Build, share, optimize & control agents

Book a demo

The agent infrastructure powering the world's most AI-native companies

Tasks Automated
0+
Agents Deployed
0+
Churn Prevention Agent

Watches every account for churn signals and runs the save play before renewal is at risk.

Claude 5.5 Opus
Stripe
HubSpot
PostHog
+5
JJoseph

Who is at risk of churning before the end of the quarter, and what should we do about each one?

Churn Prevention Agent
HubSpot
Pulling the 38 accounts renewing in the next 90 days
PostHog
Comparing 90-day usage against each account baseline
Intercom
Scanning open tickets for frustration signals
Gong
Listening for hesitation in the last QBR calls

3 accounts are at real risk, $165K ARR in total, all renewing within 60 days. The other 35 look healthy.

Northwind ($50K): Usage fell 62% after their admin left in August and nobody rebuilt her dashboards. Play: a 30-minute rebuild session this week. Note drafted from you.

Globex ($84K): Your champion moved roles and the replacement has never logged in. Play: an exec intro from Max plus an onboarding call. Task logged in HubSpot.

Initech ($31K): 4 open tickets about export speed, 2 with frustrated language. Play: an engineering update and a 10% loyalty credit. Needs your approval before I send.

Worked for 4 minutes · 38 accounts reviewed · 1 approval pending
Build

Let your experts
build the agents

Understanding a task is the only prerequisite to automating it. Let your team who already understand the problem build the necessary agents. No learning curve involved.

App triggersRecurring tasks400+ ConnectorsSkillsArtifact buildingImage generationSelf-improvement

A real browser for every agent

Agents sign in with your saved logins, browse from trusted connections, hand off to you when a site needs it, and record every session.

Anything you can do on the web

Agents open a real browser and click, type, upload, and pull data from any site your team already uses, with or without an API.

Trusted connections with CAPTCHA handling

Agents browse from trusted connections in the region you choose and clear CAPTCHAs on their own.

A password vault built for agents

Save a login once or connect 1Password, and agents sign in without ever seeing the credential.

Human in the loop with secure handoffs

When a site needs you, agents hand you the live browser to sign in or approve, then pick up where you left off.

Session recordings with audit logs

Every session is recorded step by step, with passwords masked and every login, takeover, and handoff in your audit log.

Collaborate

Meet your team
where they work

Search Gumloop

#data-analysis

MessagesAdd canvasFiles
M
Max3:12 PM

@Databot Compare this month's sales to the same month last year

💬2🔥1
G
Gabriela4:42 PM

@Databot Who are the top 3 most active users at Mostco?

💬1👀1
M
Marcelo8:50 PM

@Databot Where are we losing people in the onboarding flow?

💬1🎯1
Slack@mention an agent in any channel or thread
Search
Y
📊
data-analysis
GTM · General
Meet
M
Max3:12 PM

@Databot Compare this month's sales to the same month last year

💬2🔥1
G
Gabriela4:42 PM

@Databot Who are the top 3 most active users at Mostco?

💬1👀1
M
Marcelo8:50 PM

@Databot Where are we losing people in the onboarding flow?

💬1🎯1
Type a message
Microsoft TeamsBring agents into your chats and channels
Gmail
Search mail
Y

Sourcing video producers (SF, B2B)

Inbox
Gumloop<agent@gumloop.com>2:38 PM
to me

Done! Drafts for Tess and Maya are in your inbox, personalized to each.

GmailLet agents draft and triage right in your inbox

Built by one, used by all One teammate builds an agent and the whole org puts it to work, each in their own chats.

  • Full control over the agent.

    • Edit anything, including every trigger
    • Decide who has access
    • Transfer or delete the agent

Decide who can do what Let some people edit, some view, and some just use the agent, for one person, a team, or the whole company.

Smarter with every run Each agent reviews its own runs, spots what slowed down or cost too much, and fixes it for next time.

Always working, even while you’re away A meeting ends, a lead fills out a form, an issue gets filed. Agents pick it up the moment it happens, before anyone asks.

Held to your standard Every run is scored against the rules you set. When an agent quotes pricing it shouldn’t or guesses instead of escalating, you hear about it first.

Slack
Gmail
GitHub
Salesforce
Jira
HubSpot
Notion
Google Drive
Google Sheets
Linear

Knowledge that compounds Every tool you connect becomes part of a centralized, always up-to-date brain agents and humans can use.

Insights

Your agents. Your spend. One control plane.

See where credits go, what agents are asked to do, and which skills get used, across every team.

Spend per model

Which models cost the most to run?

  • GPT-6.1 Sol: 298K credits
  • Claude Opus 5.5: 94K credits
  • Grok 4.7: 41K credits
  • Kimi K3: 26K credits
  • GLM-5.3: 22K credits
  • DeepSeek V4.1 Flash: 19K credits
  • Gemini 3.8 Flash: 15K credits
  • Qwen3.8 Max: 11K credits

Spend by agent

Which agents spend the most credits?

412K+12%

Agent health

Which agents need attention right now?

85%109 of 128 agents
  • Healthy109
  • At risk14
  • Failing5

What agents are asked to do

What do people ask agents to do?

91%+3%

Where work happens

How does work reach your agents?

4.1K+18%

What skills get used

Which skills get used?

38never used

Adoption

Who is building with agents?

525+18%
Bella headshotMarcelo headshotAndreina headshotAlexandra headshotRahul headshot
Controls

Enterprise-grade controls

Usage monitoring

Track organization-wide credit usage in real time. Implement budget and quota controls to avoid surprises.

Audit logging

Capture detailed audit trails for actions across the organization to understand where data is flowing.

VPC deployments

Deploy Gumloop inside your own AWS, Azure, or Google Cloud environment to keep data in your network.

Roles and permissions. Manage reusable roles, credentials, and secrets with scoped access.

SAML SSO and SCIM. Securely streamline identity and access management.

AI model restrictions. Control which AI models teams can use.

SOC 2 Type II. Independently audited and compliant.

GDPR. Zero Data Retention agreements for third-party models.

Credential management. Store secrets in one place and scope who can use them.

App policies and guardrails. Write plain English rules that block, tag, or log actions before they run.

MCP client and server tracking. Trace every tool call through one logging and analytics layer.

Reporting and analytics. See AI adoption, usage, and outcomes for every team in one place.

Spend caps and approvals. Cap spend by agent, team, or org, and approve costly actions.

Model Router

Frontier quality, without the frontier bill

Gumloop's Model Router matches every task to the model it needs, so routine work runs cheap and hard work gets the best intelligence available.

In agents, they trust

Gusto powers AI agents across its go-to-market engine with Gumloop, driving revenue, reducing churn, and elevating rep performance.

$1.5M+
Additional ARR in 3 months
“Gumloop has been critical in helping all teams at Instacart — including those without technical skills — adopt AI and automate their workflows, which has greatly improved our operational efficiency.”
Fidji Simo
CEO, Instacart
“Gumloop turned our sales and marketing teams into builders, and saved us thousands of hours of manual work.”
Ryan Schwartz
VP, Marketing Systems & Intelligence, Samsara
“The partnership review agent cuts prep time from 30–45 minutes per customer to under five minutes, and it results in a 31% higher win rate and 44% larger deals.”
Matt Gould
Head of Insights and Operations, Gusto
See what’s new

Recently shipped

Chat Widgets

Agents can now show a small interactive visual right inside their reply, like a chart, a table, or a quick mock to iterate on before building the full artifact.

Oct 9, 2026

58 New Salesforce Tools

Agents can now build out your Salesforce org, not just read it. Create objects and fields, run Apex, deploy metadata, assign permission sets, work approvals, and write records in bulk.

Oct 9, 2026

Ask Gumball to Build an Agent

Tell Gumball what you want an agent to do and it builds or updates the agent, with its instructions, connectors, skills, knowledge, and triggers.

Oct 7, 2026

Claude Haiku 5.5

Anthropic's fastest Claude is now available for agents, with a 1M context and about a twentieth of Sonnet 5.5's price on prompts under 100k tokens.

Oct 7, 2026

Gumloop's Agent Browser

Agents can now open a real browser and click, type, sign in, fill forms, and download files on any site, with or without a connector.

Oct 6, 2026

Share an Agent Trigger With Its Users

Share a trigger you built and each person on the agent can turn on their own copy, running through their own account.

Oct 6, 2026

16 New Connectors

Agents now work with Amplitude, Bright Data, Buffer, Coda, Contentful, GoHighLevel, Higgsfield, Instantly, Navan, Neon, Responsive, Statsig, Tally, Vitally, Yotpo, and ZoomInfo.

Oct 1, 2026

Improved Agent Chat Performance

Agent chats are smoother and much lighter on your laptop or phone.

Oct 1, 2026

Text Your Gumball (Beta)

Text Gumball over iMessage, RCS, or SMS to kick off a task from anywhere.

Sep 29, 2026

Password-Protected Artifacts

You can now set a password on an artifact to securely share it.

Sep 24, 2026

Faster, Cheaper Agent Tasks

Agent tasks start sooner and long ones cost fewer credits.

Sep 24, 2026

Talk to Your Agents

Start a call with any agent or Gumball and just talk. Ask it to prep you for your next meeting, kick off a task while your hands are busy, or think through a problem out loud. Turn it on for your own agents in their settings.

Sep 23, 2026

Revamped Agent Builder

The agent page is reorganized so instructions, connectors, skills, triggers, and access are easier to find and edit in place. A new agent starts from one question, what should it take care of, then connects the right apps, writes its own instructions, and offers a trigger. Reflections, Evaluations, and Insights live together on a Performance page with spend by model, member, connector, skill, and subagent.

Sep 18, 2026

Agent Owners and Users

Agents now have two roles. Owners configure the agent and decide what Users can see and do, such as viewing team tasks, reading the instructions and connectors, or creating triggers. Users get a focused chat page and can request Owner access.

Sep 18, 2026

Browser Notifications for Agent Chats

Switch tabs during a long run and your browser tells you when the agent finishes or needs you for an approval, a question, or a connection. The tab icon shows whether the agent is working, blocked, or has something you have not seen. Turn it on from Settings.

Sep 17, 2026

Model Router in the API, SDK, and CLI

Ask the Model Router which model to use for a prompt from your own code. Send the prompt and the models you have, and get back the pick and why.

Sep 17, 2026

Organization Webhooks

Organization admins can register webhook endpoints under Settings, Data Export, Webhooks and get a signed request whenever someone joins the organization. Send a test event, watch the delivery log, and pause or resume an endpoint from the same tab.

Sep 17, 2026

Introducing Gumloop's Model Router

A new Auto mode for all Gumloop agents. It picks the right model for the task, so you don't have to.

Sep 14, 2026

GPT-Image 2.5

OpenAI's newest image model is now the default for agent image generation. Agents can use your uploads or earlier images as references, and they pick the size, quality, and format that fit each request.

Sep 14, 2026

Set Defaults for New Agents

Organization admins can now set the starting model, reasoning, compaction, and access settings for every new agent from the Agents tab in organization settings.

Sep 14, 2026

Introducing Gumball

Your proactive, model agnostic, self improving agent. All in your company's cloud.

Sep 9, 2026

Major Speed Improvements

Chatting with agents is much faster, from sending a message to seeing the reply.

Sep 9, 2026

Agent Emails Are Now @gumloop.ai

Agent inboxes and Gumball now use shorter @gumloop.ai addresses. Mail sent to the old @gumloopagents.com addresses still arrives.

Sep 9, 2026

GPT-6 Astra

OpenAI's most capable model is now available. It's a big step up on long, multi-step work like coding, research, and drafting documents, and it asks instead of guessing when your instructions are ambiguous.

Sep 8, 2026

Agents Can Manage Their Own Evaluations

Turn on the Manage Evaluations ability and you can ask an agent to edit its own evaluation criteria or re-run evaluations on past chats. It checks with you before changing anything.

Sep 8, 2026

Build your team of agents

Create AI agents that understand your business, work across your tools, and take work from idea to outcome.

Explore agents