Gumloop
The AI agent you train once — and it repeats the task on its own from then on.
By Thiago Lourenço Martins
If you only read one section, read this one
score 3.9 out of 5
I'll be direct: Gumloop is good for people who already know exactly which task they want off their plate and have the patience to configure it properly. Its main trick — the agent rewriting its own instructions once you correct a mistake — genuinely works, and it's the kind of thing that stops you from re-explaining the same thing every week.
The catch is credits. The math looks simple until you run a data-enrichment task: enriching 333 contacts in one go is enough to wipe out an entire Pro plan's monthly credits, by itself. I wouldn't leave a heavy routine like that running without doing the math first.
Tested and checked against official sources on July 21, 2026.
What works in its favor
- The agent writes its own instructions: you correct it once, ask it to remember, and it won't make the same mistake again — no manual prompt editing
- Connects to 150+ work tools (Gmail, Slack, Salesforce, Notion, spreadsheets) — and most of them run at no credit cost, only the AI model conversation is billed
- The free plan is enough to actually test it: 5,000 credits a month, unlimited agents and workflows, no credit card required
- Besides the conversational agent, there's a visual drag-and-drop canvas for anyone who'd rather draw the flow node by node instead of just chatting
Where it pinches
- Credits are the currency of everything and they run out fast on heavy tasks: enriching 333 sales contacts, for example, wipes out an entire month of the $37 Pro plan
- Cost swings from model to model — the same task can burn 3 to 4 times more credit if you switch to a pricier AI model, and that isn't obvious when picking one
- On the free plan, only 1 active trigger and forum-only support — for real, hands-off automation, you hit the ceiling fast
- Real learning curve: this isn't an "open it and go" tool like a plain chat app — understanding connectors, credits, and the agent-vs-workflow split takes some time
A workbench for building your own robot
Gumloop is an AI automation platform. Picture a workshop bench: instead of a hammer and screws, you have pieces that read email, write to a spreadsheet, send a Slack message — and you snap them together until it becomes a task that runs on its own. In practice, today Gumloop leans less on a fixed flowchart and more on the agent: you chat with it in plain text, explain what you need, and it decides on its own which tool to use to get the work done.
You can use both approaches, even together: an agent can call a ready-made workflow mid-conversation, whenever the task is always the same and needs no decision at all.
The tab you reopen every Monday morning
Pulling that report every Monday. Answering the same question in Slack for the fifth time this month. Copying data from one system to another by hand, knowing it's silly but nobody's had time to automate it yet.
That's the number of work tools Gumloop already knows how to use — Gmail, Slack, Google Sheets, Salesforce, Notion, and more. You connect only the two or three your routine actually uses, explain what you need once, and the agent starts doing it on its own — even adjusting how it works when you point out a mistake.
Who puts Gumloop to work behind the scenes
Every Monday you had to open three spreadsheets to put together the week's summary for your team. Now an agent reads the numbers on its own and has the summary ready in Slack before your first meeting of the day.
A support agent triages incoming tickets, opens the right one, and spots patterns in repeated complaints — without anyone on the team reading message by message.
Before calling a lead, the agent has already researched the company, summarized the conversation history, and updated the CRM on its own.
From blank screen to your first agent running
- Create a free account at gumloop.com
Click "Get Started". No credit card required.
- In the side menu, click "Create Agent"
It's the button that already shows up front and center on your account's Home screen.
- In the Connectors section, click "+ Connector"
Only connect the tools this agent will actually need.
start with 2 or 3, not your whole company at once - Give it a real task in the agent's chat
Ask it to do something from start to finish, and watch where it slips up, stalls, or asks something silly.
- Correct it, and ask it to remember the lesson
Write something like "update your instructions to always do it this way" — the agent rewrites its own internal prompt.
this is an actual product feature, called self-updating instructions - If you want it to run on its own, click "+ Trigger"
Choose between a scheduled trigger, an event from another app, or a one-time run.
the free plan only lets you keep 1 trigger active
A note it will never forget again
Every Monday morning, pull last week's numbers from our sales spreadsheet, put together a summary with up to 5 bullet points covering the highlights and the points that need attention, and send it to the right Slack channel. If any number drops more than 15% compared to the previous week, tag me directly in the message.
It runs the task once, you correct whatever didn't come out right, and ask it to remember the instruction. By next Monday, it just does it on its own, the way you agreed.
Where credit quietly leaks away
Do this
- Start with just 2 or 3 connectors — every extra tool makes the agent slower and more expensive to run
- In visual workflows, use the Resume button to pick back up from a specific node instead of rerunning everything — it saves credits while you're testing
- Ask the agent itself to turn something it got right into a "skill" — that way it reuses that way of working in future conversations, without you re-explaining it
Avoid
- Running bulk contact enrichment without watching the cost: each enriched contact eats 60 credits — 333 contacts at once wipe out an entire Pro plan's monthly credits
- Switching AI models without thinking about cost: a pricier model can burn 3 to 4 times more credit on the exact same task
- Expecting fast support on the free plan: there, support is community-forum only, no chat or direct email with the team
Teach it to do one boring thing for you
Pick that task you do every week without even thinking about it — pulling a number, answering a standard email, updating a spreadsheet. Create the free account, click Create Agent, connect only the tool that task uses, and describe the steps to it in plain, everyday language.
It worked if, the next time you ask for the same thing, it already remembers how you like it done — without you explaining it from scratch.
Is this robot worth building for your case
Definitely worth it if you…
- Manage a small team, are tired of pulling the same report every week, and want something that learns the right way to do it once and repeats it on its own
- Already use several work tools (spreadsheet, Slack, CRM) and want to connect them without hiring a developer
- Like testing and adjusting gradually — the product rewards people who correct the agent and let it learn over time
Think twice if you…
- Only need a simple, one-off automation — Zapier charges per task executed and tends to be cheaper for that
- Don't have the patience to track credit spend — under heavy use, the bill can catch you off guard
- Want to run heavy data enrichment (leads, companies) without calculating the cost beforehand — that gets expensive fast here
Where free ends and credits start counting
It's free to create an account and build unlimited agents and workflows, no card required — with 5,000 credits a month to spend. The catch is that how fast you burn those credits varies a lot by task, so "free" here isn't the same as "unlimited".
$0 — 5,000 credits/month, 1 seat, 1 active trigger, 2 concurrent runs, unlimited agents and workflows, forum support.
from $37/month — 20k+ credits/month, unlimited seats, 5 concurrent runs, unlimited teams, hosting for 1 MCP server.
custom pricing — role-based access control, SSO/SAML, admin dashboard, virtual private cloud (VPC), full audit logs.
Important detail: Gumloop bills in US dollars, and the Pro price goes up depending on how many credits you pick on the site — the $37/month is just the entry point, for 20,000 credits. Official prices as of July 2026 (subject to change).
When a competitor fits better than Gumloop
If what you want is a simple, single-purpose automation, without having to think about credits per AI call, Zapier is still more straightforward: it charges per task executed, not per token consumed, and the learning curve is much shorter. See the Zapier walkthrough →
If your team has someone who codes and wants full control over every step of the flow, n8n tends to come up in comparisons as the more flexible option for tinkering with each node's code.
And if the problem is simply "I need a chatbot that answers questions on my site or WhatsApp", Chatbase solves that on its own, without all of Gumloop's connectors-and-credits machinery. See the Chatbase walkthrough →
Questions people have before signing up
Is Gumloop really free?
Yes. The Free plan gives you 5,000 credits per month, unlimited agents and workflows, and no credit card required. The catch is that credit usage varies a lot by task — a simple question costs little, but an agent that calls several tools in a row burns through credits faster.
Do I need a credit card to sign up?
No. Signing up for the free plan doesn't ask for a card. It only comes into play if you decide to subscribe to the Pro plan, at $37 per month.
Does it work in Portuguese or other languages?
The website's interface is in English. But since you configure the agent by writing instructions and chatting in free text, you can do everything in Portuguese, Spanish, or any language you like — it understands and replies in your language without issues.
Does it work on mobile?
It's a 100% web-based tool, so it opens fine in a mobile browser. Chatting with an agent you've already set up works well; building a complex flow on the drag-and-drop canvas is more comfortable on a computer screen.
What's the difference between an "agent" and a "workflow" in Gumloop?
A workflow follows a fixed path that you draw node by node, always in the same order. An agent is conversational: you give it a goal and it decides on its own which tools to use and in what order, adapting to the situation.
Is my data safe?
According to the official site, the company doesn't use customer data to train AI models, keeps zero data retention agreements with third-party model providers, and is SOC 2 Type II certified and GDPR compliant.
Do unused credits roll over to the next month?
No. According to the official documentation, credits don't roll over from one month to the next — the exception is Enterprise plans, which are negotiated separately.
Every day, a new tool explained.
Get it on WhatsApp