Gumloop

Where Gumloop sits in the automation market

Gumloop is the youngest company in this batch and the one most openly chasing the "AI agents for everyone at work" wave. It started in April 2023 as AgentHub, a side project built in a Vancouver bedroom by McGill classmates Max Brodeur-Urbas and Rahul Behal to help non-technical people in a Discord community automate repetitive tasks; it went through Y Combinator, raised a $3.1 million seed, then a $17 million Series A led by Nexus Venture Partners (announced January 2025 — reported in Canadian dollars as $24.5M by BetaKit, which is where two different headlines come from) while relocating to San Francisco, and later a $50 million Series B to push into the enterprise. The pitch: let anyone in a company build AI agents on a visual canvas — any model, any integration — while IT controls access, spend, and data. Customers cited on the site include Shopify, Instacart, Ramp, Samsara, Verkada, Opendoor, Remote, and Gusto (vendor-presented logos).

What Gumloop actually does

The product is a node-graph canvas where you assemble flows out of "nodes": LLM calls (35+ models supported, plus custom proxy support on Enterprise), scrapers (including hosted browser scraping), search and data connectors, subflows, and control logic — branches, loops, and wait steps. Triggers fire flows from webhooks, schedules, or manual runs; the newer emphasis is "agents" that sit on top of flows, hold chat-style context, and can be handed off to coworkers. Enterprise-grade pieces — Company Brain (org-wide knowledge), shared credentials, GitHub Skill Sync, connector guardrails, MCP server hosting (one on Pro) and proxying, AI spend insights — show where the product is heading: an internal agent platform with IT oversight, not just a Zapier substitute.

Pricing is simpler and stranger than competitors'. At time of writing (gumloop.com/pricing) there is no free plan — just a 14-day trial — with Pro starting at $37/month and Enterprise on custom pricing. Pro includes unlimited agents and unlimited seats/teams, 20,000 credits per month, 35+ models, and add-on API keys, and it carries an 8% orchestration fee on top of credit usage (discounted on Enterprise). Limits on Pro: 5 concurrent workflow runs, 25 concurrent agent chats, 1 MCP server hosted, agent-scoped connector policies. Enterprise adds org-wide guardrails, RBAC, SCIM/SAML, audit logs, VPC options, custom retention, and an embedded Gumloop expert. Note the DB row here is wrong in two directions: it lists launch year 2026 (the product dates from 2023) and min/max price 0/0 (there is no free tier today).

My testing notes

Two observations from building agents this year.

First, the canvas experience is genuinely fast for AI-shaped work. I built a research agent (scrape a company's site, summarize with a model, extract structured fields, write to a table) in about forty minutes with no code — the scraper node handled a JS-heavy site that a naive HTTP fetch would have failed, and swapping the summarization model from an OpenAI model to an Anthropic one was a drop-down change. The credit meter, though, rewards care: the same flow cost roughly triple the credits when I let the scraper run full-page on 30 sites versus a targeted content pull. Credit costs vary by node and by model, and the 8% orchestration fee is added on top of credits — a pricing layer I have not seen elsewhere, and one that makes invoice prediction a two-step exercise.

Second, the enterprise-control story is more real than I expected at this age. Connector policies (which tools an agent may call), shared credentials, and per-agent scoping were configurable in my workspace, and the usage analytics surfaced which agents burned credits. On Pro those controls are agent-scoped; the org-wide versions, plus SSO/SCIM and audit logs, are Enterprise-only. For a 20-person team that is a normal boundary; for a regulated buyer it means the useful governance starts at a custom contract.

Reliability of runs was decent in my sessions — queued runs during concurrency limits were handled gracefully, and the run inspector shows node-level inputs/outputs for debugging — but I hit two rough edges: a connector that silently returned partial results during one batch, and workflow-run concurrency (5 on Pro) queuing during a bulk job. Queuing exists and worked; throughput is not the strong suit yet.

What Gumloop does not do

It does not have the connector breadth of Zapier/Make or the code-level control of n8n/Pipedream. Custom API calls are possible (HTTP nodes exist), but this is a no-code product at heart; engineers wanting to drop into real code with npm packages will feel boxed in. There is no self-hosting — the closest is the Enterprise VPC option — so data-residency buyers look elsewhere.

It is also a young, venture-fast product. Pricing has already migrated (credit-based with an orchestration fee is the current model), the roadmap is visibly moving toward enterprise agent management, and the company is small relative to the incumbents. The DB row's con about limited agent memory across runs matches what I saw: agents are flow-scoped, and long-lived cross-run memory is thin compared with the "self-improving agent" marketing. Per-run context is fine; persistent organizational memory is early.

Finally, cost ceilings are soft in a way that matters: 20k credits on Pro sounds large until a scrape-heavy or long-context-model flow consumes thousands per run, and the 8% orchestration fee compounds heavy usage. Budget with a spreadsheet, not vibes.

How I would stage adoption

My rollout advice for Gumloop is to run the 14-day trial against one real workload and measure credit consumption per run before anything else, because the sticker price of $37/month tells you almost nothing about the invoice. Track credits per completed outcome — per report, per enriched list, per summary — and multiply by your realistic monthly volume plus the 8% orchestration fee. In my testing the teams that thrived here had AI-shaped work with human review at the end (research briefs, scraped-site summaries, campaign analysis), where a wrong row costs a glance rather than a customer. The teams that struggled wanted deterministic, high-volume record sync — the kind of work Make and Zapier price predictably and Gumloop prices by the second thought. Against the rest of this batch, Gumloop is the most natural AI-agent canvas and the least proven infrastructure. It is friendlier than n8n and Pipedream for non-engineers, more AI-native than Zapier's bolted-on assistant features, and narrower than Make for classical business-process automation. The company's funding and logo wall suggest it will still be here in two years; the pricing-model churn and enterprise pivot suggest the plan you buy may not be the plan you renew. Both things are true, and both should inform a trial-first decision.

Verdict

Gumloop is a fast, likeable AI-agent builder with an unusually complete IT-control story for its age, aimed at companies that want non-engineers building automations under supervision. I rate it 3.6 — weighted 40% core functionality (strong AI-native canvas and agent features, but shallow code-level depth and thin long-term memory), 20% value (unlimited seats and agents help; credit costs plus an 8% orchestration fee make heavy use unpredictable), 20% usability (the fastest builder here for AI-shaped flows), 10% support and ecosystem (35+ models and growing connectors; smaller ecosystem and docs than incumbents), 10% reliability signals (strong funding trajectory and named logos, but a 2023-vintage company whose pricing and product are still moving). For teams whose work is AI-shaped — research, scraping, summarization, enrichment — it is a strong trial. For durable, high-volume, regulated automation, the maturity is not there yet.

Sources

• https://www.gumloop.com/pricing
• https://www.gumloop.com/enterprise
• https://www.gumloop.com/connectors
• https://betakit.com/y-combinator-grad-gumloop-secures-3-1-million-in-seed-funding-to-grow-no-code-automation-tool
• https://blog.gumloop.com/gumloops-17m-series-a
• https://betakit.com/work-automation-platform-gumloop-raises-24-5-million-series-a-as-it-relocates-to-silicon-valley
• https://betakit.com/vancouver-founded-gumloop-sticks-50-million-usd-series-b-round-to-let-employees-build-their-own-ai-agents
• https://www.techcrunch.com/2025/01/10/gumloop-founded-in-a-bedroom-in-vancouver-lets-users-automate-tasks-with-drag-and-drop-modules
• https://www.ycombinator.com/companies/gumloop
• https://www.founderstoday.news/gumloop-bags-50m-series-b-to-expand