Cronloop Review (2026): Pricing, Features & Honest Verdict

Cronloop runs AI agents on a schedule, from every five minutes to weekly, using Codex or Claude Code with memory that persists between runs. We found honest, cheap orchestration with one catch: the $25 buys scheduling only, and your model bill is separate and far larger. Best for: developers automating recurring work. Price: From $20/mo (free plan: yes). Rating: 7.5/10.

What is Cronloop?

Cronloop is cron for AI agents. You describe a job in plain Markdown, pick Codex or Claude Code as the engine, connect the tools it needs, set a cadence between five minutes and a week, and it runs. Each agent keeps a durable memory between runs, so run 200 knows what run 199 did. You watch each run live and read the full history afterwards.

The company is Cactal, Inc. and the product launched on Product Hunt on 19 August 2026. It is a small operation with one publicly listed maker, Mike Tromba, though it is a properly incorporated company with a real product rather than a weekend project.

The examples on the site are concrete in a way most agent products are not. Book outbound demos every 30 minutes using Gmail, Apollo, HubSpot and Calendar. Answer support tickets every 5 minutes across Zendesk, Intercom and Notion, escalating anything sensitive to a human. Read Search Console every morning and ship content updates to WordPress. Pick up the next assigned Linear issue every hour and open a pull request with tests. These read like jobs someone actually runs, not demo fodder.

The framing we like most is the simplest one on the page: if you can write the job description, you can run the agent. That is the correct mental model for this product, and it sets expectations honestly about who it is for.

What are Cronloop’s key features?

Scheduled runs with durable memory

Cadence runs from every five minutes to once a week. The memory layer is the part that makes scheduling useful rather than repetitive: each agent carries state forward, so a competitor monitor knows what it already reported and a prospecting agent knows who it already emailed. Without persistent memory, a scheduled agent just redoes the same work and floods you with duplicates. Cronloop exposes memory as a first-class thing you can read, write and clear.

An MCP server with a real tool surface

You can manage the whole system from inside ChatGPT or Claude rather than the web app, through a properly enumerated MCP tool set. Agents get list, get, create, update, pause, resume and delete. Runs get start, list, review, get and cancel. There are configuration, environment and secret operations, and memory gets list, read, write and delete. This is a more complete surface than most products ship, and it means Cronloop can be driven by another agent, which is the obvious use once you have more than a handful of jobs.

Secrets that cannot be read back out

Credentials and settings are encrypted on the way in and, in the vendor’s words, can never be read back out. Write-only secret storage is the correct design for a system that runs unattended jobs against your production tools, and it is a detail plenty of larger competitors get wrong. Combined with per-agent connections, it means a compromised agent config does not hand over your API keys.

Live output and full run history

Every run streams its output while it happens and stays in the history afterwards, with a per-run record of what the agent actually did. The demo shows runs that succeeded, runs that booked something, and a run that was skipped because the previous one was still going. Showing the skip case in your own marketing is a small honesty signal, because overlapping runs are the first thing that breaks a scheduled agent and most vendors hide it.

Does the $25 actually cover what Cronloop costs to run?

No, and to Cronloop’s credit it says so plainly. The pricing section states that runs use your existing ChatGPT or Claude subscription, or your own Anthropic or OpenAI API key, and that model usage stays with your chosen provider connection. Cronloop bills you for orchestration. Inference is your bill, paid to someone else.

That is the honest model and we prefer it to a marked-up credit. But it means the headline price describes a small fraction of what running Cronloop costs, so it is worth doing the arithmetic before you commit.

Take Pro’s fastest cadence. Every five minutes is 288 runs a day, or about 8,640 a month, per agent. Anthropic’s published API rates are $3 per million input tokens and $15 per million output for Claude Sonnet 5, and $5 and $25 for Claude Opus 5. Assume a modest run: 20,000 input tokens of context and instructions, 2,000 tokens of output. On Sonnet 5 that is $0.06 of input plus $0.03 of output, about $0.09 a run. Multiply by 8,640 and one agent at full cadence costs roughly $780 a month in tokens. On Opus 5 the same agent is around $1,300.

So the $25 subscription is about 3% of the true monthly cost of one agent running flat out. That is not a criticism of the price, which is cheap for what it does. It is a warning about the mental model. If you read “$25 a month, no fixed limits” and picture a fixed bill, you have misread it. The variable cost sits with your provider and scales with cadence, context size and how many agents you run.

The practical advice: start slow. An hourly agent is twelve times cheaper than a five-minute one and is fast enough for almost every job on Cronloop’s own examples list. Reserve the five-minute cadence for the support-ticket case where latency actually matters, and keep the context you feed each run tight, because input tokens dominate the bill at these volumes.

What does Pro actually remove?

Cronloop has only two tiers, and Pro’s headline benefits are phrased as the absence of limits rather than the presence of allowances. The card reads “No fixed agent limit” and “No fixed monthly run-minute limit”. Those are not the same as unlimited, and no fair-use number is published anywhere.

That leaves a gap. A buyer running twenty agents cannot tell from the pricing page whether they are a normal customer or about to get an email. There is no stated ceiling, no overage rate, and no volume table. We would rather see an honest large number, say 10,000 run-minutes with a published rate above it, than a phrase that defers the question. To be fair, the risk here is capped by the inference economics above: your provider bill will discourage runaway usage long before Cronloop’s soft limits do.

The free tier’s arithmetic deserves a closer look too, because it is tighter than it appears. Free allows up to 3 agents, runs as often as hourly, one active run at a time, runs up to 10 minutes, and 200 run-minutes a month. Hourly for a month is about 720 runs, but 200 run-minutes across 720 runs averages under 17 seconds per run. And since a single run can last 10 minutes, twenty full-length runs would exhaust the month entirely.

So the free tier is a genuine trial rather than a workable free plan. It is enough to build an agent, watch it run, and decide whether the product does what you want. It is not enough to keep one useful job running continuously. That is a reasonable place to draw the line, and the numbers are all published, but read them before you plan around them.

How much does Cronloop cost?

Cronloop is $25 a month for Pro, or $20 a month billed annually, and there is a free tier. Two tiers, no enterprise plan, no seats. Model usage is billed separately by your AI provider on every tier including Free.

Free at $0 gives up to 3 agents, hourly cadence, 200 run-minutes a month, a 10-minute maximum run length, 1 active run at a time, provider subscriptions or API keys, and live output with run history. Pro at $25 or $20 annually gives no fixed agent limit, five-minute cadence, no fixed monthly run-minute limit, runs up to 60 minutes, 3 concurrent runs, the same provider options, and priority support.

The annual discount is 20%, which is modest against the 40% several competitors offer but is at least real and clearly stated. What Pro really buys is the four constraints that make the free tier a trial: cadence from hourly to five minutes, run length from 10 to 60 minutes, concurrency from 1 to 3, and the run-minute budget. For anyone running a real job, all four bind quickly, so the upgrade is not an upsell so much as the actual product.

Against the field, $25 is cheap. Lindy charges $29.99 a month per user for 3,000 credits with model usage included. n8n’s own pricing payload showed roughly EUR 24 a month for 2,500 executions when we read it, though note that n8n’s pricing page geolocates and we were reading from an Irish IP, so your currency and figure may differ. Zapier offers a free tier with 100 tasks a month and prices paid plans on a task-volume slider rather than publishing a flat number.

The comparison is not quite like for like, and the difference cuts both ways. Lindy’s credits include inference, so its $29.99 is closer to an all-in number, while Cronloop’s $25 is not. But Cronloop takes no margin on tokens at all, so a heavy user who brings an efficient model and tight context will pay less in total, and a light user on a bundled competitor may pay less than both.

Who is Cronloop best for?

Use Cronloop if you are a developer or technical operator with recurring jobs that need judgement rather than a fixed script. Triaging tickets, reviewing changelogs, opening pull requests from issues, monitoring competitors. If a plain cron job would work, use plain cron, because it is free and deterministic. The moment the task requires reading something and deciding, this is the cheaper way to run it than building your own harness.

Use it if you already pay for Claude Code or Codex and want more out of that subscription. Bringing your own provider connection on every tier, including Free, with no markup, is the most buyer-friendly thing about the product.

Use it if you want to drive automation from inside your chat client. The MCP surface is complete enough that managing Cronloop from Claude or ChatGPT is a real workflow, not a checkbox.

Skip Cronloop if you want one predictable monthly invoice. The subscription is predictable and the inference bill is not, and for some finance teams that alone is disqualifying. A bundled competitor is the better fit.

Skip it if you are not technical. Jobs are written in Markdown and the debugging surface is run logs. There is no visual builder, and the product does not pretend otherwise.

Skip it if you need team features. There are no seats, no roles and no shared workspace at any price. This is a single-operator tool today.

What are the best Cronloop alternatives?

Lindy is the closest managed competitor at $29.99 a month per user for 3,000 credits, with inference included in the credit. You give up model choice and pay a margin on tokens. You get one bill, a friendlier builder and team-oriented plans. For a non-technical operator who wants agents that just run, Lindy is the easier purchase, and the total cost at low volume will usually be lower.

n8n is the power option, a workflow automation platform with hundreds of integrations, AI nodes and a self-hosting path that removes the subscription entirely. Its cloud pricing showed around EUR 24 a month for 2,500 executions when we checked, with the geolocation caveat above. It is far more capable than Cronloop and far more work to set up. Choose it if your jobs are mostly deterministic with some AI in the middle, rather than mostly AI.

GitHub Actions with a scheduled workflow is the free answer for developers already in GitHub. A cron-triggered workflow calling Claude Code costs nothing beyond your Actions minutes and your provider bill. You build the memory, run history and secret handling yourself, which is precisely what Cronloop’s $25 saves you. If you have one job, do this. At five jobs, Cronloop wins on time.

Final verdict: is Cronloop worth it?

Cronloop does a narrow job well and prices it honestly. Durable per-agent memory is the feature that separates a useful scheduled agent from a duplicate-generating nuisance, the MCP surface is complete enough to manage the product from another agent, write-only secret storage is the right call for unattended jobs against production tools, and showing a skipped overlapping run in the marketing screenshot is the kind of detail that suggests the team has actually operated this. Bringing your own provider connection with zero markup, on every tier including Free, is more generous than most competitors and means your costs fall as models get cheaper rather than staying pinned to a vendor’s credit rate.

The reservations are about what is not published rather than what is wrong. Pro’s two headline benefits are the removal of limits with no stated ceiling and no overage rate, so heavy users cannot predict where the soft wall sits. The free tier’s 200 run-minutes make it a trial rather than a free plan, which the numbers show but the framing does not. There are no team features at any price and one publicly named maker behind the company. And the biggest thing to understand before buying is that $25 covers roughly 3% of what a single agent at full cadence actually costs once inference is counted. Worth it at $20 annually if you are technical, already pay a model provider, and have three or more recurring jobs that need judgement. Start on Free, run hourly rather than every five minutes, and watch your provider bill for the first month.