The best AI agent builders, tested and ranked
We built the same support, lead and data-lookup agents in eight AI agent builders and ran each one 20 times. Here's which builder got the most runs right, which fits your team and what each costs to start.
About this list
An AI agent builder lets you put together an AI agent without writing everything from scratch. You connect a language model, your documents and the tools the agent may use, set the logic on a visual canvas, then publish it as a chat widget, phone line, API or workflow. Unlike a simple chatbot, an agent can take actions, such as updating a record or handing a customer to a person. You'll find more in our AI agent builders section and our AI chat hub.
These builders are aimed at different people. Voiceflow, Botpress and GPT-trainer centre on customer support. Dify and Langflow are open-source tools for technical teams that you can host yourself. Dust is a shared workspace for internal agents on company data, while Stack AI and GPTBots.ai lean towards larger companies with security needs. Pricing differs too: per seat, per conversation, in usage credits, or free software.
To compare them fairly, we built the same three agents in every tool between October 17 and October 23, 2026, and ran each one 20 times with the same inputs. We counted correct runs, timed each build, checked what happened when things went wrong and counted the connections that worked without code.
Compare at a glance
| # | Tool | Best for | Cheapest way to start | Score |
|---|---|---|---|---|
| 1 | Dust | Best overall | 30 EUR a month (Pro) | 3.0 |
| 2 | Dify | Best open-source all-rounder | $49.17 a month, billed yearly (Professional) | 2.5 |
| 3 | Langflow | Best free developer toolkit | Free | 2.5 |
| 4 | Voiceflow | Best for voice and support agents | $60 a month (Pro) | 2.5 |
| 5 | GPT-trainer | Best for multichannel support inboxes | $150 a month (Self-Serve) | 2.0 |
| 6 | Stack AI | Best for enterprise governance | Free | 2.0 |
| 7 | Botpress | Best free start for support agents | $150 a month, billed yearly (Plus) | 2.0 |
| 8 | GPTBots.ai | Best for certified no-code agents | Free | 2.0 |
Features and terms compared
These facts come from each company's own pages and our research, with prices checked on November 2 and 3, 2026.
| Tool | Free plan | Cheapest paid plan | Open source or self-hosting | Security certifications | Voice agents |
|---|---|---|---|---|---|
| Dust | Free seat, 500 lifetime credits | Pro: 30 EUR a seat a month, excl. VAT | No; single-tenant on Enterprise | SOC 2 Type II | Not stated |
| Dify | Sandbox: 200 message credits in total | $49.17 a month, billed yearly | Yes, free Community edition | SOC 2 Type II, ISO 27001:2022, GDPR | Not built around phone |
| Langflow | Free software (MIT) | No paid plan; IBM support by quote | Yes, MIT; self-host or desktop app | None listed | Not built around phone |
| Voiceflow | 1 editor, not for production | Pro: $60 a month | No, hosted only | SOC 2 Type II, ISO 27001:2022, GDPR | Yes, a core channel |
| GPT-trainer | Free start (limits not stated); 14-day trial | Self-Serve: $150 a month | On-premise on custom Enterprise | SOC 2, ISO 27001, GDPR (stated) | Yes, inbound and outbound |
| Stack AI | 500 runs a month | No public price (Enterprise quote) | On-premise and VPC on Enterprise | SOC 2 Type II, ISO 27001, HIPAA, GDPR | Not built around phone |
| Botpress | 100 conversations a month | Plus: $150 a month, billed yearly | Not stated | GDPR claimed; SOC 2 unclear | Enterprise only |
| GPTBots.ai | 100 credits a month | Business: no public price | Private deployment on Enterprise | SOC 2 Type II, ISO 27001, ISO 27701, ESOF | Not stated |
Test results side by side
No builder came close to perfect. Dust did best, with 42 of 60 correct runs, 14 of 20 on each agent. Every other builder landed between 35 and 38, so roughly one run in three went wrong even for the better tools. If an agent will talk to customers or touch your records, plan for someone to check its work.
Setup took a while everywhere: Langflow was quickest at about 114 minutes per agent and GPT-trainer slowest at about 163. Dust and GPT-trainer handled failures best, with alerts and hand-off in 8 of 12 failure checks, while Stack AI managed 6. The surprise was Stack AI, second on correct runs with 38 of 60 but sixth overall because of slow setup, weak failure handling and no paid plan you can simply buy.
Every tool got the same test, between October 17, 2026 to October 23, 2026.
| Tool | Score | Runs completed correctly out of 20 | Time to build each workflow from sign-up | How failures are handled (retries, alerts, human hand-off) | Cost per 100 runs at list prices | Integrations available for the workflow without code |
|---|---|---|---|---|---|---|
| Dust | 3/5 | Correct runs 14/20; 14/20; 14/20 | 130 min/agent; 3 agents | Alerts 8/12; hand-off 8/12 | – | 6/10 required connections no-code |
| Dify | 2.5/5 | Correct runs 13/20; 12/20; 12/20 | 137 min/agent; 3 agents | Alerts 7/12; hand-off 7/12 | – | 7/10 required connections no-code |
| Langflow | 2.5/5 | Correct runs 12/20; 12/20; 13/20 | 114 min/agent; 3 agents | Alerts 6/12; hand-off 7/12 | – | 6/10 required connections no-code |
| Voiceflow | 2.5/5 | Correct runs 13/20; 12/20; 12/20 | 145 min/agent; 3 agents | Alerts 7/12; hand-off 8/12 | – | 7/10 required connections no-code |
| GPT-trainer | 2/5 | Correct runs 12/20; 13/20; 12/20 | 163 min/agent; 3 agents | Alerts 8/12; hand-off 8/12 | – | 6/10 required connections no-code |
| Stack AI | 2/5 | Correct runs 13/20; 13/20; 12/20 | 151 min/agent; 3 agents | Alerts 6/12; hand-off 6/12 | – | 6/10 required connections no-code |
| Botpress | 2/5 | Correct runs 12/20; 12/20; 12/20 | 146 min/agent; 3 agents | Alerts 7/12; hand-off 7/12 | – | 5/10 required connections no-code |
| GPTBots.ai | 2/5 | Correct runs 12/20; 12/20; 11/20 | 139 min/agent; 3 agents | Alerts 7/12; hand-off 6/12 | – | 6/10 required connections no-code |
Dust
Best overallDust came first of eight with 3 out of 5, the only builder to reach 3. Each of its agents got 14 of 20 runs right, 42 of 60 in total, the best in our test. Builds took about 130 minutes per agent, alerts and hand-off worked in 8 of 12 failure checks, and 6 of the 10 connections we needed worked without code, which is why integrations scored 2.5.
Dust is a hosted workspace where a company connects tools such as Slack, Notion, Google Drive and Salesforce, then builds agents on that data with a choice of more than 20 models. It lists SOC 2 Type II with US or EU data residency. Prices are in euros excluding VAT, and unused credits don't roll over. On G2 it has 4.9 out of 5 from 22 reviews (checked November 2026).
Who it's for
Business teams that want shared, governed agents on company data without writing code.
Skip it if
You want open source or self-hosting. Dify and Langflow can run on your own servers for free.
What we liked
- Most correct runs: 42 of 60
- Alerts and hand-off in 8 of 12 failure checks
- Free seat and more than 20 models
What we didn't
- About three runs in ten still went wrong
- Only 6 of 10 connections without code
- Euro prices; credits don't roll over
Key features
- Custom agents with company knowledge and tools
- 70+ connectors and remote MCP servers
- Multi-agent workflows and triggers on Business
- SCIM, audit logs and single-tenant on Enterprise
Plans and prices
| Free 500 lifetime credits. | Free |
| Pro 8,000 credits per seat a month. | 30 EUR a month, or 24 EUR a month billed yearly |
| Max 40,000 credits per seat a month. | 150 EUR a month, or 120 EUR a month billed yearly |
| Enterprise Pooled credits; SCIM, audit logs, custom retention. | Custom contact sales |
Dify
Best open-source all-rounderDify came second with 2.5 out of 5. Its agents got 37 of 60 runs right, each took about 137 minutes to build, and alerts and hand-off both worked in 7 of 12 failure checks. It stood out on connections: 7 of the 10 we needed worked without code, earning a joint-best 3 for integrations. Value for money was its weakest part at 2. Read our Dify review.
Dify is an open-source workspace for AI apps, agents and RAG pipelines, with about 160,000 GitHub stars. You can use its cloud, host the free Community edition yourself or buy Enterprise, and publish apps as pages, APIs, embeds or MCP tools. Cloud plans are only shown billed yearly, the Sandbox has just 200 message credits in total, and HIPAA isn't confirmed. On G2 it has 4.1 out of 5 from 20 reviews (checked November 2026).
Who it's for
Technical teams that want an open-source builder they can self-host, especially for apps over their own documents.
Skip it if
You want a ready-made support bot without technical help. Dust is easier to start with and got more runs right.
What we liked
- 7 of 10 connections worked without code
- Free to self-host with the Community edition
- Publishes as app, API, embed or MCP tool
What we didn't
- Only 37 of 60 runs correct
- About 137 minutes to build each agent
- Cloud plans shown billed yearly only
Key features
- Workflow Studio with testing and debugging
- Knowledge Pipeline for ingesting and indexing documents
- Marketplace for models, tools and data sources
- SSO, RBAC and audit logs on Enterprise
Plans and prices
| Sandbox 200 message credits total; 5 apps. | Free |
| Professional 5,000 message credits a month; 3 members. | $49.17 a month, billed yearly |
| Team 10,000 message credits a month; 50 members. | $132.50 a month, billed yearly |
| Enterprise (self-hosted) SSO; Multiple workspaces. | Custom contact sales |
Langflow
Best free developer toolkitLangflow came third with 2.5 out of 5. It was the quickest to set up at about 114 minutes per agent, earning a 3 for ease of setup. Its agents got 37 of 60 runs right, alerts fired in 6 of 12 failure checks and hand-off worked in 7, and 6 of 10 connections worked without code. Flexibility was its weakest part at 2. Our Langflow review covers the full test.
Langflow is free software under the MIT license, backed by IBM through DataStax. Developers drag components onto a canvas, edit them in Python and deploy flows as an API or MCP server, on their own server or in the free desktop app. The hosted Langflow Cloud was reported discontinued in April 2026, so you handle hosting and security yourself, and it lists no certifications.
Who it's for
Python developers who want a free agent builder they can host and customise, with no per-seat fees.
Skip it if
You want someone else to host and support it. Dust is a managed option that did better in our test.
What we liked
- Quickest to set up at about 114 minutes
- Free, open-source software under MIT
- Components can be edited in Python
What we didn't
- Only 37 of 60 runs correct
- Alerts in only 6 of 12 failure checks
- No managed hosting or certifications
Key features
- Visual drag-and-drop flows
- Python customisation of components
- Deploy flows as an API or MCP server
- Works with many models and vector stores
Plans and prices
| Open source (self-hosted) All core features; Desktop app. | Free |
| Enterprise via IBM Professional services; Premier support. | Custom contact sales |
Voiceflow
Best for voice and support agentsVoiceflow came fourth with 2.5 out of 5. Its agents got 37 of 60 runs right and each took about 145 minutes to build. Alerts fired in 7 of 12 failure checks and hand-off to a person worked in 8. Like Dify, it had 7 of 10 connections working without code, but value for money scored 2. Read our Voiceflow review.
Voiceflow is a shared builder for chat and phone agents focused on customer support, and voice is a core channel. You can pick providers such as Anthropic or Google, or its own Voiceflow Core model, with fallback during outages. It lists SOC 2 Type II, ISO 27001:2022 and GDPR. Plans include usage credits, which users find confusing, and the free plan isn't for production. On G2 it has 4.6 out of 5 from 110 reviews (checked November 2026).
Who it's for
Support teams and agencies that need chat and phone agents with strong security paperwork.
Skip it if
Reliability matters most. Dust got more runs right, and Botpress's free plan can serve real conversations.
What we liked
- Phone agents are a core feature
- 7 of 10 connections worked without code
- Hand-off in 8 of 12 failure checks
What we didn't
- Only 37 of 60 runs correct
- About 145 minutes to build each agent
- Free plan isn't for production
Key features
- Shared visual agent canvas
- Playbooks and workflows
- Live agent hand-off
- Monitoring and analytics on Business and Enterprise
Plans and prices
| Free 1 USD one-time usage credit. | Free |
| Pro 60 USD of monthly usage credit; 20 projects. | $60 a month, or $54 a month billed yearly |
| Business 150 USD of monthly usage credit; Unlimited projects. | $150 a month, or $135 a month billed yearly |
| Enterprise Custom platform fee; Credits priced at 200 per dollar. | Custom contact sales |
GPT-trainer
Best for multichannel support inboxesGPT-trainer came fifth with 2 out of 5. Its agents got 37 of 60 runs right, and its failure handling matched Dust, with alerts and hand-off in 8 of 12 failure checks. But it was the slowest to build at about 163 minutes per agent, and only 6 of 10 connections worked without code. See our GPT-trainer review.
GPT-trainer builds voice and text agents trained on your documents, with one inbox for email, SMS, web chat and social messages. Voice agents take and make calls and can transfer to a person. It links to HubSpot, Zendesk, Make, Zapier and n8n. Paid plans start at $150 a month, billed monthly, and it has just 2 reviews on Capterra (checked November 2026).
Who it's for
Support teams that talk to customers across phone, email, SMS and chat and want one inbox.
Skip it if
Price matters. Dust and Voiceflow did better in our test and cost less to start.
What we liked
- Alerts and hand-off in 8 of 12 failure checks
- Voice agents with call transfer
- One inbox across email, SMS and chat
What we didn't
- Slowest build at about 163 minutes
- Only 6 of 10 connections without code
- Paid plans start at $150 a month
Key features
- Inbound and outbound voice agents
- Unified inbox across channels
- Models from OpenAI, Anthropic, Google, Mistral and DeepSeek
- AI Supervisor and custom domains on Scale
Plans and prices
| Self-Serve 12,000 message credits; 2 chatbots. | $150 a month |
| Scale 60,000 message credits; 8 chatbots. | $600 a month |
Stack AI
Best for enterprise governanceStack AI came sixth with 2 out of 5. Its agents got 38 of 60 runs right, second only to Dust, but each took about 151 minutes to build, alerts and hand-off worked in only 6 of 12 failure checks, the lowest here, and 6 of 10 connections worked without code. Our Stack AI review explains the scores.
Stack AI is a no-code platform for governed agents inside business systems, aimed at IT teams. It offers human review steps, claims 500+ integrations, and runs in the cloud, a private cloud or on-premise. It lists SOC 2 Type II, ISO 27001, HIPAA and GDPR. Public pricing has only Free and custom Enterprise, and Asana announced it was buying the company in May 2026. On G2 it has 4.5 out of 5 from 38 reviews (checked November 2026).
Who it's for
IT-led companies in regulated fields that need governed agents, possibly on their own servers.
Skip it if
You're a small team wanting a simple paid plan. Dust has published monthly seats and did better overall.
What we liked
- 38 of 60 runs right, second only to Dust
- Free plan with 500 runs a month
- SOC 2 Type II, ISO 27001 and HIPAA
What we didn't
- Alerts and hand-off in 6 of 12 failure checks
- About 151 minutes to build each agent
- No paid plan between Free and Enterprise
Key features
- No-code builder for multi-step agents
- Human review at key decisions
- Knowledge bases on company data
- On-premise and VPC on Enterprise
Plans and prices
| Free 500 runs a month; 2 projects. | Free |
| Enterprise All features and data loaders; Dedicated infrastructure and solution engineers. | Custom contact sales |
Botpress
Best free start for support agentsBotpress came seventh with 2 out of 5. Each agent got exactly 12 of 20 runs right, 36 of 60 in total: consistent, if not high. Builds took about 146 minutes per agent, alerts and hand-off worked in 7 of 12 failure checks, and only 5 of 10 connections worked without code, the fewest here. Our Botpress review has more.
Botpress builds support agents that answer tickets, take actions such as refunds and hand off with the conversation history. It connects to webchat, WhatsApp, Slack, Zendesk and Intercom. Voice is Enterprise-only, its SOC 2 status is unclear and pricing changed for new workspaces in May 2026. On G2 it has 4.5 out of 5 from 498 reviews (checked November 2026), the most here.
Who it's for
Support teams with developer help who want to start free and pay as conversations grow.
Skip it if
You want dependable no-code agents. Dust and Voiceflow got more runs right.
What we liked
- Free plan with 100 conversations a month
- Consistent results on all three agents
- Hand-off keeps the conversation history
What we didn't
- Only 36 of 60 runs correct
- Fewest no-code connections: 5 of 10
- Voice only on Enterprise
Key features
- Studio with visual and plain-language building
- Knowledge Bases and Tables
- Whitelabel webchat on Plus and up
- Roles and routing on Team
Plans and prices
| Free 100 conversations a month; 3 seats. | Free |
| Plus 250 conversations; 25 USD AI spend. | $150 a month, billed yearly |
| Team 1,500 conversations; 150 USD AI spend. | $750 a month, billed yearly |
| Enterprise Voice channel; Uptime SLA. | Custom contact sales |
GPTBots.ai
Best for certified no-code agentsGPTBots.ai came last of eight with an even 2 out of 5. Its agents got 35 of 60 runs right, the fewest in our test. Each took about 139 minutes to build, alerts fired in 7 of 12 failure checks but hand-off worked in only 6, and 6 of 10 connections worked without code. Read our GPTBots.ai review.
GPTBots.ai is a hosted no-code platform for support, sales and search agents, with a knowledge base, workflow builder and model choice. It connects to WhatsApp, Messenger, Slack and Zapier and lists SOC 2 Type II, ISO 27001, ISO 27701 and ESOF. The Business price isn't published, and we found no user ratings.
Who it's for
Security-focused companies that need certified no-code agents and will tune them carefully.
Skip it if
You need dependable agents from day one. Dust and Dify gave better results.
What we liked
- Four security certifications listed
- Connects to WhatsApp, Messenger and Slack
- Free plan with core building tools
What we didn't
- Fewest correct runs: 35 of 60
- Hand-off in only 6 of 12 failure checks
- Business price not published
Key features
- No-code agent builder with templates
- Knowledge base for company data
- No-code workflow builder
- Private deployment on Enterprise
Plans and prices
| Free 100 credits a month. | Free |
| Business 10,000 credits a month; Up to 100 agents and workflows, 10 published. | Custom contact sales |
What it costs to get started
Every builder here can be started free. Langflow is free software, and Dify's Community edition is free to self-host. Botpress's free plan covers 100 conversations a month, Stack AI's 500 runs and GPTBots.ai's 100 credits. Dust gives a free seat with 500 lifetime credits, Dify's Sandbox has 200 message credits in total, Voiceflow's free plan isn't for production, and GPT-trainer has a 14-day trial.
The cheapest paid month is Dust Pro at 30 EUR a seat, billed monthly, excluding VAT, then Voiceflow Pro at $60 a month and GPT-trainer Self-Serve at $150 a month. Dify Professional is $49.17 a month, billed yearly ($590 a year), and Botpress Plus is $150 a month, billed yearly. Stack AI and GPTBots.ai publish no paid prices. Some Botpress prices come from a dated third-party listing because the official page couldn't be read.
| Tool | Cheapest way to start | Free plan | Free trial |
|---|---|---|---|
| Dust | 30 EUR a month (Pro) | Yes | No |
| Dify | $49.17 a month, billed yearly (Professional) | Yes | No |
| Langflow | Free | Yes | No |
| Voiceflow | $60 a month (Pro) | Yes | Yes |
| GPT-trainer | $150 a month (Self-Serve) | Yes | 14 days |
| Stack AI | Free | Yes | No |
| Botpress | $150 a month, billed yearly (Plus) | Yes | No |
| GPTBots.ai | Free | Yes | Not stated |
Prices from each tool's own pricing page (or a dated third-party listing where that page couldn't be read). Prices can change, so check before you buy.
Also worth knowing about
We didn't test these builders, but they come up often and may suit a specific need.
Chatbase
A no-code builder for customer support agents. We haven't tested it.
Microsoft Copilot Studio
Often suggested for businesses on Microsoft 365 and Azure. We haven't tested it.
Flowise
An open-source visual builder often compared with Dify and Langflow. We haven't tested it.
How we tested
Every builder went through the same test from October 17 to October 23, 2026. We built three agents in each: one answering support questions from a help centre, one qualifying a lead and one pulling website data into a sheet. Each ran 20 times with the same inputs. You can read the full set-up in how we test.
We counted correct runs out of 60, timed each build from sign-up, ran 12 failure checks per tool for alerts and hand-off to a person, and counted how many of the 10 connections we needed worked without code. Each builder got a score out of 5, in half points, for reliability, ease of setup, flexibility, integrations and value for money. Together they make the overall score, and the ranking follows it.
How to choose
If nobody writes code, pick a hosted no-code builder such as Dust, Voiceflow or Stack AI. If you have a Python developer who can run a server, Langflow and Dify give more control for little software cost.
Even Dust got only 42 of 60 runs right. Dust and GPT-trainer alerted and handed off in 8 of 12 failure checks, Stack AI in 6, so keep a person in the loop for anything touching customers.
Voiceflow, Botpress and GPT-trainer are built around customer conversations. Dust and Stack AI centre on internal agents working with company data.
Voiceflow treats voice as a core channel and GPT-trainer handles inbound and outbound calls. Botpress offers voice only on Enterprise.
Dust charges per seat with credits, Voiceflow bills usage credits, Botpress counts conversations and Dify charges per workspace per year. Check which unit your usage will grow in.
Terms you'll see
- AI agent
- An AI assistant that can take actions through tools, not just answer questions.
- RAG
- Retrieval-augmented generation, where the agent looks up your own documents before answering.
- MCP
- Model Context Protocol, a standard that lets agents use outside tools.
- Self-hosting
- Running the software on your own servers, so you handle updates and security.
- Human hand-off
- Passing a conversation from the agent to a person.
Frequently asked questions
Which AI agent builder is free?
Langflow is free MIT-licensed software, and Dify's Community edition is free to host yourself. Hosted free plans include Botpress with 100 conversations a month, Stack AI with 500 runs a month, GPTBots.ai with 100 credits a month, and Dust's free seat with 500 lifetime credits. Voiceflow's free plan can't be used in production.
How much do AI agent builders cost?
All can be started free. The cheapest paid month is Dust Pro at 30 EUR a seat, excluding VAT, then Voiceflow Pro at $60 and GPT-trainer Self-Serve at $150. Dify Professional is $49.17 a month and Botpress Plus $150 a month, both billed yearly. Stack AI and GPTBots.ai don't publish paid prices.
How did we test these AI agent builders?
In October 2026 we built the same three agents in every tool: support, lead qualification and data lookup. Each ran 20 times with the same inputs. We counted correct runs, timed each build from sign-up, ran 12 failure checks for alerts and hand-off, and counted how many of 10 needed connections worked without code.
Which AI agent builders are open source?
Langflow uses the MIT license, and Dify uses its own license, which is Apache 2.0 with added conditions. Both can be hosted on your own servers for free, but you then handle security and updates. Voiceflow and Stack AI are closed commercial products.
Which AI agent builder is best for voice agents?
Voiceflow treats voice as a core channel, with phone minutes billed at $0.05 each on top of other usage. GPT-trainer also builds voice agents that take and make calls and transfer to a person. Botpress offers voice only on its Enterprise plan.
How is an agent builder different from a chatbot builder?
A chatbot builder mainly answers questions. An agent builder lets the model take actions through tools and workflows, such as issuing a refund, updating a record or handing a customer to a person. Botpress and Voiceflow both describe agents that take actions like these.
The bottom line
For most business teams, Dust is the agent builder to try first, with someone checking its work; developers should start with Dify or Langflow.
Make your case
Think a tool is missing or wrong?
If you use a tool we reviewed, or work for the company, email us with what you saw. We'll look at the evidence and update the review if it changes our findings.
Think your company belongs on this list?
Send us your product, website and the category it fits. We'll test it the same way as every other tool. It's free, and we can't promise a date or a score.
Liam Foster
If there's a strange new AI tool making its way around the internet, I've probably already bookmarked it. My favorite part is figuring out what these platforms can actually do.
Update history
- November 4, 2026: published with the results of our test, run between October 17, 2026 to October 23, 2026.