Gumloop

What it is

Gumloop is an AI agent platform with a no-code workflow canvas underneath it. That ordering is the change worth knowing: the product used to lead with the canvas — drag nodes, connect them, run a flow to completion on a trigger — and now leads with agents. The canvas is what an agent calls when a step needs to be repeatable rather than reasoned.

Three surfaces matter to a buyer. Agents run in chat, inside Slack, from an agent email inbox, or on schedule and event triggers. Workflows are the canvas: a trigger node, enrichment nodes, model nodes (35+ models, selectable per node, or bring your own API key), and action nodes that write results back into your apps. Brain is the retrieval layer — index Notion pages, Google Drive documents, public Slack channels, GitHub repos, Confluence spaces, Zendesk articles and tickets, Salesforce records, and uploaded files, scoped personal / team / organization. For Google Drive and Salesforce, Brain can inherit source permissions, so retrieval returns only what the asking user could already open.

The connector library is now 300+, up from the 130+ Gumloop advertised in mid-2026, with newer additions arriving as MCP servers rather than hand-built nodes. Gumloop launched in YC W24, has raised $70.1M — a $3.1M seed (First Round, July 2024), a $17M Series A (Nexus Venture Partners, January 2025), and a $50M Series B led by Benchmark (March 2026) — and names Shopify, Instacart, Gusto, Samsara, Ramp, and Opendoor as customers.

Why it shows up in RevOps, Legal, and Recruiting stacks

The same platform covers three verticals because the underlying job is identical: take unstructured inputs, run a model over them, push structured outputs into the system of record.

• RevOps runs prospect enrichment — pull a list from HubSpot, scrape each company site, extract ICP signals with a model node, write the fields back — plus outbound personalization and CRM hygiene jobs that would otherwise need a Python script and someone to own it.
• Legal ops runs document intake: ingest PDFs from a shared Drive folder, extract clause data and party names, route anything flagged to a Slack channel or the contract intake queue. This is where the per-node model choice earns its keep — clause extraction on one model, summarization on a cheaper one.
• Recruiting runs CV screening against a job description and pushes qualified rows into the ATS with a recruiter notification, replacing a manual sort that typically runs 2–3 hours a day at volume.

Pricing reality

The free plan is gone. Gumloop previously ran a $0 tier with 5,000 credits a month; the pricing page now lists two options and a 14-day trial that converts to paid unless cancelled. If your evaluation plan was “build it on free, upgrade if it works,” that path no longer exists.

• Pro — $37/month. 20,000 credits, unlimited seats, unlimited agents and teams, 5 concurrent workflow runs, 25 concurrent agent chats, 35+ models, bring-your-own-key, Brain, GitHub skill sync, shared credentials, agent-scoped guardrails, one hosted MCP server, email support.
• Enterprise — custom. Defaults rise to 15 concurrent workflow runs and 100 concurrent agent chats, both negotiable; adds request queueing, org-wide guardrails, VPC deployment, SCIM/SAML, role-based access control, audit logs, an admin dashboard, custom MCP hosting and proxying, managed tunnels for private MCP servers, a negotiated orchestration rate, dedicated Slack support, and an embedded Gumloop expert.

What a team actually pays is the credit line, not the $37. Gumloop meters at 1 credit = $0.005 and passes model token cost through at that conversion. An agent chat bills on five components: chat and reasoning tokens; a minimum of 1 credit per successful tool call; compute at 5 credits per session-minute; an 8% orchestration fee on the sum of those three; and the full cost of any workflow the agent triggers. Supply your own model key and the reasoning component disappears from your credit balance — but the orchestration fee doubles to 16%. Workflows bill more simply: 1 credit to start plus the cost of each node that runs, most native nodes free, with no compute or orchestration charge.

Budget against the 20,000 credits, not the subscription. A workflow-heavy team on mostly native nodes will not approach the ceiling. An agent-heavy team will: compute alone puts a 60-minute agent session at 300 credits before a single token or tool call is counted.

Best for

RevOps, Legal Ops, and Recruiting teams in the 5–300-person range with no engineering support, running document-heavy pipelines or multi-model work that no single point tool covers. ROI is clearest when the platform replaces at least one recurring manual task consuming 3+ hours a week.

Not for you if your automations are deterministic data movement with no reasoning step — you will pay agent-platform pricing for plumbing that Zapier or Make does for less. Also not for you if you need a durable free tier to prototype against, or if a single team’s batch jobs regularly exceed 5 concurrent runs: on Pro those requests are rejected, not queued, and the fix is an Enterprise contract rather than a plan upgrade.

Alternatives and when to pick them

• Zapier — the market-share leader, with the widest connector catalog. Pick it when the work is deterministic plumbing (move data from A to B when C fires) and no reasoning step is involved. Its per-task pricing scales badly for model-heavy pipelines, which is exactly where Gumloop’s flat $37 plus metered credits wins.
• Make — pick it for complex branching logic, a wide connector library, and lower per-operation cost at high volume, over native document intelligence. Make still runs a free tier; Gumloop no longer does, which now matters more than it did.
• n8n — the fastest-growing entrant, and the pick when you have engineering support, need data residency, or want to own the infrastructure. The self-hosted Community Edition is free; cloud starts at €20/month billed annually for 2,500 executions. Connector depth exceeds Gumloop’s, but document-intelligence steps mean writing code instead of picking a block.
• Lindy — pick it when the automation is a persistent watcher rather than a pipeline: monitoring an inbox, triaging inbound, reasoning across CRM state over time. Lindy is priced per user (Plus $29.99/user/month for 3,000 credits, Pro $99.99 for 15,000); Gumloop’s seats are unlimited and the credits are pooled, so the two invert depending on whether your constraint is headcount or volume.

Watch-outs

• The orchestration fee is charged on top of everything, and doubles under BYOK. Teams bring their own API key expecting to cut cost, then find the platform fee moved from 8% to 16% of chat, compute, and tool-call credits. Guard: run the same flow both ways on a 10-row test set and read the credit dashboard before committing to a key strategy — BYOK only wins when your negotiated model rate beats Gumloop’s pass-through by more than the extra 8%.
• Agent compute bills by wall-clock, not by work done. At 5 credits per session-minute, a long-running agent that spends most of its time waiting on slow tool calls burns credits at the same rate as one doing real reasoning. Waiting on user input is free; waiting on an API is not. Guard: move long, deterministic stretches into a workflow, which carries no compute or orchestration charge, and let the agent call it.
• Pro rejects excess concurrency instead of queueing it. Past 5 concurrent workflow runs the API returns HTTP 429 and agent chats fail with a concurrency error; queueing is an Enterprise feature. A 500-row batch fired as parallel runs will drop work silently if nothing retries. Guard: stagger triggers, break large jobs into subflows (which do not count against the limit), and build retry-on-429 into anything calling the API.
• Hosted MCP is capped at one server on Pro. Exposing several internal APIs over MCP — for Claude Code agents, cross-flow routing, or private services behind a managed tunnel — requires Enterprise. Guard: if MCP topology is a requirement rather than a nice-to-have, get Enterprise pricing before you architect around it.