Gumloop

Gumloop's most meaningful current update is Gumball Beta, released with version 10.18.0 on August 6, 2026. Gumball is a personal agent created for each user rather than a shared custom agent. It can chat inside Gumloop, respond through Slack, receive requests at a fixed Gumloop email address, use the user's own connectors, and search the user's available skill library. Its headline change is not a new visual workflow node. It is a proactive layer that tries to organise the daily flow of work before the user explicitly starts a task.

Gumball currently brings together three features: Daily Chew, Smart Inbox and Meeting Prep. Daily Chew creates a scheduled report from connected applications. Smart Inbox classifies incoming Gmail messages, applies labels, moves selected mail out of the inbox and can draft replies. Meeting Prep assembles a brief before an upcoming meeting from calendar and connected business context. Together, these features move Gumloop closer to a personal work agent rather than a platform used only to build automations for other processes.

Daily Chew turns connected activity into a scheduled briefing

The Daily Chew runs on the time and days selected by the user. It can search all available connectors or a pinned list of up to 50, follow up to 4,000 characters of additional instructions, and use a selected model. Reports always appear in Gumball, while email and Slack direct-message delivery are optional. Gumloop structures the report as a short summary plus up to four sections covering catch-up, priorities, schedule and next steps. The documentation says facts are attributed to the application they came from, which is important because a cross-app summary is useful only when a user can check the underlying evidence.

This feature could reduce the time spent scanning dashboards, inboxes, calendars and project tools. It could also create a false sense of completeness. A report can only see the connectors, accounts, permissions, sync state and records available to it. Missing, delayed, duplicated or incorrectly scoped data can change the apparent priority. A practical rollout should start with a narrow connector set and one known reporting need. The owner should compare several reports against the underlying systems, record missed and incorrect items, and decide which sources are authoritative before treating the briefing as an operational view.

Smart Inbox can classify and draft, but it should begin conservatively

Smart Inbox is the highest-risk part of the beta because it can change how incoming Gmail messages appear. Users create labels with a name, definition and colour, then decide whether matching mail should remain in the inbox or move out. Gumloop applies a conservative rule when multiple labels match: one keep label is enough to keep the message in the inbox. Messages can receive at most five labels, and labels that do not meet the system's confidence threshold are dropped rather than guessed.

The main limitation is that classification happens once, when an email arrives. Editing a label definition or changing whether it belongs in the keep or move group affects future messages, not old mail. Turning a label off does not undo earlier labels, and deleting a custom definition leaves the corresponding Gmail label on messages already processed. This makes a test period essential. A team should start with a test mailbox or low-risk labels, keep uncertain messages visible, review false moves and missed priority mail, and avoid automatic sending. Draft replies should be treated as drafts because names, commitments, prices, policy statements, tone and recipients can all be wrong even when the classification looks sensible.

Meeting Prep joins context across systems

Meeting Prep can run automatically before meetings or be started manually from the Gumball home screen. The default focuses on external meetings, defined as meetings with at least one guest whose email domain differs from the user's. Solo events and internal meetings are skipped under that setting. The generated brief contains a short summary plus possible sections for attendees, prior context, open questions and talking points. Gumloop says the prior context combines relevant history from sources such as email, Slack, meeting notes and CRM activity rather than listing each application separately.

The documentation describes Meeting Prep as read-only, which limits direct side effects. The output can still be incomplete or misleading. People with similar names, stale CRM records, private conversations, missing meeting notes and inconsistent account ownership can distort a brief. Users should verify identities, important figures, recent commitments and sensitive context before using the result in a meeting. An enterprise should also decide which sources Gumball may search and whether every participant is authorised to see the retrieved information.

Gumball uses credits and inherits the wider permission model

Gumball is not a separate free assistant. Gumloop states that classifying email, drafting replies and generating reports use models and tools like other agent activity, so these features consume credits. That matters because Gumloop has also changed its commercial structure. Current Credits documentation says there is no renewable free plan. New accounts receive a one-time, card-required 14-day Pro trial, then move to paid Pro unless cancelled. Pro starts at $37 per month with 20,000 monthly credits. Unused Pro credits do not roll over, and optional overage is billed at $0.005 per credit.

Costs can vary by model tokens, tools, processing, subagents, evaluations and the work performed. A personal agent that reviews many sources every morning, classifies continuous email and prepares multiple meetings can create steady background usage. Teams should measure the first two weeks, set a lower overage ceiling than the documented $5,000 default, and compare the time saved with the actual credit and human-review cost. The feature is valuable only when the briefing and triage reduce work more than they create.

Permissions need equal attention. Gumball uses the individual's connectors and available skills. Gumloop's broader agent system supports approval modes, including ask every time, ask for writes and deletes, custom per-tool rules and complete blocks. However, Gumloop documents always allow as the default app-level mode. Smart Inbox has an intended Gmail effect, while other connector actions available to a personal agent may reach further. Users should review every connected tool, disable unused actions and require approval for sends, writes, deletes, publication, purchases, permission changes and bulk work.

The pricing and privacy details require careful reading

Gumloop's current pricing page shows Pro and custom Enterprise, and the Credits documentation explicitly says the renewable free plan has ended. Several recent Gumloop-authored comparison articles still advertise a free plan with 5,000 monthly credits. WhatAI treats the dedicated Credits documentation as the current source of truth and records the product as paid. The same documentation says the Pro trial requires a card and converts automatically, Pro credits expire each month, and paid overage can continue until the configured cap is reached.

Gumloop also advertises SOC 2 Type II attestation, GDPR alignment, encryption at rest and in transit, zero-data-retention agreements with major model providers, and HIPAA compliance with BAAs for eligible plans. These claims should not be read as a promise that no data is stored anywhere. Agents, chats, workflows, artifacts, logs, credentials, knowledge indexes and connected systems all have their own retention and access behaviour. The homepage says Gumloop never uses customer data to train AI models, while the Privacy Policy specifically gives that assurance for premium users. Because a trial is not described with the same precision, teams using sensitive information should confirm its treatment before testing.

What the release changes for buyers

Gumball makes Gumloop easier to understand as a daily work product. A user no longer has to begin by designing a complete canvas. They can connect a few applications, observe a personal agent completing real work, then convert a successful pattern into a Skill, Agent or Workflow. That can shorten the distance between experimentation and a reusable process. It can also encourage users to grant broad access before they have defined the process, source rules or approval boundary.

The sensible starting point is narrow. Enable one Gumball feature for one measurable outcome. For Daily Chew, compare reports with source systems. For Smart Inbox, measure false moves and never auto-send during the pilot. For Meeting Prep, check identity and source completeness. Keep a written owner, a limited connector list, conservative approval settings, a credit cap and a review date. Gumball Beta is a promising personal layer on top of Gumloop's agent and workflow platform, but its value will come from disciplined configuration and measured reliability, not from assuming that proactive automatically means trustworthy.

Use Cases

Related AI in Business Tools

Explore the Network