Gumloop Buyers Checklist: What to Check Before Paying for Pro

Buy Gumloop only after you can answer four questions about it. Pro costs $37 a month for the whole organization's shared pool of 20,000 credits, with unlimited agents, seats and teams ^[1]. What that pool buys is the open question, because in Gumloop's own example the same agent spends 2 credits on a quick question and 200 on a deep research task. The other checks are the card trial that turns into paid Pro, the controls only Enterprise unlocks, and the Terms that let Gumloop train its own models on Pro content. Clear all four first.

Gumloop fits a team that wants agents working across Slack, email and its business apps, and that is comfortable reading a metered bill. It fits poorly when single sign-on, audit logs or per-member credit limits are hard requirements on a small budget, since all three sit on Enterprise, which is custom priced. The full breakdown of plans and credit lines lives in our Gumloop pricing guide; this list covers the decisions a buyer has to make before the trial converts.

Can you put a ceiling on what one agent run costs?

Gumloop bills every agent chat in credits worth $0.005 each, and the total is built from four lines: model tokens at the provider's list price, at least 1 credit per successful connector call, 5 credits per session-minute of active compute, and an 8% orchestration fee on the sum. Tools that fetch outside data, such as enrichment or search services, add their own charge on top of that 1 credit. There are no fixed per-message tiers. Cost grows with the model, the length of the conversation and the number of tools the agent touches.

Gumloop's docs give one worked run: 425 credits of work plus a 34-credit fee comes to 459 credits, which is about $2.30 by our arithmetic. Bringing your own model key changes the shape of that bill. Token credits drop to zero, but the fee doubles to 16% and is charged on the full list cost the run would have had, so the same run costs 93 credits, about $0.47 by our arithmetic. That saving holds on token-heavy runs; on a run that is mostly tool calls or compute time, the doubled fee can make the key add credits rather than save them. A key covers only the models its provider serves. Tool calls and compute stay billed.

• Run your three most common tasks during the trial and read each cost line in the chat details drawer.
• Pick a model per agent. Auto bills the tokens of the model that actually ran, and the routing decision itself is free.
• Count your AI-created triggers. Each one polls every 5 minutes by default (the shortest interval; you can slow it to once a week) and bills every poll, so a trigger left at 5 minutes spends at least 288 credits a day, about 8,640 in 30 days by our arithmetic, before any decision-model calls.
• Check that your key provider serves the models you plan to use.

What stops when the 20,000 credits are gone

The 20,000 credits reset each billing period and do not roll over; only Enterprise carries unused credits forward. Overage is off by default. When the pool empties, agents stop working, trigger runs fail until you top up or raise the cap, and AI-created triggers pause their polling until credits return. Brain searches stop as well, though a running agent carries on without them. Evaluations are skipped silently. For a team that runs scheduled reports overnight, an empty pool means the work simply does not happen.

Pay as you go is switched on from the Usage and Limits page and billed at $0.005 per credit, which is $5 per 1,000 credits by our arithmetic. On Pro it is always capped: the default ceiling is 1,000,000 overage credits a billing period, worth $5,000, and you can lower it ^[2]. Agents stop at the cap too. Enterprise sets its own cap or runs uncapped. An out-of-credits email is on by default, with usage emails at 75% and 90% that you can change.

• Decide before launch whether a stopped agent or an overage bill hurts more, and set the toggle to match.
• Lower the default cap to a figure your finance lead has approved.
• Name who receives the 75% and 90% emails.

The trial, the renewal and the refund rule

Gumloop has no free plan. Every new account starts with a 14-day Pro trial that needs a card and becomes a paid Pro subscription unless you cancel first, and the trial is one-time per customer. Canceling during the trial stops the card from being charged. You cancel through Manage Subscription on the Usage and Limits page, which opens the Stripe portal, and in an organization only the Admin role can open it. Gumloop's pricing page prints only a monthly Pro price; it shows no annual plan.

Subscriptions renew automatically until canceled, and canceling is never immediate: the plan and its credits stay usable to the end of the billing period or trial. After that, plan access ends and the credit balance stops resetting, while agents, skills, Brain sources and chats stay in the account for a later restart. Under Gumloop's Terms, paid subscription fees are non-refundable except where the law requires otherwise ^[3]. One more trap: a member who pays for the organization's subscription cannot leave or be removed until it is canceled.

• Put the trial end date in a shared calendar on day one.
• Check that the person whose card pays is an Admin who will stay.

Governance on Pro versus Enterprise

One Pro subscription covers unlimited seats and teams, and everyone draws on the same pool, with no separate billing per team or per person. That is the control gap to check first. Per-member monthly credit limits and per-chat credit warnings are set through custom roles, and custom roles are Enterprise. On Pro, admins and managers can watch usage across all members on the Usage and Limits page, but one heavy member can still drain the pool for everyone.

Pro still has real controls. Each connector on an agent can be set to Always allow, Ask each time, Ask for writes or deletes, Never allow, or custom per tool, and every tool is classed as read-only or write when it is added. Always allow is the default. Set approvals before an agent touches a CRM or a mailbox. Pro admins can also register an external MCP server once for the organization, with logging and tool-level access control, a setup our Gumloop MCP guide walks through.

If you are evaluating from an older demo, note the shift. 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. On Pro the pricing page lists 5 concurrent Legacy workflow runs and 25 concurrent agent chats. Chats over that limit are rejected and triggers are skipped, while Enterprise gets custom limits and queues the overflow. Size the limit against your busiest hour, not your average day.

Where your data goes and who can train on it

Gumloop processes data in the United States, and users elsewhere have theirs transferred there. Enterprise can instead deploy Gumloop into its own cloud on AWS, Azure or Google Cloud, which Gumloop deploys and operates. The bigger check is training. On Pro and Pro trials, Gumloop may use your inputs and outputs to develop, train and improve its own AI models . On Enterprise it does not, unless the customer asks for it in writing. OpenAI and Anthropic are contractually committed not to train on data sent through Gumloop.

Gumloop says it is SOC 2 Type II attested and GDPR compliant, and that it offers HIPAA business associate agreements for eligible plans; its security page does not name those plans. It also says it holds zero data retention agreements with major model providers, and models badged as hosted by a US provider run under them. The exception is Claude Fable models, which Anthropic keeps for 30 days for misuse checks; they stay off until an org admin allows non-zero-retention models. Custom data retention rules are an Enterprise feature.

• List the data you plan to connect and decide whether Pro's training terms are acceptable for it.
• Keep non-zero-retention models switched off unless your policy accepts 30 days.

Channels and defaults to test before rollout

If your company lives in Microsoft Teams, test there first. Teams works only with Microsoft 365 work or school accounts and only in public standard channels, with an @mention in every message, one agent per channel, no file uploads and every agent answering as @Gumloop. Slack goes further: a Custom Slack App adds direct messages to custom agents, several agents in one channel and your own branding, and setting it up needs a Slack workspace admin.

Outside audiences need a closer look. An agent's hosted page asks every visitor to sign in with a Gumloop account, and embedding it by iframe is not officially supported; the API is the route for putting an agent on your own site. An agent's email inbox answers only senders whose address matches a Gumloop account with permission to use it. In practice these channels serve internal teams, not anonymous customers, so plan a customer-facing front end separately.

One default deserves a deliberate choice. New agents on Pro and above start with Reflections on, where the agent reviews its recent tasks and proposes fixes, and Auto-Apply Suggestions is the default for new agents. Auto-apply is not limited to low-risk changes: it can change triggers immediately and remove skills. Switch to Review Queue on agents that touch customers or money. Our Gumloop AI playbook covers the rest of agent setup, from skills to triggers.

• Pro support is Gumloop University and email; dedicated Slack support comes with Enterprise.
• The Gumloop CLI runs on macOS and Linux; on Windows, use WSL or the Python SDK.
• Keep two Owners on important agents. The last Owner cannot be removed.

Who should start the Gumloop trial this month

Start the trial when your team wants agents across Slack, email and its apps, can live with one shared credit pool, and accepts that Pro content may train Gumloop's own models. Run real tasks through it, set approvals and a lowered overage cap before it converts, and cancel inside the 14 days if the cost lines surprise you. Go straight to an Enterprise conversation if SSO, audit logs, per-member credit limits or a no-training commitment are requirements. None of them is on Pro.