The question no pricing page answers
Every comparison in this category ranks tools by price tier, feature count, and ease of setup. None of them answer the question that decides whether a tool is legally usable for the buyer who intends to build a business on top of it: what does the licence permit you to do for other people?
That question reorders this list significantly. The tools that rank highest on features and star counts often carry the strictest restrictions on commercial multi-tenant operation — and the restriction lives in a file nobody reads on the vendor's pricing page.
One of the entries below — Cortex — is a visual AI agent builder sold once as source code. You deploy it under your own brand, wire agents on a node canvas with a live execution trace, and run it as a multi-tenant service for paying clients, with no licence restriction on doing so. This page groups the alternatives by what their licence actually permits: from platforms you may only rent, through tools that explicitly forbid serving third parties, to codebases that are more permissive than Cortex on that exact question.
Five questions worth answering before you choose
1. What does the licence let you do with it? The organising question in this guide, and the one most buyers discover too late. "Self-hosted" and "open source" do not mean you can serve it to paying clients as a platform. Read the actual licence file — the n8n Sustainable Use License is one example of why; the Dify multi-tenant restriction is another.
2. Do you own a codebase or a subscription? Every proprietary SaaS in this category can change its pricing, deprecate a feature, or shut down. The AI agent builder category has demonstrated this concretely: two of the most-used visual builders began winding down in the same quarter. A codebase you own cannot be deprecated out from under you.
3. Does the tool ship the commercial chassis, or do you build it? A permissive OSS canvas gives you the canvas. Multi-tenant organization management, RBAC, plan-gated billing, an admin panel, message metering, and analytics are the product-to-sell wrapper that turns a canvas into something you can charge for. Whether you want that included or prefer to assemble it yourself is a real decision with real cost.
4. Is the interface visual or code-first? This category divides almost cleanly between tools a non-technical person can use and tools that require a developer writing Python or TypeScript. The code-first frameworks — LangGraph, CrewAI, Microsoft Agent Framework — are out of scope in this guide because they serve a different audience entirely.
5. Who maintains it? A self-hosted tool without an active upstream maintainer places the security maintenance burden on you. That question has become concrete in this category — one major player has archived its repository, leaving future patch responsibility to the community or to you.
The alternatives, by what their licence permits
Rent-only — proprietary SaaS, no source licence at any tier
You build on the vendor's infrastructure under the vendor's terms. No source code is available. If the vendor changes pricing, removes a feature, or shuts down, your deployment changes with it.
Voiceflow has no self-host path and no source licence available at any tier. Its current positioning is an Agentic Context Engine for enterprise CX teams, with agent playbooks, Tools, Functions, and Guidelines primitives. Pricing is no longer published on the vendor's own page — it now shows "Request pricing" only, as of 2026-08-07 (fetched, voiceflow.com/pricing). Honest limit: Voiceflow has a large enterprise customer base, substantial community, and years of production deployment. For buyers who need enterprise CX infrastructure without any setup, it is a credible option. It cannot be self-hosted, rebranded, or operated as a white-label platform for other businesses under any tier.
Botpress — today's generation — is a proprietary SaaS. The AGPL dual-licence that appears in older comparisons applied to the archived v12 repository, which is a separate, unmaintained product. The current Botpress Cloud pricing: free Pay-As-You-Go; $89/month Plus; $495/month Team; $1,495/month Managed — with AI Spend quota bundled into every plan as of May 2026 (Botpress blog, botpress.com, May 2026). Honest limit: Botpress ships the category's strongest enterprise channel integrations — WhatsApp Business API, voice, telephony, and SMS. Those are channels Cortex does not support in v1. For an operator building into enterprise customer-support environments where those channels are the requirement, Botpress Cloud is the better-matched tool.
Stack AI is proprietary SaaS. On-premise deployment is available on Enterprise — a commercial contract, not a source licence. Its pricing has a notable gap: Free (500 runs/month, 2 projects) and Enterprise custom only — no self-serve paid middle tier (stackai.com/pricing, fetched 2026-08-07). Honest limit: Stack AI's builder is oriented toward data pipeline and enterprise workflow use cases — connecting data sources, running AI over structured data. For operators building user-facing conversational agents, it is a mismatch in product orientation.
Relevance AI is proprietary SaaS. Plans: Free, $19/month Professional (annual), $234/month Team (annual), Enterprise custom (coldiq.com, Lindy.ai, 2026). Honest limit: Relevance AI is designed around autonomous AI workforce agents — background workers that complete tasks without user interaction — rather than a user-facing chat canvas. The fit is internal automation, not a branded conversational product you can sell.
Fair-code — self-hostable, but restricted to your own internal use
n8n is distributed under the Sustainable Use License — fair-code, not OSI-approved open source (github.com/n8n-io/n8n/blob/master/LICENSE.md, fetched 2026-08-07). The operative restriction: "You may use or modify the software only for your own internal business purposes or for non-commercial or personal use." Hosting n8n as a paid service, white-labelling it for clients, or embedding it in a commercial product all require a separate commercial or embed licence from n8n. One material ambiguity: the phrase "internal business purposes" is used throughout the licence file but is never defined in it — that is the term the entire restriction turns on, and its meaning is not resolved by the text alone. n8n Cloud pricing: €20–€667/month annual, Starter to Business, as of 2026-08-07 (n8n.io/pricing, fetched at primary source). Honest limit: n8n is a workflow automation and service-integration engine — it excels at connecting APIs, automating data pipelines, and triggering actions across services. For a buyer looking for a conversational AI agent canvas to hand to a non-technical client, n8n is the wrong category of tool and the wrong licence for the commercial-platform use case.
Restricted OSS — open source with an explicit multi-tenant prohibition
Dify is distributed under Apache 2.0 with two additional conditions (github.com/langgenius/dify/blob/main/LICENSE, fetched 2026). The first prohibits removing or altering the Dify logo or copyright from the console. The second: "you may not use the Dify source code to operate a multi-tenant environment" — meaning separate workspace data and configuration per customer — without explicit written authorization (available via commercial licence, business@dify.ai). Dify Cloud: free Sandbox (200 message credits total, not per month), $59/workspace/month Professional, $159/workspace/month Team; Enterprise is annual and custom only (comparedge.com/tools/dify-ai/pricing, checkthat.ai/brands/dify/pricing, 2026). Self-hosted Community Edition: free; the multi-tenant restriction applies only when operating separate workspaces for different paying customers. Honest limit: Dify has the most mature plugin ecosystem among the self-hostable tools in this guide, the most production-hardened RAG pipeline, and substantially more deployment history than v1 Cortex. For a buyer building a single internal workspace — one team, their own agents — Dify self-hosted is the most capable free option here, and the licence restriction does not apply. The restriction kicks in when the buyer intends to serve multiple paying customers with separated data.
Permissive OSS — you can do almost anything
Langflow is distributed under the MIT Licence (github.com/langflow-ai/langflow/blob/main/LICENSE, Copyright 2024 Langflow, fetched 2026). MIT places no restriction on commercial multi-tenant use, resale, redistribution, or modification. The only obligations are retaining the copyright notice and licence text in any distribution. Langflow is IBM-backed and actively maintained. DataStax's hosted Langflow platform was retired in April 2026; the current primary deployment path is self-hosting via Docker or Kubernetes. Honest limit: on the licence question this guide organised around, Langflow wins — it is more permissive than Cortex. A buyer for whom licence freedom is the whole decision should pick Langflow. What Langflow does not include is the commercial SaaS chassis: multi-tenant organization management with data isolation, RBAC, plan-gated billing, an admin panel, message metering, conversation analytics, and agent versioning. Langflow gives you a capable agent canvas and a RAG pipeline; building the product-to-sell around it is engineering work that comes on top.
Flowise was distributed under Apache 2.0 for the community edition, with enterprise SSO and RBAC packages under a commercial licence (github.com/FlowiseAI/Flowise/blob/main/LICENSE.md, fetched 2026). Apache 2.0 — more permissive than n8n or Dify for the operator use case — permits commercial multi-tenant use, modification, and redistribution with attribution. However, the Flowise GitHub repository archived in August 2026, following the project's acquisition by Workday in August 2025 (Workday newsroom, 2025-08-14). Code freeze was July 29; core-team support ended August 31. The Apache 2.0 grant survives archiving — Flowise's own shutdown notice states the code "will still remain on Github and the Apache 2.0 licensed code is yours to keep," and it encourages community forks (flowiseai.com/sunset). What it does not have is upstream maintenance: no new features, no security patches, no dependency updates from the original team. An operator building on Flowise today owns the security maintenance burden from day one. Honest limit: Flowise was the closest architectural match to Cortex in the open-source space — a visual node canvas, self-hostable, RAG knowledge base, embeddable widget, approximately 55,000 GitHub stars at archival. It is not gone; it is current. Whether a community fork sustains the project is an open question.
Commercial own-the-code — packaged for operators who want to sell it
Cortex — $179, one-time purchase
Cortex is a visual AI agent builder sold once as source code, with no licence restriction on commercial multi-tenant use — deploy it under your own domain and brand, and charge your own clients. Your clients see only your platform. The node canvas has six types: Brain (LLM model, system prompt, temperature), Knowledge (RAG retriever; documents chunk and embed for semantic search at inference time), Tools (HTTP requests, database queries, webhook fires — no code sandbox), Decision (confidence-based routing: set a confidence threshold and the agent branches when its score falls below it), Guardrails (content filter, PII masking, response validation, max-turns escalation — visible canvas nodes, not buried config), and Output (text response, human-handoff with context summary, or end-conversation). While you test in the live-preview panel, the node-execution trace streams in real time: which node fired, what it received, and what it returned. Finished agents deploy as an embeddable vanilla JS widget (Shadow DOM, under 30 KB gzipped, with allowed-domain enforcement) or a public shareable link. The package includes the commercial SaaS chassis: multi-tenant organizations with data isolation, RBAC enforced at schema level, plan-gated billing, admin panel, message metering with 80%/100% throttle thresholds, conversation analytics, and agent versioning with snapshot rollback.
Honest limit: no code execution sandbox — there is no JavaScript or Python node, and this is a deliberate architectural constraint that keeps the product single-service and deployable on standard serverless infrastructure. Autonomous multi-agent orchestration where one agent spawns and directs others is out of scope in v1. No voice, telephony, or WhatsApp channel integrations. No custom node types beyond the six built-in categories. No SSO. These are scope boundaries, not gaps.
Try the live demo · Get Cortex
Boundary — code-first frameworks, licences not verified for this guide
LangGraph, CrewAI, and Microsoft Agent Framework — which unified AutoGen and Semantic Kernel at its 1.0 release in April 2026 — are a different audience category: code-first tools for engineers writing Python or TypeScript, not visual builders for operators who need a GUI surface to hand to a non-technical client. Their licence terms were not checked at primary source during the research for this guide and therefore cannot be placed in any licence group. Note that AutoGen as a standalone project is effectively superseded by the Agent Framework and should not be listed as a current option. Where the licence terms of any of these tools matter for your use case, verify at their respective LICENSE files before building.
How to choose
| If the goal is... | The right pick |
|---|---|
| An agent platform on someone else's infrastructure, zero setup | Voiceflow or Botpress Cloud |
| Enterprise channel integrations — WhatsApp, voice, telephony | Botpress Cloud |
| A free, mature agent builder for a single internal team | Dify Community Edition |
| Workflow automation and service integrations for your own internal operations | n8n Community Edition (Sustainable Use — internal use only) |
| Maximum licence freedom — build your own commercial chassis around it | Langflow (MIT) |
| A permissive-licence OSS canvas to fork and maintain yourself | Flowise community code (Apache 2.0; archived; security maintenance is yours) |
| A packaged agent builder you operate for paying clients, commercial SaaS chassis included | Cortex |
Honest verdict
For most buyers this guide reaches, the correct answer is not Cortex. The buyer who needs to deploy AI agents for their own organisation — one team, one workspace, their own agents — will find Dify's Community Edition more mature and more capable at no cost. Voiceflow and Botpress Cloud are usable without any infrastructure work. Langflow is MIT-licensed, actively maintained, and more permissive than Cortex on the question this entire guide organised around. If permission freedom is the deciding factor, Langflow is the honest winner.
Cortex earns its place in this guide at a specific intersection: the operator who needs a packaged, multi-tenant agent builder they can run for paying clients — the commercial SaaS chassis included, deployed under their own brand, with no licence restriction on serving those clients from day one. Among the options above, that position is distinct. Langflow is closer on licence freedom; Dify is closer on feature depth and production maturity; neither ships as a sellable product without building the wrapper yourself.
If you are building the platform other people use to deploy AI agents, this guide ends at Tier 5. If you need AI agents for your own team, it begins at Tier 3.
Common questions
Which AI agent builder licences explicitly restrict commercial multi-tenant use?
Two of the self-hostable tools in this guide carry explicit restrictions. Dify adds two conditions on top of Apache 2.0: one prohibiting removal of the Dify logo from the console, and one stating you may not use the source code to "operate a multi-tenant environment" without written authorization. n8n's Sustainable Use License restricts use to "internal business purposes" and explicitly forbids hosting it as a paid service or white-labelling without a commercial licence. Cortex, Langflow, and the Flowise community edition carry no equivalent restriction (github.com/langgenius/dify/blob/main/LICENSE; github.com/n8n-io/n8n/blob/master/LICENSE.md; both fetched 2026).
Is Langflow free to use as a commercial multi-tenant platform?
Yes. Langflow is MIT-licensed — no restriction on commercial use, multi-tenant operation, modification, or redistribution. The only obligations are retaining the copyright notice and licence text. It is IBM-backed and actively maintained. The honest counterpoint: Langflow is more permissive than Cortex on licence, but does not include the commercial SaaS chassis. Building the organization management, RBAC, billing, and admin tooling around a Langflow canvas is substantial engineering work on top of the canvas itself. For a team that wants the chassis included, Cortex is the packaged alternative (github.com/langflow-ai/langflow/blob/main/LICENSE, fetched 2026).
What happened to Flowise, and is the code still usable?
Flowise archived its GitHub repository in August 2026, roughly one year after its acquisition by Workday (Workday newsroom, 2025-08-14). Code freeze: July 29. Core-team support ended August 31. The Apache 2.0 licence survives archiving — Flowise's own shutdown notice states the code "will still remain on Github and the Apache 2.0 licensed code is yours to keep," and encourages community forks (flowiseai.com/sunset). The code is usable. What no longer exists is upstream maintenance: no new features, no security patches, no dependency updates from the original team. An operator building on Flowise today accepts full responsibility for all future security work.
What is the difference between an AI agent builder and an AI agent framework?
Agent frameworks — LangGraph, CrewAI, Microsoft Agent Framework — are code-first tools for engineers writing Python or TypeScript; flexible, built for complex agent topologies, designed for developers who work in a terminal. Agent builders — Voiceflow, Botpress, Dify, Langflow, Cortex — provide a visual canvas that a non-technical person can use or that an operator can hand to a client. Cortex sits in the builder category and adds the commercial packaging layer: multi-tenant infrastructure that turns a canvas into a product the operator can sell. If maximum technical flexibility for an engineering team is the requirement, the frameworks win. If a visual surface deployable as a branded product is the requirement, builders are the fit.
Can n8n be white-labelled and sold to clients as a managed service?
Not under the Community Edition's Sustainable Use License. The operative language: "You may use or modify the software only for your own internal business purposes or for non-commercial or personal use." White-labelling it, hosting it as a paid service for clients, or embedding it in a commercial product all require a commercial or embed licence from n8n GmbH. A material detail: the phrase "internal business purposes" is central to the entire restriction but is never defined in the licence file itself — what counts as "internal" is not resolved by the text. n8n does offer commercial licences; contact them directly for terms. n8n Cloud: €20–€667/month annual, as of 2026-08-07 (n8n.io/pricing, fetched at primary source).
What does Cortex not include in v1, and why does it matter before buying?
Cortex's six-node canvas and commercial SaaS chassis ship complete in v1. No code execution sandbox — no JavaScript or Python node, a deliberate constraint for single-service deployment. No autonomous multi-agent orchestration (one agent spawning others) in v1. No voice, telephony, or WhatsApp integrations. No custom node types beyond the six built-in categories. No SSO. For operators whose agents need to run arbitrary code, handle voice calls, or spawn child agents, Langflow (permissive, extensible) or Botpress Cloud (enterprise channels) are better fits. Cortex is the packaged choice for operators who need a multi-tenant agent builder serving paying clients from day one.