# Gumloop

> Gumloop is an AI agent platform where non-engineers build agents and automated workflows on a visual canvas, connect them to roughly 300 business apps and MCP servers, and give them access to a shared company knowledge base. Agents can be chatted with in Slack, Microsoft Teams, or email, triggered on a schedule or by an app event, and are billed by credits consumed rather than by seat.

- Category: GTM Engineering (https://saastracker.org/categories/gtm-engineering)
- Website: https://www.gumloop.com
- Starting price: $37 per month (Pro, 20,000 credits included)
- Free plan: No
- Free trial: 14 days on Pro
- Founded: 2023, HQ: San Francisco, California (founded in Vancouver, Canada), Ownership: Independent, venture-backed (legal entity AgentHub Inc.)
- Profile last reviewed: 2026-08-23
- Canonical profile: https://saastracker.org/products/gumloop

## Overview

Gumloop started in 2023 as a drag-and-drop automation canvas: you dropped nodes onto a board, wired them together, and got a repeatable pipeline that scraped a page, ran it through a model, and wrote the result into a spreadsheet. That canvas is still there, but the product's center of gravity has moved. The primary object today is an agent: a named assistant with a system prompt, a set of connected tools, a knowledge base, and reusable skill packs, which anyone on the team can talk to in Slack or leave running on a trigger.

The company's phrase for this is built by one, used by all. One person who understands a process assembles the agent, decides which tools it may touch, and writes the skills that encode how the work should be done; everyone else just @mentions it. That model is what separates Gumloop from general-purpose automation tools, where each workflow is a static pipeline someone has to maintain. It is also why the pricing has no seat charge at all: unlimited seats on every plan, with cost driven entirely by credits consumed.

For go-to-market teams specifically, the platform is most often used for the connective work that sits between a CRM and everything else: enriching inbound leads, researching accounts before a call, summarizing Gong recordings into CRM fields, drafting proposals, rolling up forecasts, triaging a shared inbox. The library of published use cases leans heavily in that direction, and the connector list covers Salesforce, HubSpot, Attio, Apollo, Gong, and the ad platforms.

The important buying caveat as of August 2026 is that the free plan is gone. The public pricing page lists exactly two options: Pro at $37 per month with a 14-day trial, and Enterprise at custom pricing. Credits do not roll over on Pro, an 8 percent orchestration fee is added on top of model and tool costs, and concurrency on Pro is capped at 5 workflow runs and 25 agent chats with rejection rather than queuing when you exceed it. None of that makes it enterprise-only, a solo operator can buy it with a card, but it does mean the real monthly cost is a usage question, not a subscription question.

## How it works

1. You create an agent, write its instructions, and connect the apps it is allowed to use. Connectors are OAuth-based and shareable across a team, so one admin can authorize Salesforce or Google Drive once and let agents built by others borrow the credential under a permission policy rather than pasting keys into individual builds.

2. Knowledge comes from Brain, a company knowledge base that indexes connected sources such as Google Drive, Notion, SharePoint, and Gong call transcripts. Drive sources can preserve original document permissions, so an agent only returns files the person chatting with it is already allowed to see, which is the detail that makes a shared agent safe to put in a public Slack channel.

3. Repeatable know-how lives in Skills, reusable instruction packs that teach an agent how a particular job is done at your company. Skills can be shared organization-wide and, on Enterprise, kept in sync with a GitHub repository so the prompt layer is version controlled like code. Reflections let an agent periodically review its own transcripts and propose changes to its own skills.

4. Workflows remain the deterministic half of the platform: a node canvas with inputs, routers, filters, loop mode for running the same branch over a list, subflows for reuse, and checkpoint history for rollback. Agents can call a workflow as a tool, which is the usual pattern when part of a process must run exactly the same way every time and the rest benefits from judgment.

5. Agents reach people where they already are. They can be @mentioned in Slack channels or DMs, added to Microsoft Teams, given their own email address, or published on a hosted page with a custom domain. Triggers start them without a human: schedules, webhooks, and app events such as a new Outlook email, a Monday.com item, or a Slack message, plus AI-built polling triggers that watch a combination of apps for a condition.

6. Everything is metered. The run log reports execution time and credit consumption per node, tool approvals and app rules let you require a human confirmation before a write action, and built-in evaluations grade each agent interaction against criteria you define so you can tell whether an agent is actually getting the work right rather than merely running.

## Best for

Operations, revenue operations, and growth people who understand a process well enough to encode it but do not want to write and host code: Gumloop gives them agents with real tool access, a shared knowledge base, and per-run cost visibility, with no seat charge as the rest of the team starts using what they built.

## Not the right fit for

- Teams that need a free tier to evaluate on; Gumloop removed its free plan and now offers only a 14-day trial on the $37 Pro plan.
- Anyone driving high-volume automation from webhooks on Pro, where hitting the 5 concurrent run cap returns an HTTP 429 rather than queuing the request.
- Buyers who need predictable flat monthly costs; credit consumption varies with model choice, tool calls, and conversation length, and does not roll over.
- Engineering teams that would rather own the orchestration layer in code, who are better served by a self-hostable framework or by n8n running on their own infrastructure.
- Simple two-app syncs with no AI component, where a general automation tool costs less and adds less conceptual overhead.

## Features

### Agents

The primary object: a configured assistant with tools, memory, and an audience.

- **Instruction and tool configuration**: Each agent gets a system prompt and an explicit allowlist of connected apps and MCP servers, so its blast radius is decided at build time rather than inferred from the conversation.
- **Subagent orchestration**: An agent can delegate to other agents, which keeps a long process from collapsing into one enormous prompt and lets specialized agents be reused across parents.
- **Skills**: Reusable instruction packs that encode how a specific job is done at your company, shareable organization-wide and syncable with a GitHub repo on Enterprise.
- **Reflections**: Agents periodically review their own past runs, identify recurring mistakes, and propose edits to their own skills for a human to approve.
- **Artifacts**: Agents produce files and documents as first-class outputs, and an artifact can be published on its own URL with a custom domain for sharing outside the workspace.
- **Human in the loop**: Tool approval settings, org-level app rules, and an Ask Question ability that pauses the run and waits for a person before a consequential write.

### Knowledge and context

What the agent knows before it starts reasoning.

- **Company Brain**: A shared knowledge base indexing Google Drive, Notion, SharePoint, and other connected sources so agents search what the company already knows instead of being briefed each time.
- **Permission-preserving indexing**: Drive sources can retain their original file permissions, including access granted through Google Groups, so a shared agent returns only what the asker is entitled to see.
- **Call recordings in Brain**: Gong recaps and transcripts sync into Brain with per-workspace selection, which is what makes pre-call prep and post-call CRM updates possible without manual copying.
- **Evaluations**: Every agent interaction can be graded automatically against criteria you write, with structured data collected and conversations tagged for reporting.

### Workflows

The deterministic node canvas, still present and callable from agents.

- **Visual node canvas**: Inputs, outputs, routers, filters, join paths, and an error shield node assembled by dragging, with no code required for the common shapes.
- **AI nodes**: Purpose-built nodes for extraction, categorization, scoring, list sorting, report generation, web research, and image and video analysis rather than one generic prompt block.
- **Loop mode**: Run a branch once per item in a list, which is how a single flow processes a spreadsheet of accounts without duplicating nodes.
- **Subflows**: Package a working sequence and call it from other flows, so a shared enrichment routine is maintained in one place.
- **Checkpoint history and node versioning**: Save points you can roll back to, plus pinned node versions, which matters because an upstream node change can silently alter output shape.
- **Custom node builder**: Write your own node when nothing in the library fits, including code that runs in a sandbox with managed secrets.

### Triggers and interfaces

How work starts and where people meet the agent.

- **Slack and Microsoft Teams agents**: Agents respond to @mentions in channels, threads, and DMs, and Enterprise organizations can expose an agent to everyone in a Slack workspace including people without a Gumloop account.
- **Agent email inbox**: Each agent can have a dedicated email address, so a process can be started by forwarding a message rather than by opening the product.
- **App and schedule triggers**: Cron schedules, webhooks, and native event triggers for Gmail, Outlook email and calendar, Slack, and Monday.com among others.
- **AI-built polling triggers**: Describe a condition across several apps and Gumloop generates the polling trigger, which covers cases where no native event exists.
- **Hosted pages and interfaces**: Publish an agent or a flow as a standalone page on gumloopagents.com, optionally on a custom domain, for use by people outside the team.
- **Gumball**: A personal agent launched in 2026 with a daily briefing newsletter, meeting prep summaries, and a smart inbox, which is the individual-productivity end of the platform.

### Connectors, models, and governance

The integration surface and the controls around it.

- **Around 300 connectors**: Native and MCP-based integrations spanning CRM, support, data warehouse, ads, SEO, and developer tools, with Salesforce, HubSpot, Attio, Apollo, Gong, Zendesk, and DataForSEO among the go-to-market relevant ones.
- **Custom and hosted MCP servers**: Connect your own MCP server, and on Enterprise deploy one from a GitHub repo, proxy an external one through a managed gateway, or reach a server inside your network through a managed tunnel with no inbound ports.
- **35 plus models with BYOK**: Claude, GPT, Gemini, Grok, DeepSeek, and open-source options, with bring-your-own-key so model tokens are billed by your provider instead of consuming credits.
- **App policies and rules**: Block or tag specific tool calls per app, require corporate email domains on OAuth connections, and claim a provider workspace so only your members can connect to it.
- **Run log and AI spend insights**: Per-node execution time and credit consumption on every run, plus organization dashboards and an analytics agent for adoption and spend.
- **Gumloop MCP and CLI**: The platform exposes itself as an MCP server and ships a CLI, so a developer can drive flows from a coding agent or a build pipeline.

## Use cases

- **RevOps lead at a 40-person B2B software company**: Inbound demo requests arrive with a work email and nothing else, and the two SDRs spend the first twenty minutes of every day hand-researching accounts before they can prioritize. Outcome: A triggered agent fires on each new form submission, enriches the company, checks whether the domain already exists in HubSpot, reads recent news and the site's pricing page, writes a three-line summary and a fit score onto the record, and posts anything above threshold into a Slack channel. Research time drops to review time.
- **Head of sales who wants call notes to actually reach the CRM**: Gong records everything and nobody updates Salesforce, so pipeline reviews run on stale fields and the forecast is assembled by hand each Friday. Outcome: Gong transcripts sync into Brain, a post-call agent extracts next steps, competitors mentioned, and stated timelines, writes them to the opportunity, and a scheduled forecast agent produces the roll-up on Thursday night with the exceptions flagged for a human to check.
- **Solo founder running content and SEO**: Competitor tracking, keyword research, and brief writing are all things that should happen weekly and in practice happen when there is time. Outcome: A weekly workflow pulls ranking changes from a DataForSEO connector, scrapes competitor pages that moved, drafts briefs into Google Docs, and posts a digest by email. The whole thing runs on the $37 plan because the volume is small and the models chosen are cheap.
- **Operations manager at an agency with shared client tooling**: Ten account managers need the same reporting and QA steps done consistently, but each has their own half-documented version and quality varies by person. Outcome: One person builds the agent and encodes the checks as Skills; the rest use it from Slack. Because seats are unlimited, adding the tenth user costs nothing, and evaluations show which client accounts the agent is getting wrong.

## Pricing

Usage-based. Seats are unlimited on every plan and the meter is credits, where one credit is $0.005. Credits are consumed by model chat and reasoning tokens, tool calls (minimum one credit per successful call), compute time at roughly five credits per session-minute, and an 8 percent orchestration fee on top of those. Workflow runs bill the same way. Bring-your-own-key removes token charges but raises the orchestration fee on agent chats.

- **Pro**: From $37 per month. 20,000 credits per month included, scaling up a slider for higher volumes; Unlimited seats, unlimited teams, unlimited agents, shared credentials; 35 plus models with bring-your-own-key support, plus Company Brain and GitHub skill sync; 5 concurrent workflow runs and 25 concurrent agent chats; One hosted MCP server, email support, and Gumloop University. This is the only self-serve plan. Credits do not roll over, and overage can be enabled with a default cap of $5,000.
- **Enterprise**: Custom annual contract. Custom credit volume with rollover, and discounts on the orchestration fee; Role-based access control, custom roles, SCIM and SAML, audit logs, custom retention rules; Org-wide connector policies, AI model access control, static egress IPs, optional VPC deployment; Custom MCP hosting, proxying, and managed tunnels into a private network; Queued rather than rejected runs when concurrency limits are hit, dedicated Slack support, optional embedded expert. Concurrency defaults are higher (about 15 workflow runs and 100 agent chats) and adjustable.

Billing notes:

- There is no free plan as of August 2026; the only way to evaluate without paying is the 14-day trial on Pro.
- One credit equals $0.005, so the 20,000 included credits on Pro represent roughly $100 of metered usage bundled into a $37 subscription.
- Unused credits expire at the end of the month on Pro. Rollover is an Enterprise-only term.
- The 8 percent orchestration fee applies on top of chat, tool, and compute charges. On bring-your-own-key agent chats the fee is higher, calculated on what the run would have cost, so BYOK saves money on tokens but is not free.
- Overage is opt-in with a default $5,000 cap on Pro accounts, which is the control that stops a runaway loop from producing a genuinely alarming invoice.
- Concurrency is a hard limit on Pro: exceeding 5 simultaneous workflow runs returns HTTP 429 rather than queuing, so anything webhook-driven needs retry logic on your side.
- Cost per task varies enormously with model choice and conversation length; the same agent can cost 2 credits for a simple answer and 200 for a deep research pass.

Value assessment: The unlimited-seat model is the genuinely unusual part. A ten-person team pays the same $37 base as one person, which inverts the economics of nearly every competing tool and makes Gumloop cheap for the pattern it is designed around: one builder, many users. The offsetting risk is that the credit meter is opaque until you have run real work through it, and the removal of the free plan means you now discover your actual cost during a 14-day paid trial rather than over a leisurely month on a free tier. For a small business, the honest budgeting approach is to assume the $37 covers exploration and light production only, and to instrument the run log early so you know which agent is eating the month.

## Strengths

- Unlimited seats on every plan, so adoption across a team costs nothing extra and the builder is not rationing access.
- Genuinely non-technical building surface that still reaches real systems, with roughly 300 connectors and first-class MCP support.
- Per-node credit and runtime reporting in the run log, which is better cost instrumentation than most automation platforms offer.
- Company Brain with permission-preserving indexing, which is what makes a shared Slack agent safe rather than a data leak waiting to happen.
- Agents meet people in Slack, Teams, and email instead of requiring everyone to learn a new interface.
- Skills plus reflections give the prompt layer somewhere to live and a mechanism to improve, rather than knowledge being trapped in one person's instructions.
- Built-in evaluations let you measure whether an agent is producing correct work, not just whether it ran.
- Well-funded and growing, with a $17 million Series A in 2025 and a $50 million Series B led by Benchmark in March 2026.

## Limitations

- The free plan was discontinued; evaluation is now a 14-day trial on a paid plan, which is a real barrier for a cautious small buyer.
- Credit consumption is hard to predict before you build. Users report a workflow going from one or two credits per run to dozens after a small change, and credits do not roll over on Pro.
- The 8 percent orchestration fee sits on top of model and tool costs, and rises on bring-your-own-key agent chats, so BYOK is less of a savings lever than it first appears.
- Pro concurrency is a wall, not a queue: 5 simultaneous workflow runs, then HTTP 429. Queuing is an Enterprise feature, which pushes any genuinely high-volume webhook use case upmarket.
- The trigger catalog is narrower than mature automation platforms; several common app events still require the AI-generated polling trigger rather than a native one.
- Custom nodes and less-used connectors can be brittle and occasionally need manual repair, which is the usual cost of a fast-moving connector library.
- The pricing page groups workflow limits under a heading marked Legacy, a signal that investment is concentrated on the agent surface and that flow-first builds may see less attention over time.
- Enterprise governance features that most compliance reviews ask for, including SSO, audit logs, RBAC, and custom retention, are entirely absent from the self-serve plan.

## Comparisons

- **Gumloop vs Cargo**: Both sit in the space between a CRM and the rest of the stack, but they start from opposite ends. Cargo is built specifically for revenue operations, with a warehouse-native model, routing, and scoring as the native vocabulary. Gumloop is a general agent platform that happens to be popular with go-to-market teams. Pick Cargo when the problem is pipeline orchestration on top of a data warehouse; pick Gumloop when the same team also needs research agents, content workflows, and a Slack assistant on one bill.
- **Gumloop vs ScrapingBee**: Not competitors so much as different layers, and worth comparing because people evaluate them for the same job. ScrapingBee is an API you call to fetch a page reliably, handling proxies, JavaScript rendering, and blocks. Gumloop is the orchestration around that fetch: what to do with the page, how to extract from it, where to write the result. Heavy scraping is cheaper and more reliable through a dedicated API, and the two are commonly used together, with ScrapingBee as a step inside a Gumloop flow.
- **Gumloop vs Clay**: The closest overlap for go-to-market buyers. Clay is a spreadsheet-shaped enrichment and outbound tool with dozens of data providers behind a waterfall, and it is better than Gumloop at the specific job of filling in contact and company data at list scale. Gumloop is broader and less opinionated: it can call the same providers but is really a general agent builder. Teams doing high-volume prospecting usually keep Clay; teams automating a wider set of internal processes end up on Gumloop, and plenty run both.
- **Gumloop vs n8n**: n8n is the choice when engineering wants ownership: it is source-available, self-hostable, has a far deeper trigger and node catalog, and its pricing is per execution rather than per credit, which is much easier to model. Gumloop is friendlier to non-engineers and further ahead on the agent surface, with Brain, Skills, and Slack-native chat that n8n does not match out of the box. If you have someone willing to run infrastructure, n8n is cheaper at volume; if you do not, Gumloop removes the operational burden.
- **Gumloop vs Zapier**: Zapier has an order of magnitude more integrations and a trigger catalog Gumloop cannot approach, and for simple app-to-app syncs it is faster to set up and cheaper to run. Gumloop is a different product for a different job: multi-step reasoning, document and knowledge work, and agents people converse with. Zapier's own AI agent features have narrowed the gap, but Gumloop's shared knowledge base, skills, and unlimited seats still make it the stronger choice when AI is the point rather than an add-on.
- **Gumloop vs Bardeen**: Bardeen began as a browser extension for automating things you do in a browser tab, and it retains an advantage on in-browser, user-triggered actions on sites without good APIs. Gumloop runs server-side, which makes it the right tool for scheduled and event-driven work that must happen whether or not a laptop is open. For a go-to-market team wanting agents that run overnight, Gumloop is the better fit; for accelerating a person's manual browser work, Bardeen is.

## Implementation

- Setup time: A first useful agent in an afternoon: connect two or three apps, write instructions, test in the canvas, then invite it into a Slack channel. A production process with triggers, error handling, and evaluations is more like a week of part-time work, most of which is spent discovering the edge cases in your own data.
- Learning curve: Low to start and moderate to do well. The canvas and the agent editor are approachable, but the skills that separate a demo from something dependable are prompt discipline, knowing when to force a step into a deterministic workflow instead of trusting the model, and reading the run log to find which node is consuming the credits.
- Onboarding: Self-serve, with Gumloop University courses, a template library, community templates, live workshops, learning cohorts, a public forum, and a directory of paid experts for hire. Enterprise adds a dedicated Slack channel and an optional embedded Gumloop expert.
- Migration: There is no import path from Zapier, Make, or n8n; flows are rebuilt by hand. Reconnecting OAuth credentials is usually the longest part of a move, and running the old and new systems in parallel for a cycle is worth the duplicated cost. Note that credits do not roll over, so timing a migration to the start of a billing month avoids paying twice for the same work.

## Platform, API & security

- Platforms: Web application, Slack, Microsoft Teams, Email (per-agent inbox), Hosted agent pages with custom domains, Chrome extension, CLI
- API: REST API for starting and monitoring runs, webhooks as triggers, a Gumloop MCP server so external coding agents can drive the platform, and a CLI. Custom MCP servers can be connected, and on Enterprise hosted from GitHub, proxied, or reached through a managed tunnel into a private network.
- Compliance: SOC 2 Type II, GDPR, Zero data retention options with model providers
- Data residency: US-hosted by default; Enterprise offers virtual private cloud deployment and static egress IPs for firewall allowlisting.
- SSO: SAML single sign-on and SCIM provisioning on Enterprise only, along with Okta integration; not available on the Pro plan.
- Security notes: Governance is concentrated in the Enterprise tier: role-based access control, custom roles, audit logging, app rules that block or tag specific tool calls, OAuth domain restrictions, provider workspace claims, model access control, and custom data retention. Code execution happens in a sandbox with managed secrets, and human-in-the-loop approvals can gate any write action.

## Support

- Channels: Email support (Pro), Dedicated Slack channel (Enterprise), Public community forum, Gumloop University courses and workshops, Optional embedded expert (Enterprise)
- Documentation: Detailed documentation covering core concepts, every node, and each connector, published with an llms.txt index so AI assistants can read it directly. Node-level pages include worked examples.
- Community: An active public forum, a template gallery with community submissions, a creator program, live learning cohorts, and a directory of independent experts available for hire, which is a meaningful signal of ecosystem depth for a company this young.

## Company

- Founded: 2023
- Founders: Max Brodeur-Urbas, Rahul Behal
- Headquarters: San Francisco, California (founded in Vancouver, Canada)
- Ownership: Independent, venture-backed (legal entity AgentHub Inc.)
- Employees: Small team by design; not formally disclosed
- Funding: Y Combinator alumnus; $3.1 million seed, a $17 million Series A in January 2025 led by Nexus Venture Partners, and a $50 million Series B in March 2026 led by Benchmark.

Funding history:

- Seed (2024): $3.1M. Following Y Combinator.
- Series A (2025): $17M. Led by Nexus Venture Partners, with First Round Capital and Y Combinator; company relocated to San Francisco.
- Series B (2026): $50M. Led by Benchmark, with Nexus, First Round, Y Combinator, Box Group, and Shopify Ventures.

Timeline:

- 2023: Founded by McGill classmates Max Brodeur-Urbas and Rahul Behal in Vancouver as a no-code, drag-and-drop AI automation canvas.
- 2024: Goes through Y Combinator and raises a $3.1 million seed round; the node library and connector catalog expand quickly.
- 2025: Raises a $17 million Series A led by Nexus Venture Partners and relocates to San Francisco.
- 2025: Product pivots toward agents: Company Brain, Skills, and Slack-native agent chat move ahead of the flow canvas as the headline surface.
- 2026: Closes a $50 million Series B led by Benchmark in March, citing adoption at Shopify, Ramp, Gusto, Instacart, Samsara, and Opendoor.
- 2026: Self-serve pricing consolidates to a single $37 Pro plan with a 14-day trial; the free tier is retired.
- 2026: Gumball personal agent ships with a daily briefing, meeting prep, and smart inbox; Outlook, Gong, SharePoint, and Monday.com triggers and connectors land through the summer.

## Integrations

Salesforce, HubSpot, Attio, Apollo, Gong, Slack, Microsoft Teams, Gmail and Google Calendar, Outlook, Google Drive and Google Sheets, Notion, SharePoint, Airtable, Linear and Jira, GitHub, Zendesk, Monday.com, Meta Ads and Snapchat Ads, DataForSEO and Ahrefs, Databricks and ClickHouse, Apify, Exa, Custom MCP servers

## FAQ

### What is Gumloop used for?

Building AI agents and automated workflows without code. A person configures an agent with instructions, connected apps, and a knowledge base, then either chats with it in Slack or Teams or leaves it running on a schedule or trigger. Common uses are lead enrichment and research, CRM hygiene from call recordings, inbox triage, report generation, SEO and content pipelines, and internal question answering over company documents.

### Does Gumloop have a free plan?

No. As of August 2026 the public pricing page lists only Pro at $37 per month with a 14-day free trial, and Enterprise at custom pricing. Older reviews still describe a free tier with a few thousand monthly credits; that plan has been retired, so plan your evaluation around the trial window.

### How much does Gumloop cost?

Pro starts at $37 per month and includes 20,000 credits, with a slider to buy more credits at higher monthly prices. Seats are unlimited on every plan, so cost scales with usage rather than headcount. Enterprise is quoted individually and adds credit rollover, governance features, higher concurrency, and discounts on the orchestration fee.

### What is a Gumloop credit worth?

One credit equals $0.005, so $1 buys 200 credits and the 20,000 credits on Pro represent about $100 of metered usage. Credits are consumed by model tokens for chat and reasoning, a minimum of one credit per successful tool call, roughly five credits per minute of active compute, and an 8 percent orchestration fee on top of those.

### Do unused Gumloop credits roll over?

Not on Pro. Unused credits expire at the end of the billing month, and rollover is an Enterprise term. If you exceed the included allowance you can enable overage, which on Pro is capped at $5,000 by default, so the runaway-loop scenario has a ceiling you control.

### Does using my own API key make Gumloop cheaper?

Partly. With bring-your-own-key, model tokens are billed by your provider rather than consuming credits, which removes the largest line item. But the orchestration fee on agent chats is higher under BYOK, calculated against what the run would otherwise have cost. Workflow AI nodes on your own key carry no orchestration fee, so the saving is real but uneven across the two surfaces.

### Gumloop vs Zapier: which should a small business choose?

Zapier for breadth of simple integrations and predictable per-task pricing; Gumloop when the automation needs judgment. Gumloop's advantages are its shared knowledge base, reusable skills, Slack-native agent chat, and unlimited seats. Zapier's are a far larger connector and trigger catalog and lower cost for straightforward two-app syncs. Many teams keep Zapier for plumbing and add Gumloop for the AI-heavy processes.

### Gumloop vs n8n: what is the real difference?

Ownership and audience. n8n is source-available and can be self-hosted, has a deeper node and trigger catalog, and bills per execution, which makes costs easy to model. Gumloop is fully hosted, easier for non-engineers, and considerably further along on agents, knowledge, and evaluation. If you have engineering capacity and volume, n8n is cheaper; if you do not, Gumloop removes the infrastructure question.

### Can Gumloop agents access my company documents safely?

Yes, through Company Brain, which indexes connected sources such as Google Drive, Notion, SharePoint, and Gong transcripts. Google Drive sources can preserve original file permissions, including access via Google Groups, so an agent shared in a public Slack channel returns only what the person asking is already allowed to see. Enterprise adds app rules, domain restrictions, and audit logging on top.

### How many concurrent runs does Gumloop allow?

Pro allows 5 concurrent workflow runs and 25 concurrent agent chats. Exceeding those on Pro produces an HTTP 429 for API calls and an error in the interface rather than a queued run, so webhook-driven use cases need retry logic. Enterprise defaults to about 15 and 100 respectively, is adjustable, and queues excess runs instead of rejecting them.

### Does Gumloop support MCP servers?

Extensively. Many of its connectors are MCP-based, you can connect your own MCP server on any plan, and Gumloop itself exposes an MCP server so an external coding agent can drive it. Enterprise adds hosting an MCP server from a GitHub repository, proxying an external one through a managed gateway, and reaching servers inside a private network via managed tunnels without opening inbound ports.

### Who owns Gumloop and is it financially stable?

It is an independent, venture-backed company operating as AgentHub Inc., founded in 2023 by Max Brodeur-Urbas and Rahul Behal, now headquartered in San Francisco. It went through Y Combinator, raised a $17 million Series A in January 2025, and a $50 million Series B led by Benchmark in March 2026. That is a well-capitalized position, though as with any three-year-old company the product surface is still moving quickly.

### Is the workflow builder being deprecated?

It is not being removed, and agents can still call workflows as tools, which is the recommended pattern for steps that must run identically every time. But the pricing comparison table now groups workflow limits under a heading marked Legacy, and recent releases have concentrated on agents, Brain, and skills. If you are choosing Gumloop today, build with agents as the primary object and use flows for the deterministic parts.

## Editorial verdict

Gumloop is the most credible no-code agent platform for a go-to-market team that has processes worth encoding and nobody free to write code. The unlimited-seat model is the standout commercial decision: one person builds, the whole company uses it, and the base subscription does not move. Company Brain with permission-aware indexing, skills that give the prompt layer a home, and evaluations that grade output are the parts competitors have not matched. The reasons for caution are commercial rather than technical. The free plan is gone, so evaluation now happens inside a 14-day paid trial; credits are hard to forecast, do not roll over, and carry an orchestration fee on top; and the governance features a compliance review will ask about live entirely on Enterprise. Buy it if the work you want automated genuinely needs judgment, budget for usage rather than a flat subscription, and instrument the run log from day one so you find out which agent is expensive before the invoice does.

---

Source: SaaSTracker (https://saastracker.org), an independent editorial project. This profile is compiled from public information, carries no peer reviews or paid placement, and was last reviewed 2026-08-23. Awards are judged on published criteria: https://saastracker.org/methodology