The best AI agent builder depends on who's building: n8n if a technical team wants visual workflows with agents and the option to self-host, Zapier Agents if non-developers want an agent that acts across their existing apps, Microsoft Copilot Studio if you live in Microsoft 365, and Lindy if you want an AI teammate in Slack rather than a canvas. Developers who want code should start with LangGraph, the OpenAI Agents SDK, the Claude Agent SDK or CrewAI, and Google Cloud teams with Gemini Enterprise Agent Platform. One name you may expect is missing: OpenAI is shutting down its Agent Builder on November 30, 2026.
"AI agent builder" covers two products. One is a no-code or low-code tool where you describe or draw the agent and the vendor runs it. The other is a code framework where you write the agent and host it yourself. Pick the category first.
How we compared
We grouped tools by who builds the agent, then compared how each one bills (per execution, credit, seat or model call), where the agent runs, and whether you can self-host or switch models. Prices were checked on each vendor's own pricing page and docs on October 4, 2026, and are stated as of October 2026 with the billing basis the vendor gives. Code frameworks are free libraries; the cost is the model usage.
At a glance
| Tool | Type | Best for | Pricing (as of October 2026) | Notable limit |
|---|---|---|---|---|
| n8n | Low-code workflows | Technical teams, self-hosting | Cloud from €20/month billed annually (2,500 executions) | Prices shown in euros |
| Zapier Agents | No-code | Agents acting across business apps | Free 400 activities/month; Pro $50/month ($33.33 billed annually) | FAQ still refers to a beta |
| Microsoft Copilot Studio | Low-code | Microsoft 365 organisations | 25,000-credit packs at $200/month, or pay as you go | External agents need a Copilot Studio plan |
| Gemini Enterprise Agent Platform | Developer platform | Google Cloud teams | Pay for usage; up to $300 credits for new customers | Built for developers |
| Lindy | AI teammate | Delegating work from Slack | Free $50 credits for 7 days; Team from $29.99 per active user a month | Every active user is billed |
| LangGraph | Code framework | Controlled, stateful agents | MIT library; LangSmith Plus $39/seat a month | Hosting and tracing are separate |
| OpenAI Agents SDK | Code framework | Lightweight agents on OpenAI | MIT library; model usage at API rates | Few built-in services |
| Claude Agent SDK | Code framework | Agents that work on files and commands | Library; Claude API usage | Anthropic's Commercial Terms apply |
| CrewAI | Code framework plus platform | Multi-agent crews | MIT framework; platform free for 50 executions a month | Enterprise pricing is custom |
The 9 best AI agent builders
1. n8n – visual workflows with agents for technical teams
n8n is a workflow automation tool where an AI agent is one node among many, so you can mix deterministic steps (fetch, filter, branch) with model calls. Every plan includes agents, MCP and AI nodes, unlimited users and workflows, and all integrations; you pay by workflow executions, not steps. On our check, prices were shown in euros: Starter from €20 a month billed annually for 2,500 executions, and Pro from €50 a month billed annually for 10,000. A self-hosted Community Edition is on GitHub under n8n's Sustainable Use License.
Strengths: deterministic steps and AI in one workflow, and self-hosting keeps data on your own servers. Limits: n8n's licence page says it doesn't call itself open source, and the licence covers internal business, non-commercial or personal use, so embedding n8n in a product you sell needs an agreement with n8n. Our Zapier alternatives guide compares n8n with the other workflow tools.
2. Zapier Agents – no-code agents across your existing apps
Zapier Agents lets you describe an agent, connect it to company knowledge, and have it take actions in the apps Zapier connects to (9,000+, by Zapier's count), on command or automatically. On the pricing page, Agents Free covers 400 activities a month and Agents Pro is $50 a month billed monthly or $33.33 a month billed annually ($400 a year) for 1,500 activities. An activity is an action the agent takes, a web browse or a knowledge lookup.
Strengths: a short path for non-developers whose work already runs through Zapier-connected apps. Limits: activity caps are small for busy agents, and the Agents FAQ still has a section on support "during the Beta".
3. Microsoft Copilot Studio – agents for Microsoft 365 organisations
Copilot Studio builds agents that work across Microsoft 365 and can be published to websites and other channels. Microsoft 365 Copilot ($30.00 per user a month, paid yearly) includes building agents for use inside your organisation. Publishing to outside channels needs a standalone Copilot Studio plan: a tenant-wide licence with capacity packs of 25,000 Copilot Credits at $200.00 a pack per month, a pre-purchase plan, or a pay-as-you-go meter that requires an Azure subscription.
Strengths: if you already pay for Microsoft 365 Copilot, building internal agents is included. Limits: credit consumption varies by what the agent does, so budgeting means testing real usage first.
4. Gemini Enterprise Agent Platform – Google Cloud's developer platform
Google's agent offering is now Gemini Enterprise Agent Platform, formerly Vertex AI. Google describes it as a platform for developers to build, scale, govern and optimise agents, with Agent Studio for designing and testing prompts for Gemini models and the open-source Agent Development Kit (Apache-2.0 on GitHub) for code. Pricing is pay for what you use, and new customers get up to $300 in free credits.
Strengths: fits teams already building on Google Cloud and Gemini models. Limits: this is a developer platform, not a drag-and-drop tool for business users, and usage pricing needs a cost estimate before you scale.
5. Lindy – an AI teammate in Slack
Lindy describes itself as an AI teammate that lives in your company's Slack and connects to your tools. You give it work, set scheduled routines and build your own skills, and you can choose the model per task. Free gives $50 in credits that last 7 days. Team starts at $29.99 a month for 3,000 credits, every active user in the workspace is billed, and top-ups are $10 per 1,000 credits.
Strengths: no workflow to design; you delegate in plain language where your team already talks. Limits: when the shared credit pool runs out, Lindy pauses credit-using work until the next cycle, and per-user billing adds up in larger teams.
6. LangGraph – low-level control for developers
LangGraph is LangChain's agent runtime and orchestration framework, MIT-licensed on GitHub. You design the control flow (single, multi-agent or hierarchical), with built-in memory, human-in-the-loop checks and token-by-token streaming, and it works with any model provider. Tracing, evaluation and hosting are sold separately as LangSmith: Developer is $0 for one seat with 5,000 base traces a month, and Plus is $39 per seat a month, billed monthly, with 10,000 traces and one free small serverless deployment.
Strengths: explicit control flow, which matters when an agent must follow a process rather than improvise. Limits: more code and concepts than the SDKs below; you'll want LangSmith or your own tracing for production.
7. OpenAI Agents SDK – a lightweight framework from OpenAI
The OpenAI Agents SDK is a small, MIT-licensed library for Python (and JavaScript) built on three primitives: agents with instructions and tools, handoffs between agents, and guardrails. It includes tracing and sessions, and third-party adapters let it call other model providers. The library is free; you pay for the model calls. OpenAI also documents a hosted Agents API billed at model and tool rates.
Strengths: few abstractions, fast to learn. Limits: it's a library, so hosting, storage and UI are yours to build. If you started in OpenAI's visual Agent Builder, its deprecation notice sets shutdown for November 30, 2026 and points to the Agents SDK or ChatGPT Workspace Agents.
8. Claude Agent SDK – agents that work on files and commands
The Claude Agent SDK gives you the agent loop, tools and context management behind Claude Code as a Python or TypeScript library: built-in tools to read and edit files, run commands and search the web, plus hooks, subagents, MCP connections, permissions and sessions. You authenticate with a Claude API key, and its use is governed by Anthropic's Commercial Terms of Service, including in products you ship to customers.
Strengths: built-in tools for agents that act on a codebase, documents or a sandbox, not only chat. Limits: it's built around Claude models, and you host and secure the process yourself (Anthropic documents a hosted Managed Agents option separately).
9. CrewAI – multi-agent crews, open source plus a platform
CrewAI is an MIT-licensed Python framework on GitHub for teams of agents that collaborate on a task. The hosted platform's Basic plan is free with a visual editor, an AI copilot, GitHub integration and 50 workflow executions a month; Enterprise is custom-priced and can run on CrewAI's cloud or your own infrastructure.
Strengths: the multi-agent pattern is the default, not an add-on. Limits: 50 free executions is a trial, not a production allowance, and anything bigger means a sales conversation.
Got an idea? Build it now!
Just start with a simple prompt. No coding required: Greta.sh turns your idea into a working app in minutes.
Sales-led platforms worth knowing
Two well-known builders don't publish self-serve prices. Relevance AI lists only an Enterprise plan and asks you to talk to sales. Voiceflow, which focuses on voice and chat agents for customer experience, offers agencies a free trial with usage-based billing and asks businesses to request pricing. Budget time for a sales cycle.
How to choose
- Who maintains it? If nobody on the team writes code, choose Zapier Agents, Lindy or Copilot Studio. If someone does, n8n gives you a visual tool with an escape hatch into code.
- How much should the agent decide? For a process with fixed steps, use a workflow tool (n8n) or a graph (LangGraph). For open-ended tasks, the SDKs and Lindy give the model more room.
- Where does the data live? Self-hosting (n8n Community, the open-source frameworks) keeps it with you; hosted tools keep it with the vendor.
- How will it connect to your tools? n8n, Lindy, CrewAI and both SDKs mention MCP support; our MCP explainer covers what that buys you.
The part an agent builder doesn't build
An agent is rarely the whole product. Users need to sign in, see what the agent did, approve its actions and fix its mistakes, and all of that needs an app with a real database. If you're building that app around an agent, Greta.sh can generate it: describe the app and it builds the frontend, backend, database and authentication, with clean Next.js code you can export. It's an app builder, not an agent framework, so the agent itself still comes from one of the tools above. Pricing: Free is 5 credits a day; Starter is $5/month on the welcome offer (normally $20/month) and adds deployment; the Database feature is on Pro, $20/month on the welcome offer (normally $50/month). If you're choosing the backend yourself instead, see our guides to Supabase alternatives and Firebase alternatives.
Got an idea? Build it now!
Just start with a simple prompt. No coding required: Greta.sh turns your idea into a working app in minutes.
Frequently asked questions
What is the best AI agent builder?
For non-developers, Zapier Agents or Lindy; for Microsoft 365 organisations, Copilot Studio; for technical teams that want visual workflows and self-hosting, n8n. Developers should start with LangGraph, the OpenAI Agents SDK, the Claude Agent SDK or CrewAI.
Is there a free AI agent builder?
Yes, with limits. Zapier Agents Free includes 400 activities a month, CrewAI's Basic plan includes 50 workflow executions a month, and LangGraph, CrewAI's framework and the OpenAI Agents SDK are MIT-licensed libraries; you still pay for model usage. n8n's Community Edition can be self-hosted.
What happened to OpenAI's Agent Builder?
OpenAI announced its deprecation on June 3, 2026, and it's scheduled to shut down on November 30, 2026. OpenAI points existing users to the Agents SDK or ChatGPT Workspace Agents, and ChatKit remains available.
What is Google's AI agent builder called now?
Google Cloud's agent offering is Gemini Enterprise Agent Platform, formerly Vertex AI. It includes Agent Studio and works with the open-source Agent Development Kit.
Do I need to code to build an AI agent?
No. Zapier Agents, Lindy and Copilot Studio are built for non-developers. Code frameworks give more control over how the agent behaves and where it runs, at the cost of hosting and maintaining it yourself.
Start with the job, then the tool
Write down the one task the agent should finish and who fixes it when it fails. That picks the category, and the category narrows nine tools to two or three.
