Cronloop

About Cronloop

Cronloop lets you build autonomous AI agents that run in a loop on a schedule — anywhere from every five minutes to once a week — so recurring work happens on its own while you sleep. Instead of prompting an AI once and manually deciding every next step, you describe the job in plain Markdown, choose Codex or Claude Code as the engine (running on your existing subscription or your own provider API key), connect the tools the agent needs, and set a schedule. From that point the agent works independently: it observes the current state, reasons about what to do, calls the tools it needs, records what happened, and picks up where it left off on the next run.

What makes Cronloop different from a plain cron job or a stateless automation is memory and intelligence. Every agent runs inside an ephemeral, isolated sandbox and keeps a durable, Markdown-based memory between runs — so it can compare the current run to previous ones, adapt its behaviour, and self-improve over time by recording learnings that other agents can benefit from. You can watch each run live as the agent's logs stream into the app, and it intelligently skips a scheduled run if the previous one is still working, avoiding overlap.

With 200+ ready-to-go connectors, agents can plug into virtually any tool that exposes an MCP, CLI, SDK or API — so a single agent can research prospects and log them in HubSpot, triage and answer Zendesk tickets from your help docs, autonomously SEO-optimize a website using Search Console data, monitor deployments, or run daily engineering checks. Agents are also reachable directly from the assistants you already use: add Cronloop as a custom MCP connector in Claude or ChatGPT, approve access via a standard OAuth sign-in, and ask in plain language for a new agent, a fresh run, or a post-mortem on last night's failures — no dashboard required.