer, the branch name or the failure where the team will see it.

Integration details
Notion

Notion

Create a page, or append to one, with the agent’s write-up.

Integration details
Workflows

Four agent workflows, step by step

Each one is a chain of jobs. The agents run as Claude (CLI), Codex (CLI) or Gemini (CLI) jobs, and lighter AI work runs as model steps.

A branch waiting for review every morning

When the repo goes quiet for the night…

  1. 1A daily schedule at 02:00 runs a Git step that pulls the latest main.
  2. 2Only if the pull succeeded, a Claude (CLI) step runs Claude Code in the repo with your instructions: take the top issue labelled for the agent, implement it on a branch, run the tests.
  3. 3On success, a Git step pushes the branch, and the agent’s final answer, what it changed and why, goes to Slack.
  4. 4If the agent fails or hits its timeout, its whole process group is stopped, the push never runs, and a failure handler posts the error.

You review a branch over coffee instead of babysitting a terminal.

Every new issue triaged within minutes

When someone opens an issue on GitHub…

  1. 1The GitHub webhook trigger fires on the issues event, delivered through the hosted relay.
  2. 2A Claude (API) step reads the title and body, labels it a bug, a feature or a question, and drafts a first reply as JSON.
  3. 3A Code step parses the label, and a Switch step routes on it.
  4. 4Bugs get a GitHub comment asking for the missing details. Everything else goes to Slack for a maintainer.

Reporters get an answer the same hour, and maintainers see a sorted queue.

A weekly security sweep with Codex

When the week wraps up…

  1. 1A weekly schedule runs a Codex (CLI) step in the read-only sandbox: audit the code and dependencies, and return the findings as JSON.
  2. 2A For each step runs once per finding, and a GitHub step opens an issue for each one.
  3. 3A Collect step waits for the last one and counts what was filed.
  4. 4A Slack step posts a one-line summary of the sweep.

Findings arrive as issues you can assign, not a wall of terminal output.

Private documents, summarized on your own hardware

When a new file lands in a folder…

  1. 1A Folder watch trigger fires once the file has arrived and finished writing.
  2. 2A Local Model (Ollama) step, given read access to that one folder, summarizes the document. The full text never leaves the machine.
  3. 3A Notion step files the summary as a new page in your team’s database.

A model on your own hardware reads the document. Only the summary goes where you send it.

What changes

What changes when agents have a scheduler

Agent work stops depending on an open terminal and a good memory.

Agents work when you’re not there

On a clock, on a GitHub event or on a new file, as jobs on your own machine, under the logins it already has.

Only good runs move on

The next step fires only when the agent’s run succeeded. A failure or a timeout stops the chain and runs a handler instead.

Every run on the record

Each agent run keeps its status, timing, the full streamed log and its final answer. You can see what it did last Tuesday, and what it said about it.

Costs you can predict

Cronable doesn’t meter runs — except the AI Assistant job, which shares the daily assistant allowance. Agents use your own Claude, OpenAI or Gemini account, so the model bill comes straight from your provider, with no markup.

The numbers you report on

  • Agent branches merged
  • Time from issue opened to first response
  • Agent runs that fail or time out
  • Security findings opened and closed
  • Hours of agent work per week

We don’t promise figures. Cronable records every run, so you can measure these yourself, before and after.

Questions

Running AI agents, answered

Which AI agents can it run?

Three coding agents run as their own job types: Claude (CLI) runs Claude Code, Codex (CLI) runs OpenAI’s Codex, and Gemini (CLI) runs Google’s Gemini CLI. Each runs headlessly in the working directory you set, under the account the machine is already signed in to. For lighter steps, model jobs call Claude, OpenAI, Gemini, a local Ollama model or any OpenAI-compatible endpoint, and you can give them tools: folders to read, hosts to fetch from, jobs to run.

How much can an agent do on my machine?

As much as the account Cronable runs as. Claude (CLI) runs with bypassPermissions, so it can read and change anything that user can; the working directory is where it starts, not a sandbox. Codex runs in the sandbox mode you pick. Gemini CLI’s approval mode decides what it may do without asking, but it is not a sandbox. On a shared server, run Cronable as a dedicated low-privilege user. Business adds SSO, role-based access, the audit trail and four-eyes change approval (when you turn it on), so no one adds or changes an agent job alone, and exec-tiering limits jobs that run commands to admins.

Where do API keys and secrets live?

On your machine. The CLI agents use the login or API key their own CLI already has, and Cronable stores nothing for them. Keys for model APIs and integrations are encrypted with AES-256-GCM in a local database, referenced by name, and masked out of logs, output and the live stream.

What happens when an agent run fails or hangs?

Each job has a timeout. When it’s reached, or you cancel the run, Cronable stops the agent’s whole process group, including any tools it spawned. The run retries if you set a retry policy, and then the failure handler runs: post to Slack, open an issue or email you. Steps that needed the agent to succeed never start.

Can I see what an agent did last month?

Yes. The run record, with status, timing, trigger and the agent’s final answer, is kept until you choose to prune it. The full streamed log is kept for 30 days by default, or longer if you set it. Changes to the workflow itself are commits in your git repository.

What does the built-in assistant do?

It builds workflows. Describe one in a sentence and it drafts the jobs, schedule and failure handling on the canvas for you to review. It has no shell, browser or file tool and never sees a secret value. Anything it runs, and any change that could make a job run on its own, asks you to confirm first, and goes through the same roles and approvals as your own changes. Use the hosted Cronable assistant, your own Anthropic or OpenAI key, or a local CLI. Want a reviewer too? An AI Assistant (scheduled) job looks over your jobs and runs on a schedule and summarizes what it finds, read-only by default.

Why not just use cron?

Cron starts a command. It doesn’t start the next step only on success, retry with backoff, mask secrets in the output, alert you on failure, or keep a history of what the agent said that you can open later. Cronable does all of that, and you edit it on a canvas instead of a crontab.

Other solutions

IT & operationsFinance & accountingE-commerce

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.

Join the waitlist

It used to mean “schedulable”. Now an agent pushes a branch before you wake up. Still cronable.