How to Use Gumloop: From the 14-Day Trial to a Team Agent

Using Gumloop comes down to six moves: start the 14-day Pro trial, get a first agent from Gumball or the ready-made gallery, attach connectors on that agent, set what needs your approval, add a trigger, and read the credit cost of each run. The trial needs a card and turns into Pro at $37 a month for the whole organization's shared pool of 20,000 credits unless you cancel first ^[1]. For structured learning, Gumloop University's Getting Started course runs 6 lessons. Our view is simple. A first working agent comes quickly. The care goes into approvals and cost, the two settings that decide whether an agent stays useful after week two.

Start the trial with its end date in view

Every new account begins with the 14-day Pro trial. It needs a card, it turns into paid Pro unless you cancel before it ends, and it is one-time per customer, so plan the two weeks before you click. Canceling during the trial stops the card from being charged; you do it from Manage Subscription on the Usage and Limits page, which only an Admin can open inside an organization. Canceling is never immediate. Access runs to the end of the trial, and your agents, skills and chats stay in the account if you come back later. Once charged, paid fees are non-refundable under Gumloop's Terms except where the law requires otherwise.

Organizations and teams come with Pro, while the personal space exists on every plan. That matters on day one: if colleagues will use the agent, create the organization during the trial so you test sharing, shared credentials and the single credit pool the way they will really work. There is no separate bill for each person. Everyone draws on the same 20,000 credits, and those reset every month with no rollover outside Enterprise. Overage is off by default, so a team that empties the pool sees its agents stop rather than an unexpected invoice.

Get a first agent without a blank page

The quickest start is Gumball, the personal agent that comes with every account. It runs on your own connectors and skills, answers DMs to the Gumloop Slack app with no setup, and can build a custom agent from a plain-language request after showing you an approval card. It has two gaps. Gumball cannot delete agents or create app triggers, so you finish those parts in the agent editor yourself. Think of it as a fast first draft of an agent, not the whole build.

The other start is the use-case gallery, sorted into Marketing, Sales, Operations, Customer Support and Ecommerce. Clicking Use this agent creates a preconfigured agent in your workspace with its instructions loaded into the system prompt and its apps ready to connect. You authorize each tool with your own credentials, so the agent sees and changes only what you already can. A few gallery agents need nothing connected, such as the GEO audit agent and the SEO keyword research agent, because web search, web fetch and a code sandbox are built in. Our Gumloop templates guide walks through the gallery in detail.

Older tutorials talk about nodes and flows. Gumloop's older step-by-step builder, Workflows (once called Flows), still appears on the pricing page marked Legacy; Gumloop's current docs and product are built around agents. Custom nodes are part of that Legacy Workflows builder, and so are the resume and error-email tips on Gumloop's tips page. Gumloop says agents can call workflows, which lets a team keep existing ones behind a new agent. A new learner should start with agents.

Wire in the apps, then decide what needs a yes

Connectors are the most important part of an agent's setup, because they decide what it can reach. Gumloop offers 300+ built-in connectors, and the step people miss is this: the connector must be added on the agent itself, plus a sign-in. Connecting an app in Settings alone does not attach it ^[2]. Extra connectors are cheap to keep, since the agent searches its connectors and loads only the tools a step needs. By default each person who runs the agent uses their own account; team agents can use a team default, and you can pin specific accounts.

Then set approvals. Each connector has five modes: Always allow, Ask each time, Ask for writes/deletes, Never allow and Custom per tool. Always allow is the default. For anything that sends email, edits a CRM or deletes records, switch to Ask for writes/deletes, which lets read-only tools run freely and pauses every write for a click. Approval requests arrive in the chat, in notifications, on the Tasks page and as Approve and Reject buttons in a Slack thread. Rejecting is non-destructive, so a cautious setting costs you little.

Two self-improvement settings start on and deserve a look. An agent can edit its own instructions, and those edits keep no version history. New agents on Pro also start with Reflections, which review recent tasks and propose fixes; Auto-Apply Suggestions is the default for new agents and is not limited to low-risk changes, so it can change triggers and remove skills. Switch to the Review Queue if you want to approve each suggestion. For the model, new agents normally default to Auto, Gumloop's router, which picks a lower-cost model per message. Our Gumloop AI playbook covers skills, subagents and Brain in depth.

Give the agent a schedule or an event

A trigger is what turns a chat agent into one that works while you are away. Gumloop has four kinds: scheduled (recurring or one-time), app triggers that fire on events in services such as Gmail, Slack or Salesforce, webhooks with their own URL, and AI-created custom triggers. Recurring schedules can run as often as every minute, and you can describe the schedule in plain language while the AI writes the cron expression. When a trigger fires, the agent receives your configured prompt exactly as if you had typed it in the chat.

Three behaviors catch beginners. Triggers run with the credentials of the person who created them, not whoever chats with the agent, though team agents can use team credentials instead. A trigger switches off after 3 consecutive failed runs and notifies its owner. And if the previous run is still going when the next one fires, the new run is skipped. So test a trigger by hand first.

AI-created custom triggers need the most care, because they bill every poll whether or not they fire ^[3]. They poll every 5 minutes by default, which is also the shortest interval, and can be slowed to once a week. Left at 5 minutes, by our arithmetic, that is 288 polls and at least 288 credits a day, about 8,640 credits in 30 days, before any decision-model calls. That is a large share of a 20,000-credit month spent on watching alone. Pick the slowest interval the job tolerates.

Check the cost of a run before you share it

Credits vary by run. Gumloop's own examples put the same agent at 2 credits for a quick question and 200 for deep research, so test with real tasks rather than toy prompts. The chat details drawer lists each charge, from Chat & Reasoning and Tool Calls to Compute and the orchestration fee, with waived credits shown struck through as Free. The agent's Performance page adds Insights, Tasks and Evaluations tabs, with average credits per task and credits by model, trigger, member and connector. Credits per task is the number to watch while you tune.

The docs list four ways to lower that number: a cheaper model, your own provider API key to waive model credits, new tasks instead of long threads, and fewer tools. Your own key pays off on token-heavy runs; on runs that are mostly tool calls or browser and compute time it can add credits, because the orchestration fee doubles. Out-of-credits emails are on by default, and usage emails arrive at 75% and 90% of the pool, thresholds you can change. Read them early. A single busy trigger can move the pool faster than a whole team chatting.

Sharing comes last. An agent has two roles, Owner and User: Owners configure it, Users run it and see what Owners allow. Keep two Owners on any agent the team depends on, because the last Owner cannot be removed. In Slack, the standard Gumloop bot holds one agent per channel; a second agent in the same channel needs its own Custom Slack App. Every message a teammate sends to a shared agent draws on the organization's one pool; people outside the organization pay from their own account. Developers who want agents inside their own tools can use webhooks, the REST API or the SDKs, covered in our Gumloop API guide.

Where to keep learning: University, cohort, workshops

Gumloop University is the structured path. Its Getting Started course has 6 lessons, and an Admin Guide course has 8 for whoever runs the organization . Pro support pairs University with email. Gumloop's docs cover core concepts, help articles, the API, the CLI and MCP. If you already work in an MCP client such as Claude or Cursor, the Gumloop MCP server can search that documentation and return a cited answer without leaving your editor.

For live help, the Learning Cohort is a guided one-week program with live sessions, practical challenges and direct access to the Gumloop team. The workshops page lists a cohort starting in late November 2026: one week from idea to deployed automation, two live sessions and two challenges, by application. Past workshops stay on demand as recordings, typically 60 minutes, on Brain, artifacts, the browser, evals and reflections, open-source models to cut token costs, and rollout guardrails. Enterprise buyers also get a 30-day Launch program, with two outcomes agreed on day one, the first agent live by day 10 and the second by day 20.

Can a team learn Gumloop inside the 14-day trial?

It can, if the two weeks have a plan. Spend the first days on Gumball and one gallery agent, the middle days on connectors, approvals and a single trigger, and the last days reading credits per task on the Performance page. By then you know whether a 20,000-credit monthly pool fits your real work, and you can still cancel before the trial turns into paid Pro and the card is charged. Teams that only chat with Gumball for two weeks learn the least, because triggers and approvals are where Gumloop either earns its place or does not.