Cronloop AI
AI agents that run in a loop
Turns Claude Code or Codex into unattended agents that run on a schedule, from every five minutes to weekly. Instructions are written as plain markdown, each run executes in an ephemeral isolated sandbox, and a durable memory carries state between runs. Tools connect over MCP, CLI, SDK and API, and inference is bring-your-own via an existing Claude Code / Codex subscription.
What is Cronloop AI?
Cronloop is a hosted scheduler and runtime for coding agents. You describe a job in Markdown, pick Claude Code or Codex, connect the tools it needs, and set an interval — as often as every hour on the free tier, every five minutes on Pro. Each agent carries a durable memory from one run to the next, so recurring work such as issue triage, support replies, or content updates continues without a human opening a terminal. Cronloop also exposes a hosted MCP server, so the same agents can be created, run, and reviewed from Claude, ChatGPT, or any MCP client that supports remote servers. It runs on your own Anthropic or OpenAI subscription or API key rather than reselling inference.
Key features
- Agents defined in plain Markdown — "An agent is whatever you describe in Markdown" — rather than in a visual builder or DSL
- Choice of Claude Code or Codex as the underlying coding agent, driven by your own provider subscription or API key
- Scheduled unattended runs: as often as every hour on the free plan, as often as every five minutes on Pro
- Durable per-agent memory carried between runs, readable and writable through the memory_list / memory_read / memory_write / memory_delete tools
- Live output and run history, with the ability to start a run on demand, follow it, cancel it, and review what recent runs actually did (runs_start / runs_list / runs_review / runs_get / runs_cancel)
- Encrypted secrets and environment settings that are write-only — "Values are encrypted on the way in and can never be read back out"
- Hosted MCP server with OAuth sign-in, so agents can be created, paused, resumed, or post-mortemed from Claude or ChatGPT without opening the dashboard
- Connects to anything with an API, MCP server, or CLI, with listed integrations spanning GitHub, GitLab, Linear, Jira, Notion, Zendesk, Intercom, Gmail, Google Search Console, Vercel, Sentry, Supabase, and more
Who it's for
- Turn assigned issues into pull requests on a schedule — pick up the next Linear or GitHub issue, write the fix and the tests, hourly
- Answer inbound support tickets from your own help docs and escalate anything sensitive to a human, every five minutes
- Run a recurring SEO loop that reads fresh Google Search Console data and ships measured content updates to the CMS each morning
- Watch competitors' changelogs, pricing pages, and launches and post a digest to Notion twice a day
- Drive scheduled agent operations conversationally from Claude or ChatGPT via the hosted MCP connector, instead of through a dashboard
When not to use it
If your work is interactive and you want to steer the agent turn by turn, a scheduler adds nothing — Cronloop's value is unattended repetition, and an agent you are already watching does not need a cron interval. It is also the wrong tool if you cannot supply your own Anthropic or OpenAI subscription or API key, since Cronloop schedules and runs agents but does not resell model inference.
FAQ
How much does Cronloop cost?
There is a free plan at $0/month covering up to 3 agents, an hourly minimum interval, 200 run-minutes per month, a 10-minute run limit, and 1 concurrent run. Pro is $25/month, or $20/month billed annually, and lifts that to unlimited agents, a 5-minute minimum interval, unlimited run-minutes, a 60-minute run limit, and 3 concurrent runs. Model usage is separate: both tiers run on your own Claude or ChatGPT subscription or provider API key.
Which coding agents and models does Cronloop support?
Cronloop runs Claude Code or Codex, using your own Anthropic or OpenAI subscription or API key. You choose the agent when you create the job; the instructions themselves are written in Markdown.
How are API keys and credentials handled?
Settings, environment variables, and secrets are set through Cronloop and, per the vendor, are encrypted on the way in and cannot be read back out — they are write-only once stored. Cronloop does not publish a third-party security audit, so treat this as a vendor statement rather than a verified control.
Can I use Cronloop from Claude or ChatGPT instead of its dashboard?
Yes. Cronloop ships a hosted MCP server that works with any MCP client supporting remote servers, including Claude and ChatGPT. You add Cronloop as a custom connector, approve access through a standard OAuth sign-in with no key to copy, and can then create agents, start runs, and review failures in plain language from the assistant.
Share this launch
Embed this badge
<a href="https://orangebot.ai/product/cronloop-ai" target="_blank" rel="noopener noreferrer"> <img src="https://orangebot.ai/api/badge/cronloop-ai.svg" alt="Featured on OrangeBot" width="200" height="54" /> </a>