The best automation platforms in 2026, compared

A visual builder for agents and workflows on the same platform as the Connect model gateway and Nexus policy layer.

Best for: Teams that want AI agents and business workflows built, tested and governed in one place, with published prices and managed hosting in AWS Ireland, and who can live with a smaller connector catalogue.

Connectors and integrations: 100+ tool integrations, plus an API for custom systems. That is far fewer than the largest catalogues in this guide.[1]

AI-native capability: Builds agents and workflows in one canvas. The Pro tier lists multi-agent orchestration. Agents can connect to external MCP servers (a Pro feature). Swfte does not expose your own agents as an MCP server today.[1][2][3]

Governance and approvals: Role and tenant controls, and human-approval steps in workflows. A workflow step can pause until a named person approves. Policy applied while an agent runs is handled by Nexus. SAML SSO and SCIM are in development.[1][2][3]

Deployment and data control: Managed service, with customer data at rest in AWS Ireland. Dedicated or private deployment is designed for and scoped with Swfte as an engagement. There is no self-serve self-hosting.[1][2]

Pricing transparency: Free (10 workflows, 5 agents, 2,000 executions per month); Pro $39; Team $99 (3 users included); Enterprise custom. Pro lists 25 agents and 25,000 executions a month; Team lists unlimited agents and 100,000 executions.[1]

Extensibility: An API for custom systems is listed on the Team tier. Model access goes through the Connect gateway, which takes your own provider keys.[1][2]

Built and designed, kept apart:

Watch-outs:

• The connector catalogue is a small fraction of the largest in this guide. Check your own systems against the list first.
• SAML SSO and SCIM are in development, not generally available.
• No self-serve self-hosting. Private deployment is an engagement with Swfte.
• Swfte does not publish an uptime SLA figure and does not claim to be “GDPR compliant” or “EU AI Act compliant”.
• Studio drives tools and APIs. It does not operate a desktop screen the way RPA does.

Sources: Swfte Studio page, Swfte pricing, Swfte trust page (read 2026-10-06).

An integration and automation platform covering integration, API management, data management and AI agents, with 1,000+ connectors.

Best for: Large organisations that need a broad integration platform with agent building and governed MCP tools, and that may run runtimes on-premises or choose a North American or UK platform instance.

Connectors and integrations: Boomi states 1,000+ prebuilt connectors across categories such as CRM, ERP and data platforms.[1][2]

AI-native capability: Agentstudio is for building, governing and orchestrating agents. Boomi documents an MCP Server (early access or tech preview) that exposes deployed processes as tools, and agents can connect to MCP servers. Boomi Connect offers 1,000+ governed MCP tools.[1][2][3][4]

Governance and approvals: SAML 2.0 single sign-on is part of Advanced User Security (Enterprise and Enterprise Plus, or an add-on to Professional and Professional Plus). Documented user roles carry sets of privileges. Audit log entries are kept for up to one year, with the UI showing only the last month. Agent tools can be set to 'Requires Approval' so the agent asks the user before acting.[1][2][3][4]

Deployment and data control: Runtimes can be Boomi-hosted (public, dedicated single-tenant or managed cloud) or customer-hosted on-premises or in a public cloud, including single-tenant runtime clusters. The platform is offered as separate North American and European (UK) instances that customers can choose between.[1][2][3]

Pricing transparency: The pricing page lists plan names but no prices for most. It shows Pay-As-You-Go at '$99 per month plus usage, billed monthly', and a 30-day free trial. How usage is measured is not defined on the page. Other plans go through sales.[1]

Extensibility: A public Platform REST API is documented. Data Process steps accept custom JavaScript or Groovy scripts. The Connector SDK is for building custom connectors with REST, OpenAPI, SOAP and authentication support.[1][2][3]

Score 93.3 out of 100. Connectors and integrations: Vendor publishes '1,000+ prebuilt connectors' on its connectors page, matching the 1,000 or more band. AI-native capability: Agent building (Agentstudio) and MCP support (server and client sides) are both published. Governance and approvals: SSO, role-based access, audit logs and a human approval option for agent tool actions are all documented. Deployment and data control: Customer-hosted (on-premises) runtimes and a choice of platform region (North American or European) are both published. Pricing transparency: One price point is published ($99 a month plus usage) but the usage unit is not defined and other plans are not priced. Extensibility: Public API, custom scripting step and a custom connector SDK are all published.

Watch-outs:

• SSO sits in Advanced User Security: Enterprise and Enterprise Plus, or an add-on to Professional and Professional Plus.
• The documented MCP Server is labelled early access or tech preview.
• European and North American platform instances are separate; European users cannot use North American runtime clouds.
• The audit log UI shows only the last month, though entries are kept for up to a year.
• Pay-as-you-go usage charges are not defined on the pricing page.

Sources: Boomi pricing, Boomi homepage, Boomi integration platform, Boomi connectors, Boomi Connect, Boomi trust centre, Boomi MCP overview, Boomi Agentstudio, Boomi building an agent, Boomi SAML SSO, Boomi user roles and privileges, Boomi audit log entries, Boomi integration runtimes, Boomi platform regions, Boomi European Platform Instance blog, Boomi Platform API reference, Boomi custom scripting step, Boomi Connector SDK (read 2026-10-06).

A workflow automation tool, source-available and self-hostable, with hosted cloud plans, AI agent nodes, and MCP server and client nodes.

Best for: Technical teams that want to self-host under the Sustainable Use License or use EU-hosted cloud, with code nodes, AI agents and MCP. Audit logging and custom roles need Enterprise.

Connectors and integrations: n8n's integrations page states "2307 integrations".[1]

AI-native capability: An AI Agent node connects a chat model to tools that the agent chooses between. An MCP Server Trigger node exposes n8n tools to MCP clients, and an MCP Client Tool node lets agents use MCP tools.[1][2][3]

Governance and approvals: Pricing lists SSO and LDAP from Business, and custom roles and audit logging on Enterprise. Project roles apply on all cloud plans and on Registered Community, Business and Enterprise self-hosted; custom roles are Enterprise only. Human-in-the-loop for tool calls pauses an agent until a person approves or denies.[1][2][3][4]

Deployment and data control: Self-hosted editions run from a Docker or other install; the free Community edition is self-hosted. Cloud plans store data in the EU on servers in Frankfurt, Germany. The cloud page names one region, not a choice of regions. Air-gapped use is not addressed in the docs read.[1][2][3]

Pricing transparency: Cloud: Starter 20 euro a month and Pro 50 euro a month, both billed annually (24 and 60 euro on monthly billing), for 2,500 and 10,000 executions a month. Business is 667 euro a month billed annually (40,000 executions, self-hosted). Enterprise is contact sales. Community edition is free.[1]

Extensibility: A public REST API is available on Starter, Pro and Enterprise cloud and on all self-hosted editions. The Code node runs JavaScript and Python. Custom nodes can be built and shared. Source code is public under the Sustainable Use License.[1][2][3][4][5]

Score 91.7 out of 100. Connectors and integrations: Published count of 2,307 integrations, above the 1,000 threshold. AI-native capability: Agent building (AI Agent node) and MCP as both server and client are documented. Governance and approvals: SSO, role-based access and audit logging all published (audit logging and custom roles on Enterprise) plus approve or deny step for agent tool calls. Deployment and data control: Self-hosting published, but only one named cloud region (Frankfurt) rather than a region choice, so not 3. Pricing transparency: Self-serve tiers priced with a defined unit (executions); Enterprise is custom, so not 3. Extensibility: Public API plus Code node, plus a custom node framework and source-available code.

Watch-outs:

• The Sustainable Use License allows use only for internal business purposes or non-commercial use; it is not an OSI open-source licence.
• Business is self-hosted only and, per the pricing page, comes with community forum support.
• Audit logging and custom roles appear only on Enterprise.
• Cloud data is stored in Frankfurt; no other cloud region is listed.

Sources: n8n pricing, n8n integrations, n8n AI page, n8n docs sitemap, n8n AI Agent node, n8n MCP Server Trigger, n8n MCP Client Tool, n8n Code node, n8n choose how to use, n8n self-hosting, n8n human in the loop for tools, n8n public API, n8n RBAC, n8n log streaming, n8n community licence, n8n GitHub repository (read 2026-10-06).

A workspace combining chat, agents, flows and tables with 760+ apps; the core is MIT-licensed and can be run on managed cloud, self-hosted or air-gapped.

Best for: Teams that want an open-source automation platform with agent building, MCP support and a self-hosted or air-gapped option, and that accept credit-based pricing.

Connectors and integrations: The homepage and MCP page cite 760+ apps; the MCP page's directory shows 765 apps. Pieces are described as MIT-licensed and mostly community-contributed.[1][2]

AI-native capability: Activepieces has an AI agent builder where you set a goal and pick the apps, flows and files the agent may use. It acts as an MCP server (free on every plan) and agents can use external MCP servers.[1][2]

Governance and approvals: The homepage lists SSO (Okta/Entra), granular role-based access and audit logging to SIEM. Pricing shows SSO and standard roles on Team, and SCIM, custom roles and audit logs on Ultimate. A Wait for Approval step pauses a run for a person, and the agent builder page describes approval steps.[1][2][3][4]

Deployment and data control: Managed cloud, self-hosting (Docker or Kubernetes) and air-gapped deployment are published. The security page says the cloud runs in Frankfurt by default with other regions available; it does not name those other regions on that page.[1][2][3]

Pricing transparency: Free $0 (1,000 credits/month, up to 5 users); Plus $20/month (10,000 credits, overage $0.007/credit); Team $200/month (50,000 credits, 25 users). Ultimate is custom and sales-led. Embed starts at $36k/year.[1]

Extensibility: A REST API covers projects, flows, runs, connections and more (API keys are generated in the platform dashboard, which the docs tie to Platform and Enterprise editions). Flows can include TypeScript code, and users can build custom pieces in TypeScript. The core is MIT-licensed.[1][2][3]

Score 90 out of 100. Connectors and integrations: Vendor publishes 760+ apps (directory 765), which falls in the 200-999 band. AI-native capability: Agent building is published on the AI agent builder page and MCP server and client support is published on the MCP page. Governance and approvals: SSO, RBAC and audit logs are all published (audit logs and custom roles on the Ultimate plan), and a Wait for Approval step is documented. Deployment and data control: Self-hosted and air-gapped options are published, and the security page states a default cloud region with other regions available. Pricing transparency: Self-serve tiers are priced against a defined unit (credits), and the Ultimate tier is custom. Not every tier is priced publicly. Extensibility: Public REST API, TypeScript code in flows, a custom piece framework and an MIT-licensed core are all published.

Watch-outs:

• API keys are generated in the platform dashboard, which is only available in the Platform and Enterprise editions.
• Audit logs, SCIM and custom roles are listed under the sales-led Ultimate plan.
• Paused approval runs are kept for 30 days; the agent builder page says approvals time out in 24 hours.

Sources: Activepieces homepage, Activepieces pricing, Activepieces MCP, Activepieces AI agent builder, Activepieces security, Activepieces technical limits, Activepieces REST API, Activepieces building pieces, Activepieces architecture overview, Activepieces Human Input piece (read 2026-10-06).

A visual automation builder where scenarios connect apps, with AI agents built on the same canvas and a hosted MCP server.

Best for: Teams that want visual scenario building with a large app library and credit-based pricing from a low monthly price. SSO, audit logs and custom roles need Enterprise.

Connectors and integrations: Make's integrations page states "3,000+ Integration Apps".[1]

AI-native capability: Agents are built in Make's visual builder and run and debugged on the same canvas. Make provides a Make-hosted MCP server (streamable HTTP) that lets clients such as Claude and ChatGPT run scenarios.[1][2]

Governance and approvals: The pricing table lists SSO, audit logs and custom roles on Enterprise only. The help centre says audit logs are an Enterprise feature, kept for 12 months. A Human in the loop app inserts manual review or approval steps into scenarios; the page read does not state its plan.[1][2][3]

Deployment and data control: Pricing lists hosting on AWS (EU/North America) for every plan. The security page says Enterprise runs in a separately managed AWS environment isolated from self-service customers, and mentions on-prem agents. The on-prem agent only reaches systems on a local network; Make itself stays in the cloud. Zones (eu1, eu2, us1, us2) are assigned to an organisation.[1][2][3][4]

Pricing transparency: Monthly-billing prices are Free $0 (1,000 credits), Core $9, Pro $16 and Teams $29, each shown with 10,000 credits a month. Annual billing saves 15% or more, with no exact annual prices on the page. Enterprise is custom. A credit is one module action.[1]

Extensibility: A REST API exists, but most endpoints need a paid plan. The Make Code app runs JavaScript or Python on paid plans, at two credits per second of execution. Users can build custom apps for services Make lacks.[1][2][3]

Watch-outs:

• Most API endpoints are reserved for paid plans.
• The Make Code app costs two credits per second and runs in a sandbox limited to 30 seconds outside Enterprise.
• SSO, audit logs and custom roles are shown on Enterprise only in the pricing table.

Sources: Make pricing, Make integrations, Make AI agents, Make MCP server docs, Make security, Make API documentation, Make custom apps documentation, Make API structure (zones), Make audit logs help, Make HITL app, Make on-prem agent help, Make Code app help (read 2026-10-06).

Microsoft's workflow and robotic process automation service, with cloud flows, desktop flows, approvals and Copilot help for building flows.

Best for: Organisations already on Microsoft 365 and Entra that want flows, approvals and desktop RPA with more than 1,000 connectors, and can accept per-user or per-bot pricing.

Connectors and integrations: The product page states more than 1,000 API connectors and, in another passage, more than 1,400 prebuilt, certified connectors. Users can also create their own custom API connector.[1]

AI-native capability: Published AI features are Copilot for creating and editing flows (cloud and desktop), AI recommendations, built-in AI models for document and image processing, and built-in GPT support for content generation. Agent building and MCP support were not seen on the Power Automate pages read; the pricing page lists Microsoft Copilot Studio as a separate product.[1][2]

Governance and approvals: Human approval steps are documented (Start and wait for an approval, with approve/reject, custom-response and sequential types). Access to create flows is controlled by security roles within environments, users are authenticated by Microsoft Entra ID, and Power Automate activity logs (flow create, edit, delete, permission changes) are viewable in Microsoft Purview once auditing is switched on. The Purview log does not cover individual flow runs.[1][2][3]

Deployment and data control: Environments can be created in a chosen region, and an app and its data stay in the region that hosts the environment (for example, EU and EFTA data centres). The pages read do not say whether a self-hosted or on-premises option for the platform exists; desktop flows run attended or unattended on machines, and an on-premises data gateway is mentioned as unavailable in the India region.[1][2]

Pricing transparency: Published prices (USD): Premium $15.00 per user per month, paid yearly; Process $150.00 per bot per month, paid yearly; Hosted Process $215.00 per bot per month, paid yearly; Process Mining add-on $5,000.00 per tenant per month, paid yearly; 30-day free trial. Microsoft Copilot Studio is listed as a pre-purchase of Copilot Credit Commit Units with 'up to 20%' saving and no fixed price figure.[1]

Extensibility: Cloud flows can be managed with code through the Dataverse SDK for .NET and the Dataverse Web API. Desktop flows have scripting actions for PowerShell, Python, VBScript, JavaScript and C#/VB.NET. Custom API connectors can be created.[1][2][3]

Watch-outs:

• Approval flows need a Microsoft Dataverse database; in non-default environments the first run needs an administrator to provision it.
• Purview audit logs cover flow lifecycle and permission events only, not individual runs or action executions.
• The unsupported api.flow.microsoft.com API is not recommended; Microsoft points to the Dataverse Web API instead.
• Only solution-aware flows can be managed with code; flows under My Flows are not supported.
• Python scripting action default version is Python 2.7, and VBScript is being deprecated in Windows.

Sources: Power Automate product page, Power Automate pricing, Learn: approvals, Learn: getting started (home page, environments, roles), Learn: Dataverse security, Learn: Purview activity logs for Power Automate, Learn: environment regions, Learn: scripting actions, Learn: work with cloud flows using code (read 2026-10-06).

A developer-oriented workflow platform where you build workflows with prebuilt app triggers and actions plus your own Node.js, Python, Go or Bash steps.

Best for: Developers who want code steps alongside thousands of app connectors, priced by credits, and who also want a hosted MCP server for AI tool calls.

Connectors and integrations: The pricing page and docs both state 3,000+ connectors or integrated apps.[1][2]

AI-native capability: Pipedream runs a hosted remote MCP server (mcp.pipedream.com) giving access to over 10,000 pre-built tools across 3,000+ apps, for developers to add to their own agents or for end users to connect accounts. The docs refer to AI agents as a use case, but a Pipedream agent builder was not seen.[1][2]

Governance and approvals: SSO (Okta, Google or any SAML 2.0 or Google OAuth provider) is Business plan only, and SCIM is Business plan only. Workspace owners and admins can perform all project actions while members are restricted; per-project access restriction is Business plan. The suspend and rerun functions can pause a workflow until someone approves it by link. An audit log API is documented for Conduit, but an audit log for the hosted product was not seen.[1][2][3][4]

Deployment and data control: The security page says Pipedream is hosted on AWS in the us-east-1 region, with no region selection. Separately, the docs describe Conduit as a self-hosted application on customers' own infrastructure (Docker or Kubernetes), but the pages read did not show it is a self-hosted version of the workflow platform.[1][2]

Pricing transparency: Free: $0, development only, no production environment. Startup: $99 per month billed monthly or $90 per month on annual billing, including 100 external users and 10,000 usage credits per month; overage $2 per extra user and $0.012 per credit. Business: custom pricing, annual contract only, contact sales.[1]

Extensibility: Workflows accept custom code steps in Node.js (with npm packages), Python, Go and Bash. A public REST API and TypeScript, Python and Java SDKs exist for Connect. A component framework builds custom sources and actions, and the component registry is published on GitHub.[1][2][3][4]

Watch-outs:

• The Free plan is development only, with no production environment.
• SSO, SCIM and per-project access restriction are on the Business plan, which is custom-priced with an annual contract.
• Hosting is a single AWS region, us-east-1, with no region selection.
• HTTP trigger payloads are capped at 512KB unless the upload parameter is used.

Sources: Pipedream pricing, Pipedream docs home, Pipedream docs index, Pipedream MCP, Pipedream Node.js code, Pipedream Connect API reference, Pipedream access controls, Pipedream SSO, Pipedream HTTP triggers, Pipedream suspend and rerun, Pipedream components, Pipedream privacy and security, Pipedream Conduit install, Pipedream Conduit access control (read 2026-10-06).

A workflow platform it calls the secure environment for AI agents, apps and automation, with IT and security governance controls.

Best for: Security and IT teams that want agents, MCP, Python scripting and human approval pages in one governed tool, on Tines cloud in a chosen region or on-premises, and accept sales-led pricing.

Connectors and integrations: Tines says it offers '1000s' of preconfigured action templates and can connect to any tool with an API. The vendor gives no exact count and templates are per-action, not per-app, so a connector band is not set.[1][2]

AI-native capability: The AI agent action runs in task or chat mode and can use templates, Send to Story, custom tools and remote MCP servers as tools. A story can also be exposed as an MCP server.[1][2]

Governance and approvals: SSO/SAML is stated as available by default across all plans. Audit logs are captured when users change data and are available in the UI and API. Default roles are viewer, builder and manager; custom roles are an Enterprise feature not on all plans. Tines Pages pause a story for human approval.[1][2][3]

Deployment and data control: Tines-hosted cloud or on-premises deployment. Tines Cloud runs in seven AWS regions: Tokyo, Sydney, Dublin, London, Stockholm, Zurich and Oregon.[1][2]

Pricing transparency: A Free edition ($0, 3 live workflows, one-time $50 AI allowance). Paid editions show no price; 'Get pricing' goes to a demo booking.[1]

Extensibility: A public API with bearer-token auth covering stories, cases, records, audit logs and more. Run script executes Python 3.13. Users can create private templates from an API spec and use a Build your own option for products without a template.[1][2][3]

Watch-outs:

• Paid editions are not priced publicly; the pricing page links to a demo booking.
• Custom roles are part of Enterprise Tenant Management and are not available on all plans.
• Run script supports Python 3.13 only, with a default 10 second timeout and a 110 second maximum.

Sources: Tines homepage, Tines pricing, Tines security, Tines AI agent action docs, Tines MCP server template docs, Tines action templates docs, Tines custom roles docs, Tines human-in-the-loop blog, Tines Cloud AWS regions, Tines API info, Tines Run script docs, Tines integrations blog, Tines docs quickstart (read 2026-10-06).

An enterprise orchestration platform for data and AI: an iPaaS with 700+ connectors, an agent builder and managed MCP servers.

Best for: Larger teams that want an iPaaS with an agent builder and governed MCP tooling, and are happy to buy through sales and add SSO, regional hosting and agent features as paid extras.

Connectors and integrations: The pricing page and homepage both state 700+ connectors, on every plan (Pro, Team, Enterprise).[1][2]

AI-native capability: Merlin Agent Builder is described as a no-code builder for conversational or autonomous agents. Agent Gateway exposes Tray workflows and connectors as managed MCP server tools. The pricing page lists Agent Development (Gateway for MCP plus Merlin Agent Builder) as a paid add-on.[1][2][3]

Governance and approvals: SAML 2.0 single sign-on is available on Enterprise plans or as an add-on. The governance page describes one role-based access model across workflows, agents and MCP tools, and audit logging with streaming to SIEM tools. It mentions approval gates between dev, staging and prod, but no human approval step inside a workflow was found.[1][2][3]

Deployment and data control: The trust centre lists AWS hosting with US West by default and EU and APAC regions available. The pricing page lists regional hosting as a paid add-on and the homepage mentions on-premise connectivity on Enterprise. The security statement describes a multi-tenant design. No self-hosting option was found.[1][2][3][4]

Pricing transparency: Three plans (Pro, Team, Enterprise) with no prices published. The page says to contact sales for a quote based on expected usage. Add-ons include Agent Development, SSO, HIPAA and regional hosting.[1]

Extensibility: A Script connector runs JavaScript (ES6) steps. A GraphQL API is documented, including a callConnector mutation. Documentation says you can build your own connector with the Tray.ai connector SDK.[1][2][3]

Watch-outs:

• No prices are published for any plan; a quote from sales is needed.
• SSO is Enterprise-only or a paid add-on.
• Regional hosting, the agent features (Merlin Agent Builder, MCP gateway) and longer log retention are paid add-ons.
• The security statement describes a multi-tenant design.

Sources: Tray.ai pricing, Tray.ai homepage, Tray.ai trust centre, Tray.ai governance and trust, Tray.ai security statement, Tray.ai profile and login (SSO) docs, Tray.ai Agent Gateway overview, Tray.ai Script connector docs, Tray.ai call connector API docs, Tray.ai connectors page, Tray.ai documentation home (read 2026-10-06).

A platform for business orchestration and automation, coordinating AI agents, robots and people.

Best for: Enterprises that need governed agents and human approval steps, a choice of cloud regions or self-hosting, and are comfortable with sales-led pricing.

Connectors and integrations: Integration Service has a catalogue of pre-built connectors (Salesforce, SAP, ServiceNow, Workday and others named), but neither the product page nor the docs give a connector count.[1][2]

AI-native capability: Agent Builder is a low-code canvas for building agents, with Python SDKs for coded agents. Agents can use MCP servers as tools, and UiPath-hosted MCP servers can expose workflows and agents.[1][2][3]

Governance and approvals: The pricing page lists SSO with your own identity provider, custom roles and access policies, and searchable, exportable audit events, without tying them to a single plan. Maestro documents a human approval step that pauses a process and routes on approve or reject.[1][2]

Deployment and data control: Cloud in 13 listed regions (including EU, US, UK, Switzerland, Japan, India), plus self-hosted on Kubernetes or Linux VMs, and a dedicated cloud add-on for Enterprise. The Basic plan is listed as Europe region only.[1][2]

Pricing transparency: Community is free; Basic starts at $25/month (up to 5 Basic, 5 Plus and 1 Pro user, 2 robots), with a note that an updated plan is coming. Standard and Enterprise are contact sales.[1]

Extensibility: A public Orchestrator API, Python and TypeScript SDKs (plus an MCP SDK), and a Connector Builder for custom connectors to third-party APIs.[1][2][3]

Watch-outs:

• Standard and Enterprise pricing is contact sales; the Basic plan page says an updated offering is coming.
• Enterprise data retention for audit events is stated as 5 years, Standard 2 years.
• If a service is not available in the tenant region, it is hosted in a fallback region (per the data residency docs).

Sources: UiPath homepage, UiPath pricing, UiPath Integration Service product page, UiPath Integration Service docs, UiPath Agents MCP servers, UiPath Agents SDKs, UiPath global cloud regions, UiPath Maestro human approval, UiPath Orchestrator API guide, UiPath trust and security (read 2026-10-06).

An open-source, code-first orchestration platform for internal software, with scripts in many languages, flows and apps, that can be self-hosted.

Best for: Engineering teams that write scripts in code, want open-source software they can self-host (including air-gapped) or run on Windmill Cloud, and are comfortable with a smaller pre-built integration list.

Connectors and integrations: The integrations page says '+50 pre-made integrations', while the docs say the Hub lists '200+ resource types'. The two counts use different units and fall either side of the 200 threshold.[1][2]

AI-native capability: AI agent steps can be added to flows, with a range of model providers and script, web-search and nested-agent tools. Windmill exposes an MCP server and can also act as an MCP client; agents can use MCP servers as tools.[1][2][3]

Governance and approvals: Docs describe roles (Superadmin to Operator), folders and ACLs. SAML and SCIM sit in Enterprise Edition. The pricing page lists audit logs, SAML with group sync and approval steps on Enterprise, and none on the free tier. Docs say approval steps exist, with forms and permission controls limited to Cloud and Enterprise.[1][2][3][4][5]

Deployment and data control: Self-hosting is documented with Docker Compose, Kubernetes (Helm) and several cloud providers, and the Windmill site says air-gapped deployments are supported with offline images. Enterprise Cloud instances can be placed in US (us-east) or EU (eu-central) at no extra cost, per the data processing addendum.[1][2][3]

Pricing transparency: Free and open-source tier at $0. Enterprise starts from $120 a month, with developer seats at $20 a month, operators at $10 a month and compute units for workers from about $50 a month. Monthly or annual billing. Larger deployments are custom.[1]

Extensibility: Scripts can be written in many languages. An OpenAPI-described API and clients for TypeScript, Python and Rust are documented. Core code is open source under AGPLv3 (some enterprise code is proprietary).[1][2]

Watch-outs:

• The free tier lists no audit logs, no approval steps and SSO capped at 10 users, per the pricing page.
• SAML and SCIM are Enterprise Edition features.
• Approval-step forms and permission controls are limited to Cloud and Enterprise self-hosted.
• The open-source repository also contains proprietary enterprise code.
• The EU/US region choice is documented in the Enterprise Cloud data processing addendum.

Sources: Windmill homepage, Windmill pricing, Windmill docs intro, Windmill integrations, Windmill integrations on Windmill docs, Windmill AI agents docs, Windmill MCP docs, Windmill roles and permissions, Windmill SAML and SCIM, Windmill audit logs, Windmill approval steps, Windmill self-host docs, Windmill self-host platform page, Windmill Enterprise Cloud data processing addendum, Windmill GitHub repository (read 2026-10-06).

An enterprise integration and automation platform that adds AI agents (Agent Studio genies) and MCP servers, with regional data centres.

Best for: Enterprises that need many connectors, regional data residency, governed AI agents and MCP servers, and are prepared to talk to sales for pricing.

Connectors and integrations: The homepage says Workato connects to "17,000+ apps, APIs, databases, files, MCP servers". The integrations page says only "thousands" of apps, so the number is a mixed count rather than apps alone.[1][2]

AI-native capability: Agent Studio is for building interactive AI agents called genies. Workato MCP servers expose endpoints to AI agents as tools, built from project assets, an API collection or a proxy to external MCP servers.[1][2]

Governance and approvals: The security page lists role-based access control with custom roles and audit logging with configurable 90-day retention. Workspace SAML SSO is documented. Agent Studio lets admins set approval workflows for critical genie actions.[1][2][3]

Deployment and data control: Enterprise customers choose a data centre region: US, EU, JP, SG, AU, IL, CN, KR or UK. Self-service users cannot choose and sit in a US data centre. Virtual Private Workato is a dedicated environment in its own AWS VPC, run by Workato, at extra pricing. The on-prem agent only connects local systems; it does not host Workato. No customer self-hosting is described.[1][2][3]

Pricing transparency: The pricing pages read show no prices, plans or billing unit. They mention "pay-as-you-go", a Capped Recipes model, a free start and a talk-to-sales option, but give no rates. The data-centre page refers to self-service Free and Pro plans.[1][2][3]

Extensibility: The Developer API manages recipes, connections and jobs (bearer-token auth, access depends on plan). A recipe action runs custom JavaScript on Node.js 20.11.0. The Connector SDK builds custom connectors.[1][2][3]

Watch-outs:

• Region choice is for Enterprise customers; self-service accounts sit in a US data centre.
• Virtual Private Workato carries additional pricing and data privacy requirements; the pages read give no price.
• The JavaScript action allows 256 MB of memory, a 30-second timeout and no user-supplied libraries.
• Developer API and SDK access can depend on the pricing plan and contract.

Sources: Workato homepage, Workato integrations, Workato pricing, Workato pricing tiers, Workato predictable pricing, Workato security, Workato Agent Studio docs, Workato MCP servers docs, Workato SSO docs, Workato data centre overview, Workato Virtual Private Workato, Workato on-prem agent, Workato Developer API, Workato JavaScript connector, Workato Connector SDK (read 2026-10-06).

A hosted automation platform for connecting apps with workflows (Zaps), with AI agents and an MCP service for AI clients.

Best for: Teams that want the widest app coverage on a self-serve, task-priced plan, with AI agents and MCP access to those apps. Enterprise controls sit on the custom plan.

Connectors and integrations: Zapier's app directory states "10207+ apps". Its MCP and agents pages cite 9,000+ apps.[1][2]

AI-native capability: Zapier Agents lets users create agents that work across 9,000+ apps. Zapier MCP connects AI clients such as Claude, ChatGPT and Cursor to 9,000+ apps and 66,000+ triggers and actions.[1][2]

Governance and approvals: Pricing page lists SAML SSO on Team (audit logs kept 6 months) and on Enterprise (audit logs kept 1 year, Super Admin and advanced admin permissions, SCIM). The enterprise page describes role-based access and an audit trail. A Request approval step in Zaps, and agents that pause for a human, are described.[1][2][3]

Deployment and data control: The pages read describe a hosted service. The enterprise page mentions VPC peering for private connections and a bring-your-own-model option, but states nothing about self-hosting, single-tenant deployment or region choice. The pricing page lists no data region.[1][2]

Pricing transparency: Free is $0 a month for 100 tasks. Professional is $19.99 to $3,389 a month on annual billing ($29.99 to $5,099 monthly), from 750 up to 2,000,000 tasks. Team is $69 to $3,999 a month on annual billing ($103.50 to $5,999 monthly). Enterprise is custom, contact sales. The unit is tasks.[1]

Extensibility: A public API at api.zapier.com (OAuth 2.0) covers Workflow, Actions, Trigger Inbox and Connections. Code steps run JavaScript or Python. Developers can build their own integrations with a Visual Builder or the Zapier CLI, and a Zapier SDK is documented.[1][2][3][4]

Watch-outs:

• Pricing is by tasks, which count successful automation steps and programmatic calls, so costs rise with multi-step Zaps.
• Audit log retention is 6 months on Team and 1 year on Enterprise.
• Enterprise pricing is custom and needs a sales conversation.

Sources: Zapier pricing, Zapier app directory, Zapier MCP, Zapier Agents, Zapier Enterprise, Zapier blog: human in the loop, Zapier docs index, Zapier API overview, Zapier platform docs, Zapier help: code in Zaps (read 2026-10-06).