Puente OS vs Gumloop: Comparison and When to Use Each

An AI agent builder is a platform where a person describes a job and gets an agent: a language model with access to tools, like email, the CRM or a spreadsheet, that decides case by case which step to take to reach the goal. Puente OS is the platform that builds the systems that run your business: apps, workflows, databases and agents that read from your systems, cross-reference the information none of them holds in full, decide on your rules and execute the action inside them. The agent is one of those four pieces, and in a well-designed process it takes the steps that call for judgment.

I'm Cris Ugarte, CEO and co-founder of Puente OS. We work with retail, eCommerce and logistics companies in Chile, Peru and Mexico, and since AI agents became the topic of the moment, Gumloop comes up in our conversations: it is one of the platforms pushing hardest on the idea that every person at a company builds their own agents. The next question is direct: how is Puente different?

My answer is that Puente delivers the system that runs your process, up and running and with someone accountable for it, and that in that system each step gets the piece that fits it. Almost everything that runs your company has a known right answer and runs best on rules; the agent pays off in the steps where the next move depends on what turns up. Here I go through what each one does, with public sources for everything I say about Gumloop, and when each one makes sense.

What is Gumloop and what does it solve?

Gumloop is a platform for building AI agents, operated by AgentHub Inc. Its llms.txt and its home page present it as "the multiplayer AI agent builder": anyone at a company builds agents with any model and any integration, and IT controls access. Its About page sums it up in one idea: understanding a task should be the only prerequisite for automating it.

What it offers, according to its website and its documentation:

Agents. Its agents page presents them as agents your whole team can use, with control over performance, security and observability. Each agent is built from a model, instructions, connectors, skills, knowledge sources and subagents it delegates work to in parallel. Gumball, its personal agent, builds or updates an agent from what you ask for.
Triggers. Triggers run an agent on a schedule, from a webhook or when something happens in an app, like a new email in Gmail, a message in Slack or a record that changes in Salesforce. With Create With AI, you describe in natural language what to watch across one or several apps, and the AI builds the trigger.
Skills. Its skills page describes them as reusable playbooks that agents write, correct and share, with a file-level change history and permissions to decide who edits them and who only uses them.
Artifacts. From a prompt, an agent creates presentations, documents, spreadsheets, code and web apps that read and write data from your connected tools in real time. Each artifact has its own URL, versions and permissions, and interactive ones load their data with Python scripts every time they are opened.
Brain. A company knowledge base that indexes your tools so agents can search them and answer with citations.
Human control and evaluation. With Human in the Loop, each connector is set so the agent always asks for approval, asks only to write or delete, or asks tool by tool. Evaluations grade every agent interaction against the criteria you define.
Channels. Agents are used in Slack, Microsoft Teams, email and the Gumloop interface, according to its agents page, which shows WhatsApp as "Coming soon," and also on a hosted page anyone can open.
Models. It is model-agnostic: its Enterprise page names Claude, GPT, Gemini, Grok, DeepSeek, Qwen, GLM, Kimi, Llama, MiniMax and Perplexity, and its documentation adds Auto, a router that picks the model for each message, and the option to bring your own API keys.
For developers. A REST API, Python and JavaScript SDKs, a CLI and an MCP server to use Gumloop from Claude, Cursor or VS Code.
Enterprise features. According to its Enterprise documentation, SSO with SAML and OIDC, SCIM provisioning, custom roles that set which connectors, tools and models each group uses, connector policies, audit logs and adoption dashboards. Its Enterprise page adds deployment in your own VPC on AWS, Azure or Google Cloud and SOC 2 Type II certification.

Its focus is on agents. On its plans page, the workflows the platform started with are listed under the heading "Workflows (Legacy)," and its MCP server keeps tools to list and run those saved flows. Its agents page talks about orchestrating complex processes with multiple agents "inside deterministic workflows."

It speaks to marketing, sales, operations, engineering and support teams, and its use cases sit in that territory: meeting prep, lead enrichment and qualification, CRM updates from calls, SEO research, ticket triage and competitor monitoring, according to its llms.txt. Its published customer stories include Opendoor, Gusto, fal and Samsara. In March 2026 it announced its Series B, led by Benchmark.

Anyone at your company building agents connected to their tools, with IT in charge of access and control: that is Gumloop's job.

What is Puente OS and what does it solve?

Puente OS is the platform that builds the systems that run your business. Your ERP records what happened. Your BI shows you what happened. Puente makes it happen. In your stack it is the system of action: it reads from your systems, cross-references the information none of them holds in full, decides on your rules and executes the action inside them.

What it builds comes from four pieces:

Workflows with explicit, auditable rules that start on their own when data changes or at a set time.
Databases that bring together context from several systems: the order, its receipt, its shipment and its settlement in a single record.
Apps where your team reviews, decides on and approves the exceptions.
Agents for the steps that take judgment, like an ambiguous complaint or a failed delivery.

With Puente OS you describe in natural language what your operation needs and get software running: apps, workflows, databases and agents. Puente's Growth Ops team builds each process with you and maintains it, and each one reaches production in 2 to 4 weeks. Puente acts on your systems, and where the operation lives in spreadsheets, it also keeps the record.

The discipline behind it is autonomous operations design: designing your operation so the repeatable work runs on its own, on your rules, and your team decides what takes judgment. For a closer look at the layer, see what a system of action is.

Which piece fits each step of a process?

Each step falls into one of four layers: a workflow for the rule, a model for reading, an agent for judgment and a person for the exception. The useful question for your operation is which step goes in each layer, and the answer defines how much of your process comes out the same every time.

The difference lies in who decides the next step. In a workflow, the code decides, with the rules your company defined, and the same data always produces the same result. In an agent, a language model decides, case by case, and that is exactly what you want in the steps where the path depends on what turns up. Anthropic, OpenAI, Microsoft and Google agree on the advice: start with the simplest solution, use rules for what has a known answer and save agents for what has no fixed path. The detail, with sources, is in how to implement AI in your business.

Take an order that comes in through a marketplace and has to ship with its label:

A workflow for the rule. The shipping cutoff, the carrier chosen by zone, cost and insurance, the label generated and the tracking number written to the order. These decisions have a right answer, and your team and your auditor expect to see them applied the same way on every order.
A model for reading. The address written a thousand different ways, the order that arrives over WhatsApp or the supplier's PDF. A model turns it into fields, and a rule validates them before anything is written to your systems.
An agent for judgment. The failed delivery, where the customer's reply decides whether to reschedule, change the address or pick up at a collection point.
A person for the exception. The refund above the amount you set or the case that falls outside every rule, which reaches an app with the order and its history in view.

Around those steps, the process needs its database, where the order, its receipt, its shipment and its return sit in a single record, and the screen where your team decides. When those pieces live in different tools and with different people, coordinating them eats up your team's hours: that is the coordination tax. Puente OS delivers all four layers, the data and the apps on a single platform, designed around your process.

How do Puente OS and Gumloop compare, dimension by dimension?

Dimension by dimension:

Who builds. Gumloop: the people at your company, describing in natural language what they want, with templates, Gumball to build or adjust an agent and an expert directory of people you can hire. Puente OS: the Growth Ops team builds each process with you. You describe in natural language what your operation needs and get software running, and every activation or irreversible change goes through your confirmation.
Who maintains. Gumloop: whoever built the agent, with agents that review their own work and propose improvements to their skills for someone to approve. Puente OS: the Growth Ops team maintains each process as your operation changes, with a new channel, a new rule or a report that changes format, and responds when something fails.
The core piece. Gumloop: the agent. When a trigger fires, the agent receives a prompt and processes it just as if you had typed it in the chat, according to its triggers documentation. Puente OS: the four pieces, with each step on the one that fits it. What has a known answer runs in workflows with explicit rules, and the agent takes the steps that call for judgment.
Triggers. Gumloop: schedules, webhooks and app triggers like Gmail, Slack, Microsoft Teams, Google Sheets, Notion, Airtable, Zendesk, Salesforce, HubSpot, Linear, Jira and Monday.com, plus triggers the AI builds from a description and that span several apps. Puente OS: workflows that start on their own when data changes in your systems or at a set time, with rules that are written down, versioned and edited.
Apps and interfaces. Gumloop: artifacts, from a document to a web app with live data from your tools, with their URL, versions, permissions and the option to host them on your own domain; and agents that work in Slack, Teams, email or a hosted page. Puente OS: web apps tailored to each process, hosted on Puente with their own link, where your team reviews, decides on and approves the exceptions, like a discrepancy dashboard for finance or a stalled-shipments inbox for operations.
Data. Gumloop: Brain, the knowledge base that indexes your documents and conversations so agents answer with citations; files that persist across conversations with the same agent; and connectors to read and write databases like PostgreSQL, Snowflake or BigQuery. Puente OS: your business databases, with typed columns validated before saving, queries with filters, grouping and aggregation, and loads of up to 10,000 rows in a single operation. The workflows and apps of the same process read and write them.
Integrations. Gumloop: over 100 MCP integrations according to its llms.txt and over 300 natively supported integrations according to its About page, with names like Salesforce, HubSpot, Slack, Gmail, Notion, Shopify, Stripe, QuickBooks and NetSuite, and the option to connect any MCP server or deploy your own. In the connector list of its llms.txt and in its documentation, as of October 11, 2026, Mercado Libre, Bsale, Defontana, the SII (Chile's tax authority) and Falabella do not appear. Puente OS: native connectors for Mercado Libre (orders and shipping details), Bsale (stock and documents), Multivende, WhatsApp, Gmail and Google Sheets, among others. The Growth Ops team connects the rest, like the SII, your invoicing system or your ERP, through its API, over HTTP or Python, or by reading the ERP's database.
Channels with your customers. Gumloop: Slack, Microsoft Teams, email and hosted pages, with WhatsApp shown as "Coming soon" on its agents page. Puente OS: agents on WhatsApp, built on the official WhatsApp Business API with Meta-approved templates, on Instagram, on email and in a web portal's chat, working with text, images, files and voice, with handoff to a person on your team when the conversation calls for it.
Human control. Gumloop: approval per connector or per tool, with modes like asking permission only to write or delete, notifications through Slack or in the app, and the option for the agent to ask a question before moving on. Puente OS: irreversible actions, like issuing a tax document, are designed in two steps, a draft and a person's confirmation, and exceptions reach an app with full context.
Error handling and monitoring. Gumloop: a trigger turns itself off after 3 consecutive failed runs and notifies its owner by email, Slack and in the app, with a link to the latest failed run; evaluations that grade every interaction; and an activity log for connectors. Puente OS: each execution's detail shows the step where it stopped and, per step, the error, the traceback and the logs; there are failure alerts with an excerpt of the error, and failure policies per step and per iteration. Behind it is the team that responds: at an electronics seller in Mexico, every incident is recorded in a support log, and from August 1 to 12, 2026, the median resolution time was 10 minutes.
AI. Gumloop: agents on any model, a router that picks the model per message, skills that agents write and improve, subagents, Brain and evaluations. Puente OS: the complete system is built with AI. You describe in natural language what your operation needs and get apps, workflows, databases and agents running: that is how, at an electronics seller in Mexico, one person built 20 apps and 75 processes in ten months. Inside each process, a model reads what arrives unstructured and agents take the steps that call for judgment: they run inside workflows, with tools, history and an iteration cap, and a decision agent classifies and routes with closed questions.
Governance and deployment. Gumloop: SSO with SAML and OIDC, SCIM, custom roles, connector policies written in natural language, model restrictions, audit logs and deployment in your own VPC on the Enterprise plan, plus SOC 2 Type II certification; immutable versions of each agent, according to its API. Puente OS: every workflow save creates a new version, and rolling back means activating the previous one; in apps, every publish is saved as a version, with earlier ones kept as history. Each app is assigned the tables its process uses, with read, write and delete permissions. Puente manages the platform where everything runs.

What does it look like in a real operation?

It looks like processes where rules do almost all the work and AI shows up in the steps that need it, with a team that maintains them. Three Puente OS customers, from least to most judgment:

Example 1: a multichannel electronics seller in Mexico, from order to shipping label

An electronics seller in Mexico runs without an ERP, and Puente has seven of its sales channels connected. Every order follows its path on rules: the address is validated, the carrier is chosen by zone, cost and insurance, the label is generated and the tracking number is written to the order. When the address arrives written irregularly, an AI model normalizes it, and the rule validates it before moving on. Other processes run around it: a report that consolidates 213,362 Mercado Libre orders via API and a support log where every incident is recorded.

Of the 38,628 orders processed between November 2025 and August 2026, 97.9% got their shipping label with no human intervention: 37,815 labels. Generating the labels for a cutoff went from 5 to 7 hours to 15 minutes, and monthly volume grew 40% with the same operations team. From August 1 to 12, 12 of 2,051 orders had an incident, 0.59%, with a median resolution time of 10 minutes. Behind it are 20 apps, 75 processes, 7 channels and 19 APIs, built in ten months by one person.

Example 2: a pet retailer in Chile, with the margin on every sale in view

A pet retailer sells on four channels. A workflow calculates the margin of every sales line each day: as of August 2026 it had covered 247,626 lines, and 3,830 of them, 1.5%, showed a negative margin that nobody saw before. Every shipment is measured against the delivery promise using the courier's API: 33,192 shipments measured and 85.0% on-time delivery among those not canceled. It is a calculation with a right answer, so it runs on rules.

It runs 10 apps in production for six departments, with 167 versions deployed, built and maintained by one person. The time from requesting a new process to having it in production dropped from 105 days, on the first project, to 7.

Example 3: a multibrand fashion retailer in Chile, with an agent where the conversation leads

A Chilean multibrand fashion retailer uses each piece according to what the process calls for. In customer service, where every request arrives differently over WhatsApp, Instagram or email, an AI agent handles the first line, checks each order's tracking in the OMS, resolves exchanges and returns inside the chat and hands the conversation to a customer service rep when needed. In alterations, fixing a garment always follows the same path (a ticket at the store, routing to the area's workshop, six statuses and an email to the customer with each change), and that process runs on rules on the same platform.

The agent takes the conversation, where every case calls for judgment, and the rules take the path that repeats.

Why choose Puente OS?

Because you get the system that runs your business, up and running: each step of your process on the piece that fits it, the apps where your team works and your business database, connected to your local systems and built and maintained by the Growth Ops team. Gumloop is an excellent tool for the people at your company to build their agents, and Puente OS delivers the operation running, with someone accountable for it. Six concrete reasons:

You get the system working. Each process reaches production with its screen, its data, its rules and its agents, and your team works the exceptions in an app built for that process. At the pet retailer in Example 2, that adds up to 10 apps in production for six departments.
Each step runs on the piece that fits it. A workflow for the rule, a model for reading, an agent for judgment and a person for the exception. What has a known answer comes out the same every time and is audited by its rule: at the electronics seller in Example 1, 97.9% of orders go out with their label with no human intervention. Where judgment leads, an agent works, as in the customer service of the fashion retailer in Example 3.
Your process has its database and its screen. The order, its receipt, its shipment and its margin sit in a single record, with typed columns, and your team decides in apps with per-table permissions. At the pet retailer, that database surfaced 3,830 sales lines with a negative margin that nobody saw before.
It works on the systems you already use. Puente has native connectors for Mercado Libre, Bsale and Multivende, and the team connects your invoicing system, your courier or your ERP through its API, over HTTP or Python, or by reading the ERP's database. At the electronics seller, a report consolidates 213,362 Mercado Libre orders via API.
A team builds it with you and responds when something fails. Growth Ops maps what triggers each process, what it decides and what its exceptions are, builds it and maintains it. At the electronics seller, from August 1 to 12, 2026, 0.59% of orders had an incident, with a median resolution time of 10 minutes.
Each process builds on the last. Each one reaches production in 2 to 4 weeks and uses the connections and data the previous ones left in place. At the pet retailer, the time from requesting a new process to having it in production dropped from 105 days to 7.

That way your operation runs on your rules, and your team spends its time deciding what takes judgment.

When does each one make sense?

Gumloop fits when you want the people on your teams to build their own agents on the tools they already use. Puente OS fits when you want your operation running end to end, with each step on its piece and a team that builds it with you and maintains it.

Gumloop, if your sales, marketing, support or operations teams want to build agents for meeting prep, lead qualification, CRM updates, research or reporting, and work with them in Slack, Microsoft Teams or email. Also if your stack lives in tools like Salesforce, HubSpot, Gong, Notion or Google Workspace, if you want to choose the model for each agent or if your IT team requires SSO, SCIM, auditing and deployment in your own VPC. Its breadth of connectors, its product pace and its focus on letting anyone build are real strengths.
Puente OS, if you want the process left running with its apps, its data and its rules, connected to Mercado Libre, Bsale, your invoicing system or your ERP, with a team that designs it with you, maintains it and responds when something fails. Also if your operation needs what has a right answer to come out the same every time, with the rule in view, and the agent to work where judgment leads, for example on WhatsApp with your customers.

How do you start?

With a process that hurts, with volume and manual handoffs: generating shipping labels, reconciling a marketplace or matching documents. With Puente OS, each process reaches production in 2 to 4 weeks.

Pick the process. With volume, manual handoffs and a visible cost when it fails.
Take the baseline. Cases per month, hours of manual work per week and the share of cases that come out right the first time.
Take inventory of what already runs. The agents, flows, spreadsheets and scripts that touch that process, with what they do, who maintains them and which systems they use. That list defines what gets connected and what gets redesigned.
Separate the rule from the judgment. Mark which steps have a known answer, which arrive unstructured and need a reading, and which depend on what turns up along the way.
Write down the rules your team applies from memory. For example: "if an order still has no label two hours after the cutoff, it goes to operations with the full order in view."
Decide where each exception goes. The system resolves it, or it reaches a named person with full context in an App.
Measure from the first week and add the next process on the same platform.

That is autonomous operations design applied process by process: the repeatable work runs on its own, on your rules, and your team decides what takes judgment. If you are comparing automation platforms, the same approach is in Puente OS vs n8n; if your focus is agents that talk to your customers, in Puente OS vs Vambe, and if you sell on marketplaces, in Puente OS vs Multivende.

We believe every company in Latin America can be operationally autonomous, and that starts with having every process running end to end, with each step on its piece and a team that keeps it running.

Key takeaways

Gumloop is an AI agent builder for business teams: agents on any model, triggers in natural language, skills, artifacts with live data, Brain, human approval and controls for IT, like SSO, auditing and deployment in your VPC.
Each step of a process falls into one of four layers: a workflow for the rule, a model for reading, an agent for judgment and a person for the exception.
Puente OS delivers the complete system up and running: apps, databases, workflows and agents on a single platform, designed around your processes and connected to your systems.
Puente has native connectors for Mercado Libre, Bsale and Multivende, and the Growth Ops team connects the rest of your systems via API, such as the SII, your invoicing system or your ERP.
The Growth Ops team builds each process, maintains it and responds when something fails: at an electronics seller in Mexico, 97.9% of orders go out with their label with no human intervention, and from August 1 to 12, 2026, incidents were resolved with a median time of 10 minutes.
Gumloop fits when your teams want to build their own agents on their tools. Puente OS fits when you want your operation running end to end, in 2 to 4 weeks per process.

Want to see which steps of your operation run on rules and which on agents? Let's talk about your processes, and we'll show you Puente OS running with cases like the ones in this article. Book a conversation with our team.

Frequently asked questions

What is Gumloop and what is it used for?

Gumloop is an AI agent builder, operated by AgentHub Inc., that presents itself as "the multiplayer AI agent builder." People at a company use it to build agents that connect to their apps, run on triggers, use skills and produce artifacts like dashboards, documents and web apps, on any AI model. Its typical use cases are meeting prep, lead enrichment and qualification, CRM updates, SEO research, support ticket triage and reporting. For IT, it offers SSO, auditing, usage monitoring and deployment in your own VPC.

Does Gumloop have workflows?

It has workflows, and today its product is organized around agents. On its plans page, workflows are listed under the heading "Workflows (Legacy)," and its MCP server keeps tools to list saved flows, run them and review their results. Its agents page describes orchestrating processes with multiple agents inside deterministic workflows, and its triggers run an agent that processes the event as a chat message. In Puente OS, the workflow with explicit rules is one of the four pieces, and the agent runs inside it in the steps that call for judgment.

Does Gumloop integrate with Mercado Libre, Bsale or the SII?

In the connector list of its llms.txt and in its documentation, as of October 11, 2026, Mercado Libre, Bsale, Defontana, the SII and Falabella do not appear, and the connector catalog URLs for those names return 404. Shopify, Stripe, PayPal, QuickBooks and NetSuite do appear, among others. For a service without its own connector, Gumloop lets you connect any MCP server or deploy your own. Puente OS has native connectors for Mercado Libre (orders and shipping details), Bsale (stock and documents) and Multivende, and the Growth Ops team connects the rest via API.

Does Gumloop work for automating operations processes?

Gumloop has a line for operations teams and use cases like processing invoices that arrive by email, where an agent reads the PDF, extracts the data, checks the math and logs it to Google Sheets, according to its llms.txt. Each agent is built and adjusted by your team, or by an expert from its directory. Puente OS is made for the company that wants the operational process running end to end: the Growth Ops team designs it with you, puts each step on its piece, connects it to your ERP, your marketplace, your courier and your invoicing system, and maintains it.

Does Puente OS have AI agents?

Yes, and Puente OS's AI starts earlier: it's in how the system gets built. You describe in natural language what your operation needs and get software running; at an electronics seller in Mexico, one person built 20 apps and 75 processes in ten months. Inside that system, the agent is one of the four pieces, along with apps, workflows and databases, and it runs inside a workflow: a simple agent to draft or summarize, an advanced agent with tools, history and an iteration cap, and a decision agent that classifies and routes with closed questions. Puente also builds agents that talk to your customers over WhatsApp, Instagram, email or a web portal's chat, and that work with text, images, files and voice. Each step gets what fits it, as I explain in how to implement AI in your business.

What happens to the agents I already have in Gumloop?

They go into the inventory first. The Growth Ops team reviews with you the agents and flows that touch each process, what they do and who maintains them, and decides which ones keep running where they are and which get redesigned within the process in Puente OS, with its app, its data and its rules. Gumloop offers a REST API and webhooks to run its agents from another system. Each process reaches production in 2 to 4 weeks.

External sources cited

Everything this post says about Gumloop comes from its website, its llms.txt, its documentation and its official forum, reviewed on October 11, 2026. Puente OS customer figures are production measurements as of August 2026; label generation time and the request dates for each process come from follow-up with the customer.

Gumloop: product

llms.txt (Gumloop): "Gumloop is the multiplayer AI agent builder. Anyone at a company can build agents with any AI model and any integration, while IT controls access"; operated by AgentHub Inc.; agents that connect to apps, run on triggers, use skills and produce artifacts like dashboards, documents and web apps; model-agnostic; enterprise controls: SSO, audit logging, usage monitoring and VPC deployments; "100+ MCP integrations"; typical use cases (lead enrichment and qualification, CRM updates, meeting prep, SEO, support triage, competitor monitoring, data analysis, scheduled reports, invoice processing); MCP connector list; REST API, MCP server, Python and JavaScript SDKs and CLI; solutions for Marketing, Sales, Operations, Engineering, Support and Security.
Home page (Gumloop): "Gumloop is the multiplayer AI agent builder."
About (Gumloop): "over 300 integrations natively supported," connection to any external MCP or deployment of your own MCPs; six beliefs, among them "Everyone should be able to automate": understanding a task should be the only prerequisite for automating it; deployment in your own VPC.
Agents (Gumloop): agents for the whole team, with control over performance, security and observability; "Orchestrate complex processes with multiple agents inside deterministic workflows"; agents in Slack, Microsoft Teams and Gmail, with WhatsApp marked "Coming soon."
Triggers (Gumloop): describe in natural language what to watch on the web or across several apps, cross-reference several tools in a single check and detect conditions like percentage drops or mismatches between systems.
Skills (Gumloop): playbooks agents write, correct and improve; automated performance reviews with suggestions to approve; reuse across the organization with permissions; change history with file-level diffs.
Artifacts (Gumloop): web apps, presentations, code, documents, spreadsheets and images from a prompt; apps that read and write data from connected tools in real time; versions, a dedicated URL and permissions.
Experts (Gumloop): "Gumloop Expert Directory," with vetted experts to help automate workflows with AI.
Customers (Gumloop): stories from Opendoor, Gusto, fal and Samsara.
Pricing (Gumloop): the plans table groups the workflow features under the heading "Workflows (Legacy)." Cited only for that heading.
Enterprise (Gumloop): Claude, GPT, Gemini, Grok, DeepSeek, Qwen, GLM, Kimi, Llama, MiniMax and Perplexity models; bring your own keys; audit logging; "VPC deployments: Deploy Gumloop inside your own AWS, Azure, or Google Cloud environment"; roles and permissions; SAML SSO and SCIM; AI model restrictions; SOC 2 Type II; GDPR and zero data retention agreements with third-party models; plain-English policies that block, tag or log actions; MCP client and server tracking.

Gumloop: documentation

Agents (Gumloop documentation): an agent is made of a model, instructions, connectors, skills, knowledge sources, subagents, secrets and abilities; subagents run as their own tasks, up to 10 concurrently per task.
Agent triggers (Gumloop documentation): scheduled, app and webhook triggers; when a trigger fires, the agent receives the configured prompt and processes it "exactly as if you had typed it in the chat"; triggers for Gmail, Slack, Microsoft Teams, Google Drive, Google Sheets, Google Calendar, Notion, Airtable, Zendesk, Salesforce, Linear, Jira, Monday.com and Parallel Web Monitor, plus Outlook, Google Forms, Typeform, HubSpot and incident.io; triggers deactivate after 3 consecutive failed runs and notify their owner by email, Slack DM and in-app.
Create triggers with AI (Gumloop documentation): the AI builds triggers that monitor any combination of connected services and fire the agent when conditions are met.
Agent artifacts (Gumloop documentation): automatic versioning, access permissions, hosting on a custom domain; interactive artifacts are HTML with Python scripts that load live integration data each time they are opened.
Brain (Gumloop documentation): company knowledge base that indexes connected tools so agents answer with citations.
Human in the Loop (Gumloop documentation): per-connector approval modes (Always allow, Ask each time, Ask for writes/deletes, Never allow and Custom), notifications in-app, by Slack DM or in the chat, and the Ask Question ability.
Evaluations (Gumloop documentation): grades every agent interaction against the defined criteria and tags conversations for reporting.
Hosted Pages (Gumloop documentation): a standalone page for the agent that anyone can open.
AI models (Gumloop documentation): models from Anthropic, OpenAI, Google, DeepSeek, MiniMax and Z.ai, among others; Auto, the router that picks the model per message; bring your own key.
Custom MCP servers (Gumloop documentation): connection to any MCP server.
Enterprise features (Gumloop documentation): SSO with SAML, OIDC and SCIM; custom roles controlling connectors, tools and models; connector policies; managed tunnels; static egress IPs; audit logging; Organization Insights; usage data export.
Getting started, API (Gumloop documentation) and List agent versions: REST API to create agents, start sessions and trigger automations; immutable versions of each agent.
MCP server (Gumloop documentation) and Flow and run tools: MCP server for Claude, Cursor, VS Code and other clients; tools to list workbooks and saved flows, run them and view their history.
llms-full.txt (Gumloop documentation): full text of the documentation. Search on October 11, 2026: "Mercado Libre," "MercadoLibre," "Bsale," "Defontana," "Falabella," "SII" and "Servicio de Impuestos Internos" return 0 results; they also do not appear in the connector list of the site's llms.txt, and the gumloop.com/connectors/ URLs for mercadolibre, bsale, defontana, falabella and sii return 404.

Gumloop: company

Announcing our Series B (Gumloop official forum, Mar 12, 2026): Series B announcement, led by Benchmark.

Puente OS vs Gumloop: From the Agent to the System That Runs Your Business