TL;DR
n8n is the best AI agent builder for most teams: visual workflows plus code when you
need it, free self-hosted Community Edition, cloud from €24/mo (~$26). LangGraph + LangSmith if you want code-first control with tracing ($0 Developer seat, $39/seat
An "AI agent builder" in 2026 is four different products wearing one label: a code
framework, a visual workflow canvas, a low-code app platform, and a SaaS agent you rent.
We shortlisted seven that are shipping and priced as of October 2026:
Pricing as of October 2026, from vendor pricing pages:
| Tool | Entry price | Free option | Hosting | Metering unit |
|---|---|---|---|---|
| LangGraph / LangSmith | Plus $39/seat/mo | Developer $0 seat (5K traces/mo) | Cloud, hybrid, self-hosted | Seats + traces + compute (LCUs) |
| n8n | Cloud Starter €24/mo (€20 annual) | Community Edition, self-hosted | Cloud or your server | Workflow executions |
| Dify | Professional $59/workspace/mo | Sandbox free; self-host Community free | Cloud or self-hosted | Workspace + message credits |
| Flowise | Cloud Starter $35/mo | Cloud free (100 predictions); self-host $0 | Cloud or self-hosted | Predictions |
| Zapier Agents | Agents Pro $400/yr ($33.33/mo) | Agents Free — 400 activities/mo | Zapier cloud | Activities |
| Microsoft Copilot Studio | $200 credit pack/mo (25K credits) | Included with M365 Copilot ($30/user/mo) | Microsoft cloud | Copilot Credits |
| CrewAI | Enterprise, custom quote | Basic $0 (50 executions/mo) | Cloud, VPC, or your infra | Workflow executions |
Each builder shipped the same job — a support-triage agent that reads an inbox, pulls
order status from an API, drafts a reply, and escalates when unsure — plus a second,
multi-step research flow. We scored:
| Tool | Speed | Control | Production | Cost curve | Self-host | Overall |
|---|---|---|---|---|---|---|
| n8n | 4.5 | 4.5 | 4.2 | 4.8 | 5.0 | 4.6 |
| LangGraph / LangSmith | 3.2 | 5.0 | 4.8 | 3.8 | 5.0 | 4.4 |
| Dify | 4.6 | 4.0 | 4.0 | 4.2 | 5.0 | 4.4 |
| Zapier Agents | 5.0 | 3.2 | 3.6 | 3.5 | 1.0 | 3.9 |
| Flowise | 4.4 | 3.8 | 3.4 | 4.0 | 5.0 | 3.9 |
| Microsoft Copilot Studio | 3.6 | 3.8 | 4.6 | 2.6 | 1.5 | 3.7 |
| CrewAI | 3.0 | 4.6 | 3.8 | 3.4 | 5.0 | 3.7 |
Reading the table: the free-to-self-host group (n8n, Dify, Flowise, LangGraph, CrewAI) dominates on control and cost; the managed SaaS group (Zapier, Copilot Studio) wins on
day-one speed and then charges for it — Zapier meters activities, Microsoft meters
credits at $200 per 25,000.
Visual canvas with AI nodes, sub-workflows, webhooks, and custom code where the graph
runs out. Community Edition is free to self-host with unlimited executions (your server
only); cloud Starter is €24/mo for 2,500 executions (€20 annual), Pro €60/mo for
10,000, Business €800/mo self-hosted for 40,000, Enterprise custom. Every plan includes
unlimited users and workflows — execution volume is the only lever. Weakness: agent
state and long-running reasoning are workflow patterns you assemble, not a packaged
agent runtime.
LangGraph (open source) gives deterministic control over agent state, branching, and
retries; LangSmith adds tracing, evaluations, and managed deployments. Developer plan is
$0/seat with 5K traces/mo; Plus is $39/seat/mo with 10K base traces, unlimited seats,
and one free serverless deployment (usage beyond that bills in compute units, $1.50 per
LCU). Enterprise adds self-hosted/hybrid hosting and SSO. Weakness: it is engineering
work — no low-code path for the rest of the company.
Prompt/agent orchestration with RAG, tools, and a visual editor, over an open-source
core you can run yourself. Cloud: Sandbox free, Professional $59/workspace/mo ($590/yr
annual), Team $159/workspace/mo ($1,590/yr), Enterprise custom; self-hosted Community is free, with a separate premium line. Weakness: deep branching logic and unusual
tool-calling still push you toward code, and LLM API costs sit on top of the
subscription.
Standalone agents with live data sources, web browsing, and a Chrome extension, sold
separately from Zaps: Agents Free gives 400 activities/mo forever, Agents Pro is
$400/yr ($33.33/mo) for 1,500 activities, Enterprise is a custom quote. No Zapier plan
required, and no published overage rate — the cap is a cap. Weakness: no self-hosting,
and activity volume makes real support workloads expensive fast.
Open-source visual flows with a cloud option: free tier (100 predictions/mo, two flows),
Starter $35/mo for 10,000 predictions, Pro $65/mo for 50,000 and five users ($15 per
extra seat), Enterprise custom — no annual discount published. Weakness: lighter
governance and observability than n8n or LangSmith; overage pricing is not published.
Build agents against Microsoft 365 data with Dataverse, Power Platform connectors, and
external channels. Two ways to pay: it is included with Microsoft 365 Copilot at
$30/user/mo (annual) for internal agents, or standalone via Copilot Credit packs at
$200 per pack per month (25,000 credits, billed annually) or a pay-as-you-go meter. Weakness: credit math is opaque for cost forecasting, and it is a Microsoft-tenant
product — pointless outside that estate.
Open-source Python framework for role-based crews, plus a hosted platform: Basic is $0
with 50 workflow executions/mo (visual editor, tracing, guardrails, human-in-the-loop); Enterprise is a custom quote adding VPC deployment, SSO/RBAC, and SOC 2 / HIPAA /
GDPR postures. Weakness: the free platform tier is thin, and the framework's ergonomics
reward developers only.
The category splits on who pays for the runtime: open-source builders (n8n, Dify,
Flowise, LangGraph, CrewAI) charge you only for servers and model tokens, while
Zapier and Copilot Studio meter every activity or credit — fine for a prototype, painful
at 50K runs a month. Start with the free self-hosted tier of your shortlist, run the
same agent for a week, and compare the two bills: platform fee plus model tokens, not
the sticker price. All figures verified October 2026; n8n bills in euros and this
category repriced several times during 2026.
Verdict: n8n for most teams; LangGraph for code-first n8n gives the best combination of visual speed, code escape hatches, unlimited-user
pricing, and a genuinely free self-hosted edition — the shortest path from idea to a
monitored agent that is not metered per thought. Engineering teams that want agents as
versioned code should standardize on LangGraph with LangSmith tracing instead. Pick
Copilot Studio only if your agents must live inside Microsoft 365.