The best AI agent builders in 2026, from no-code to enterprise

A visual builder for agents and workflows, with the Connect gateway for model choice and Nexus for policy applied while agents run.

Best for: Teams that want governed agents and the workflows around them in one place, with approvals, traces and published prices, and managed hosting in AWS Ireland.

Build surface: A visual canvas for agents and workflows. The Team tier lists API access. Agents and workflows are two models under one editor.[1][2]

Tools, integrations and MCP: 100+ tool integrations, plus an API for custom systems. Agents can connect to external MCP servers (a Pro feature). Swfte does not expose your own agents as an MCP server today.[1][2]

Multi-agent, workflows and evaluation: The Pro tier lists multi-agent orchestration, and workflows are the core of the builder. Evaluation suites with an automated judge exist in the platform service. The Studio page also describes golden sets and regression runs; their status is listed as to confirm below.[1][2]

Governance and approvals: Role and tenant controls, human-approval steps in workflows, and a workflow step that pauses until a named person approves. Nexus applies policy while agents run. SAML SSO and SCIM are in development.[1][2]

Deployment and data control: Managed service with customer data at rest in AWS Ireland. Dedicated or private deployment is designed for and scoped as an engagement. 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. Studio meters executions.[1]

Built and designed, kept apart:

Watch-outs:

• SAML SSO and SCIM are in development, not generally available.
• No self-serve self-hosting. Private deployment is an engagement with Swfte.
• The tool catalogue is 100+ integrations, a fraction of the largest in this guide.
• Agents and workflows share an editor but are not one graph model.
• Swfte does not publish an uptime SLA figure.

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

Botpress describes itself as an AI agent platform for customer support, with a visual Studio, a TypeScript development kit and a platform API.

Best for: Teams building customer-support agents who want a visual studio plus a TypeScript kit, and who can live with per-conversation pricing and role-based access only on the Team plan or above.

Build surface: Botpress Studio is described as a visual, drag-and-drop interface for building agents. The docs also list the ADK, a TypeScript library for building agents from code, and a Platform API for Botpress Cloud.[1][2]

Tools, integrations and MCP: Agents can connect an MCP server (server URL and authentication set under Build > Tools) and can make HTTP requests. A live integration catalogue exists, but the docs and integrations page give no total count.[1][2]

Multi-agent, workflows and evaluation: The ADK lists workflows among its building blocks and describes evals as automated conversation tests. The homepage highlights multi-step workflows. Multi-agent orchestration was not found in the pages read.[1][2]

Governance and approvals: The docs say access control requires the Team plan or higher, and the pricing page lists Teams and RBAC on Team and a log of workspace changes on paid plans. The trust page lists 'Internal Single-Sign-On (SSO)' as an access-control feature. No human-approval step was found; human handoff is documented via Botpress Desk.[1][2][3]

Deployment and data control: Botpress Cloud is the documented hosting route; the ADK deploys to Botpress Cloud. The trust page names hosting only as 'Major Cloud Provider'. No region choice, self-hosting or on-premises option was found.[1][2]

Pricing transparency: Free is $0 with 100 conversations a month. Plus is $150 a month billed annually (page says save 20%) with 250 conversations and $0.65 per extra conversation. Team is $750 a month billed annually with 1,500 conversations and $0.50 per extra. Enterprise is custom.[1]

Watch-outs:

• Access control (roles) needs the Team plan or higher, which the pricing page lists at $750 a month billed annually.
• Plus and Team plans include a capped number of conversations a month, with a per-conversation overage charge.
• The Free plan allows 3 AI agents and community support only.

Sources: Botpress pricing, Botpress homepage, Botpress docs home, Botpress integrations, Botpress trust centre, Botpress ADK introduction, Botpress tools and integrations, Botpress Autonomous Node, Botpress control access (read 2026-10-06).

CrewAI is an open-source framework for multi-agent systems, with a commercial platform (AMP) for deploying, monitoring and governing agents, including a no-code Crew Studio.

Best for: Teams that want a Python framework for crews of agents and flows, with an optional paid platform for deployment, governance and a visual studio.

Build surface: The open-source framework is code-first. CrewAI AMP adds Crew Studio, a visual no-code interface for creating crews, and a REST API for deployed crews.[1][2]

Tools, integrations and MCP: CrewAI supports MCP as a client (stdio, streamable HTTP and SSE) and custom tools via BaseTool or a decorator. The tools page lists named tools but states no total count.[1][2]

Multi-agent, workflows and evaluation: The docs describe Crews (teams of collaborating agents) and Flows (state and control flow), and a crewai test command that runs a crew over several iterations and scores tasks, agents and the crew from 1 to 10.[1][2]

Governance and approvals: The pricing page lists SSO, RBAC, workload identity, PII redaction and policies under Enterprise. The AMP announcement says 'RBAC + audit logs'. Human-in-the-loop is documented through a human_feedback decorator and, for Enterprise, webhook-based review.[1][2][3][4]

Deployment and data control: The pricing page says Enterprise can deploy on CrewAI cloud, your own VPC, or your own infrastructure. The Basic plan runs on CrewAI cloud. No cloud region choice was found on the pages read.[1]

Pricing transparency: Basic is free with 50 workflow executions a month. Enterprise is custom with a quote required. No dollar figure and no overage price is shown.[1]

Watch-outs:

• The pricing page shows only Basic (free, 50 workflow executions a month) and Enterprise (custom).
• SSO and RBAC are listed under Enterprise only; the Basic plan lists no access control.
• The pricing page does not state billing cycle, audit-log detail or region options.
• The CrewAI test command currently supports OpenAI as the only provider, per the testing docs.

Sources: CrewAI pricing, CrewAI homepage, CrewAI docs introduction, CrewAI AMP introduction, CrewAI RBAC, CrewAI AMP blog post, CrewAI MCP, CrewAI human in the loop, CrewAI testing, CrewAI tools (read 2026-10-06).

Dify is an open-source platform for building agents, agentic workflows and chatbots on your own data, available as a managed cloud, a community edition or an enterprise deployment.

Best for: Teams that want a visual workflow studio with an open-source self-hosted option, MCP in both directions and a human-input step, and can accept enterprise-only SSO and audit logs.

Build surface: Workflow Studio is a visual block-based builder. Apps can be published as web apps, APIs, embeds or MCP tools, and the docs mention the difyctl command-line tool and plugin development.[1][2]

Tools, integrations and MCP: Dify connects to MCP servers as a client (HTTP transport only) and can publish its apps as MCP servers. A Marketplace lists model providers, tools, data sources and MCP integrations, but no total count is stated.[1][2][3]

Multi-agent, workflows and evaluation: Workflows and Chatflows are documented, and an Agent can run as a workflow step. Multi-agent orchestration and an agent evaluation framework were not found on the pages read; the docs index shows retrieval testing and single-node runs.[1][2]

Governance and approvals: Cloud docs list four workspace roles (Owner, Admin, Editor, Normal). The enterprise page lists SSO via SAML or OIDC, fine-grained RBAC, tamper-evident audit logs streamed to a SIEM. A Human Input node pauses a workflow for human review.[1][2][3]

Deployment and data control: Dify Cloud is managed hosting. The free Community Edition is self-hosted and open source. The enterprise page lists a Helm chart for Kubernetes, Terraform modules for AWS, GCP and Azure, and air-gapped deployment. No Cloud region choice was found.[1][2]

Pricing transparency: Sandbox is free (200 message credits). Professional is $590 per workspace a year and Team is $1,590 per workspace a year, billed annually with a monthly toggle. Community Edition is free; Enterprise is custom via sales.[1]

Watch-outs:

• The Sandbox plan allows 1 team member and 5 apps; Professional allows 3 members.
• Dify's MCP client supports servers using HTTP transport only, per the docs.
• SSO and audit logs appear on the enterprise page, not on the self-serve cloud plans.
• Message credits are for trying hosted models; the pricing page says to switch to your own API key once they run out.

Sources: Dify pricing, Dify homepage, Dify enterprise, Dify docs introduction, Dify key concepts, Dify MCP, Dify Human Input node, Dify manage members, Dify cloud docs index, Dify self-host docs index, Dify Marketplace, Dify docs home (read 2026-10-06).

Dust describes itself as multiplayer AI for human-agent collaboration, connecting company knowledge, tools and teams to create, share and run agents.

Best for: Companies that want shared team agents over their own data, with SSO and a US or EU data location, and that can pay per seat.

Build surface: The homepage says teams can create, share and run agents and links a developer platform, JavaScript SDK and OpenAPI documentation. The pages read do not describe the agent-building interface itself.[1][2][3]

Tools, integrations and MCP: The homepage states 70+ connectors. Dust can expose a workspace as a remote MCP server and can add external MCP servers as tools; the pricing page lists 5 remote MCP servers on Business and unlimited on Enterprise.[1][2][3]

Multi-agent, workflows and evaluation: Agents can run other agents as a tool (maximum recursion depth of 4), and the pricing page lists multi-agent orchestration and triggers. No evaluation feature or separate workflow builder was found.[1][2][3]

Governance and approvals: SSO is on Business and Enterprise (SCIM is Enterprise). Roles are Admin, Manager and Member with groups and spaces. Audit logs are Enterprise only. An API endpoint approves or rejects an agent action in a conversation.[1][2][3][4]

Deployment and data control: Data can be hosted in the EU or the US, on Business and Enterprise. A single-tenant deployment option is listed under Enterprise. No self-hosting option was found.[1][2]

Pricing transparency: Business: Free at 0 euros with 500 lifetime credits; Pro at 24 euros per seat per month on yearly billing (8,000 credits); Max at 120 euros per seat per month on yearly billing (40,000 credits). Enterprise: talk to sales.[1]

Watch-outs:

• Audit logs, SCIM and single-tenant deployment are Enterprise only.
• Business includes up to 3 data-source connectors and 5 remote MCP servers.
• The pricing page states Business is for teams of up to 100 people.
• The Free tier gives 500 credits for life, not per month.

Sources: Dust homepage, Dust pricing, Dust security, Dust docs home, Dust docs index, Dust audit logs, Dust roles and permissions, Dust run agent, Dust validate action API, Dust MCP server, Dust JavaScript SDK (read 2026-10-06).

Gumloop describes itself as a multiplayer AI agent builder where staff create agents from plain-language descriptions, with IT controlling access.

Best for: Company-wide teams that want non-developers to build agents on top of connected apps, with admin controls, approvals and an API or SDK for developers.

Build surface: Agents are created in a UI with four tabs (Agent, Triggers, Access, Settings) and guided setup, and can also be created and run through a REST API, Python and JavaScript SDKs and a CLI.[1][2][3]

Tools, integrations and MCP: The homepage states 100+ MCP integrations. Gumloop can connect custom, proxied and hosted MCP servers and is itself available as an MCP server for Claude, Cursor and VS Code.[1][2]

Multi-agent, workflows and evaluation: Agents can hand work to subagents (up to 10 concurrent per task). Evaluations grade agent interactions against set criteria. A separate multi-step non-agent workflow feature was not found in the docs index.[1][2]

Governance and approvals: Organisation roles need Pro or above; custom roles need Enterprise. SSO (SAML, OIDC, SCIM) needs Enterprise. Audit logging is documented, with no plan stated. Tool approval settings let a human approve or block agent actions.[1][2][3][4]

Deployment and data control: The homepage lists VPC deployments among its enterprise controls. No data-region choice or self-hosting option was found, and no docs page describes the VPC option.[1][2]

Pricing transparency: Docs list Pro at $37 per month with 20,000 credits, and Enterprise as custom. A credit is $0.005. Agent chats are billed in credits for model use, tool calls, compute time and an 8% orchestration fee. New accounts get a 14-day Pro trial.[1][2]

Watch-outs:

• SSO login page settings need an Enterprise subscription, and SCIM must be requested from support.
• Custom roles, Security and Analytics organisation roles, and organisation-level evaluations are Enterprise only.
• API keys need the Pro plan or above; Brain needs Pro or Enterprise.
• Using your own model key cuts model credits to zero but raises the orchestration fee from 8% to 16%.
• Pro overage is capped at 1,000,000 credits ($5,000) per billing period.

Sources: Gumloop homepage, Gumloop docs home, Gumloop docs index, Gumloop docs full text (first 100,000 characters only), Gumloop credits, Gumloop subscription management, Gumloop agents, Gumloop human in the loop, Gumloop audit logging, Gumloop user roles, Gumloop SSO (read 2026-10-06).

LangGraph is an open-source orchestration framework for stateful agents; LangSmith adds observability, evaluation, deployment, a Studio and a no-code builder called Fleet.

Best for: Engineering teams that want to write agent logic in code and need tracing, evaluation and a choice of cloud, hybrid or self-hosted deployment, with a no-code route for other teams.

Build surface: LangGraph is a code framework. LangSmith Studio is a visual interface for developing and testing agents locally, and LangSmith Fleet is a no-code platform for creating agents, with guidance on calling those agents from an app.[1][2][3]

Tools, integrations and MCP: LangChain's docs state 1000+ integrations across models, tools, loaders and vector stores. An MCP adapter turns a server's tools into LangChain tools for agents; Fleet also supports remote MCP connections.[1][2][3]

Multi-agent, workflows and evaluation: Docs describe multi-agent patterns (subagents, handoffs, router, custom workflow), LangGraph graphs mixing deterministic and agentic steps, and LangSmith offline and online evaluation with datasets and evaluators.[1][2][3]

Governance and approvals: The pricing page lists SSO (Google and GitHub on Developer and Plus, custom SSO on Enterprise) and organisation roles on Plus and Enterprise. Audit logs are Enterprise only, kept up to 400 days. LangGraph interrupts pause a run for human approval of tool calls.[1][2][3]

Deployment and data control: Docs list cloud (managed on AWS and GCP), self-hosted with a control plane, hybrid, and a standalone server. Cloud regions are GCP US, GCP EU, GCP APAC and AWS US, available on all plans.[1][2]

Pricing transparency: Developer is $0 per seat a month with up to 5k base traces; Plus is $39 per seat a month with up to 10k base traces; both then pay as you go. Enterprise is custom. Usage is metered in LangChain Standard Units, where 1 LSU is $1.[1]

Watch-outs:

• Audit logs are Enterprise only, according to the LangSmith audit-logs page.
• Organisation roles are limited to User and Admin on Plus; custom RBAC and ABAC are listed for Enterprise.
• Usage beyond included traces is pay-as-you-go in LSUs, so cost depends on consumption.

Sources: LangSmith pricing, LangGraph overview, LangSmith platform, LangSmith deployments, LangSmith regions FAQ, LangSmith audit logs, LangGraph interrupts, LangChain multi-agent, LangChain MCP, LangSmith evaluation, LangSmith Fleet, LangSmith Studio, LangChain integrations overview, LangSmith docs home, LangSmith Agent Server (read 2026-10-06).

Lindy presents itself as an AI teammate that connects to your tools, learns how you work and takes repeatable work off your plate.

Best for: Individuals and teams that want a Slack-based AI assistant for inbox, meetings and routine tasks, rather than a developer-oriented agent-building platform.

Build surface: Users work with Lindy through chat and Slack, saving repeatable work as Skills and scheduling Routines. The pages read describe no visual builder and no public API or SDK.[1][2][3]

Tools, integrations and MCP: Docs state 1,000+ tools (the pricing page says 1,500+), connected by OAuth or API key. Lindy can use any hosted MCP server a team runs. Where no integration exists it can drive a browser.[1][2][3]

Multi-agent, workflows and evaluation: Multi-agent orchestration, multi-step workflows and agent evaluation are not described on the pages read. Lindy offers scheduled Routines and reusable Skills.[1]

Governance and approvals: Workspace roles are Owner, Admin and Member. The pricing page lists SSO and SCIM and audit logs under Enterprise, and 'built-in approvals' on all plans.[1][2][3][4]

Deployment and data control: The security page states encryption in transit and at rest and no model training on data. It names no data region and no self-hosting option.[1]

Pricing transparency: Lindy's two own pages differ. lindy.ai/pricing shows Free ($50 credits, 7 days), Team from $29.99 per month for 3,000 credits, and custom Enterprise. The docs show Plus $30, Pro $100, Max $200 per user per month.[1][2]

Watch-outs:

• The marketing pricing page and the docs pricing page list different plan names and prices; check at purchase.
• SSO, SCIM, audit logs and a signed HIPAA BAA are Enterprise only.
• Monthly credits do not roll over; purchased top-ups do. Top-ups are $10 per 1,000 credits.
• Computer use is Pro, Max and Enterprise only on the docs pricing page, so not on the entry Plus tier.

Sources: Lindy homepage, Lindy pricing, Lindy docs home, Lindy docs index, Lindy docs pricing, Lindy docs security, Lindy docs integrations, Lindy docs team setup (read 2026-10-06).

Microsoft describes Copilot Studio as a graphical, low-code studio for building and managing AI agents and workflows, published to channels such as Teams and Microsoft Copilot.

Best for: Organisations already on Microsoft 365, Entra and Power Platform that want a low-code agent studio with Purview audit logging, 1,500+ connectors and MCP support.

Build surface: Copilot Studio is a graphical low-code studio for agents and drag-and-drop workflows. Agents can be reached from code through the Copilot Studio client using the Microsoft 365 Agents SDK (.NET, JavaScript, Python) or Direct Line.[1][2]

Tools, integrations and MCP: The Microsoft product page states 1500+ pre-built data connectors and connection to MCP servers. The docs say Copilot Studio currently supports MCP tools and resources, and that generative orchestration must be on.[1][2]

Multi-agent, workflows and evaluation: Docs describe child agents, connected agents and A2A connections for multi-agent work; workflows built in a drag-and-drop designer; and evaluations using test sets and a grader library.[1][2]

Governance and approvals: Agents can sign users in with Microsoft Entra ID and a documented SSO setup. The overview cites role-based access for administration. Audit events go to Microsoft Purview and Sentinel. Workflows include human-in-the-loop controls and agent flows have human review steps.[1][2][3][4]

Deployment and data control: The geographic data-residency page says organisations can choose where data is stored across Azure datacentres, and describes EU Data Boundary compliance. It is a Microsoft-hosted service; no self-hosting or on-premises option was found on the pages read.[1]

Pricing transparency: Copilot Credits are sold as 25,000-credit tenant packs at $200.00 per pack per month, or pay as you go through Azure with no upfront commitment. A pre-purchase plan saves up to 20%. Microsoft 365 Copilot is $30.00 per user per month, paid yearly.[1][2][3]

Watch-outs:

• Credits are enforced monthly and unused Copilot Credits do not carry over, per the licensing page.
• An Azure subscription is required for the options that need agents, per the pricing page.
• Connections to Foundry, Fabric and Microsoft 365 Agents SDK agents are public preview and not meant for production use.
• MCP needs generative orchestration turned on, and Copilot Studio supports MCP tools and resources only.

Sources: Copilot Studio product page, Copilot Studio pricing, Copilot Studio overview, Standard harness licensing, Copilot Studio MCP, Security and governance, Purview audit logs, User authentication, Geographic data residency, Add other agents, Integrate with Agents SDK, Use connectors (read 2026-10-06).

MindStudio offers a visual, no-code builder for AI agents that can be extended with code and deployed as API endpoints, web apps or MCP servers.

Best for: Teams that want to build and expose agents visually, call them from code through an API or npm package, and need self-hosting on a Business plan.

Build surface: The homepage describes a visual builder needing no coding but extensible with code. Agents can be run through an API and a Node.js npm package, and custom JavaScript or Python run-function blocks are documented.[1][2][3]

Tools, integrations and MCP: The homepage states 1,000+ pre-built integrations. Docs say agents can be exposed as an MCP server for external applications; whether agents can consume MCP servers as a client was not stated.[1][2]

Multi-agent, workflows and evaluation: Workflows are the core build unit, and an Evaluations feature runs test cases against workflows. Multi-agent orchestration was not confirmed on the pages read.[1][2]

Governance and approvals: Workspace roles are Owner, Admin, Member and Guest, with edit or use-only agent access. The pricing page lists SSO, audit logs and granular collaborator permissions under Business. A Checkpoint block asks a person to approve or reject a step.[1][2][3]

Deployment and data control: The Business plan lists 'Self-host MindStudio'. No data-region choice was found, and the trust centre page did not render.[1]

Pricing transparency: Free is $0 per month plus usage (one agent, 1,000 runs). Individual is $20 per month, or $16 per month billed yearly, plus usage. Business is custom. Model usage is billed per token with no markup.[1]

Watch-outs:

• SSO, audit logs, budgets and self-hosting are listed under the Business plan only, which has no published price.
• Individual and Free prices are 'plus usage', so total cost depends on model token use.
• The Free plan is limited to one agent and 1,000 runs per month.

Sources: MindStudio homepage, MindStudio pricing, MindStudio docs home, MindStudio docs index, MindStudio team settings, MindStudio MCP, MindStudio Checkpoint block, MindStudio Evaluations, MindStudio npm package (read 2026-10-06).

Agent Builder is a visual canvas for multi-step agent workflows on the OpenAI platform, with export to code for the Agents SDK. OpenAI's docs say it is deprecated.

Best for: Not a safe new choice: the docs schedule Agent Builder's shutdown for 30 November 2026. The Agents SDK and Responses API MCP tool remain for teams that code their own agents.

Build surface: Agent Builder is a visual canvas where nodes are linked by typed edges. The Code option lets you embed the workflow via ChatKit or copy code to run on your own infrastructure with the Agents SDK. Published versions can be targeted in API calls.[1]

Tools, integrations and MCP: Agent Builder has an MCP node, and the Responses API connects to remote MCP servers. Eight connectors are listed (Dropbox, Gmail, Google Calendar, Google Drive, Microsoft Teams, Outlook Calendar, Outlook Email, SharePoint), and connector_id is deprecated for models released after 1 September 2026.[1][2]

Multi-agent, workflows and evaluation: Agent Builder chains multi-step workflows with if/else and while nodes. The Evaluate option runs trace graders over selected traces. The Agents guide describes agents, tools and handoffs in the Agents SDK.[1][2][3]

Governance and approvals: A human approval node defers to the end user before the workflow continues, and a Guardrails node checks for PII, jailbreaks and hallucinations. SSO, role-based access and audit logs for the platform were not read on any page.[1]

Deployment and data control: Agent Builder runs on the OpenAI platform. Exported code can run on your own infrastructure with the Agents SDK. No data-region or hosting-option page was read.[1]

Pricing transparency: The API pricing page lists no separate Agent Builder fee and does not say either way. Related items shown: ChatKit file storage at $0.10 per GB-day after 1 GB free, web search at $10.00 per 1k calls, file search at $2.50 per 1k calls. Model tokens bill at model rates.[1]

Watch-outs:

• The Agent Builder guide says OpenAI is phasing it out, with shutdown scheduled for 30 November 2026; ChatKit stays available.
• The connector_id route is deprecated for models released after 1 September 2026, per the MCP guide.
• The Agent Builder guide flags prompt injection and data leakage risks and points to a separate safety guide.

Sources: Agent Builder guide, Agent Builder node reference, MCP and connectors guide, Agents guide, OpenAI API pricing (read 2026-10-06).

Relevance AI is a platform for building specialist AI agents and multi-agent 'workforces' that teams, including non-technical ones, can deploy.

Best for: Go-to-market and operations teams that want no-code agents and multi-agent workforces with evaluations, plus a JavaScript SDK, and that can accept a US, EU or AU region choice.

Build surface: Agents are built in a no-code Tools builder and a visual Workforce canvas. A JavaScript SDK starts conversations with agents and triggers workforces from code, and an API is documented.[1][2][3][4]

Tools, integrations and MCP: The pricing page lists 2,000+ integrations (the homepage says 1,000+ apps). An MCP and Plugins docs page describes MCP support for connecting agents to MCP-compatible clients.[1][2][3]

Multi-agent, workflows and evaluation: Workforces are multi-agent teams. Tools are multi-step automations with API, code and LLM steps. Evals test agents and workforces with simulated conversations and reusable scoring checks.[1][2][3]

Governance and approvals: The pricing page lists SSO (SAML), RBAC and audit logs under Enterprise. RBAC docs confirm Enterprise-only roles at organisation, project and asset level. Approvals and escalations let humans approve or reject agent actions.[1][2][3]

Deployment and data control: The security page offers three regions at signup: US (N. Virginia), EU (London), AU (Sydney), fixed once chosen. It says single-tenant options are in development and Enterprise customers get isolated service instances.[1]

Pricing transparency: Pro is $19 per month billed annually or $29 monthly (2,500 Actions per month). Team is $234 annually or $349 monthly (7,000 Actions per month). Enterprise is custom. Extra Actions are $80 per 1,000.[1][2]

Watch-outs:

• SSO (SAML), RBAC, audit logs and agent evaluations appear under Enterprise on the pricing page, but the evals docs say evals are on every plan.
• RBAC is being rolled out gradually to Enterprise customers.
• The data region is permanent once an account is created.
• Plan Actions reset at each renewal; Vendor Credits roll over while subscribed.

Sources: Relevance AI homepage, Relevance AI pricing, Relevance AI docs home, Relevance AI docs index, Relevance AI docs pricing, Relevance AI docs security, Relevance AI docs RBAC, Relevance AI docs approvals, Relevance AI docs evals, Relevance AI docs MCP, Relevance AI docs tools, Relevance AI docs SDK (read 2026-10-06).

StackAI calls itself an Enterprise AI Agent Factory: a drag-and-drop no-code platform for building and governing agents, with multi-tenant, VPC and on-premise deployment.

Best for: IT-led enterprise teams that want a no-code builder with a REST API and VPC, on-premise or air-gapped deployment, and who accept contact-sales pricing beyond a small free tier.

Build surface: The homepage describes a drag-and-drop no-code platform. The API docs list endpoints to create and manage runs, and a legacy endpoint to run a flow by ID. No SDK is mentioned.[1][2]

Tools, integrations and MCP: The homepage states 300+ enterprise integrations. StackAI has an MCP section where StackAI is exposed as an MCP server to outside assistants; nothing read describes agents calling third-party MCP servers as tools.[1][2]

Multi-agent, workflows and evaluation: Docs list a Workflow Builder guide and an Evaluations and Observability guide; the API lists evaluation datasets, evaluators and experiments. Multi-agent orchestration was not found on the pages read.[1][2]

Governance and approvals: The security page describes custom SSO over SAML with role mapping; the pricing page lists SSO and access control for Enterprise. The homepage lists audit logs and human review at critical decision points. Named roles were not found.[1][2][3][4]

Deployment and data control: The homepage lists multi-tenant, VPC and on-premise deployment. The security page mentions four deployment models up to air-gapped on-premise. The pricing page lists on-prem and VPC under Enterprise. No region choice was found.[1][2][3]

Pricing transparency: Free is $0 a month with 500 runs a month, 2 projects and 1 seat. Enterprise is custom per month with custom runs and seats. No mid-tier paid plan is shown.[1]

Watch-outs:

• SSO, on-premise and VPC deployment are listed under Enterprise on the pricing page, not on the Free plan.
• The Free plan has 500 runs a month, 2 projects and 1 seat.
• The homepage carries a banner saying Asana has acquired StackAI, so product and pricing may change.

Sources: StackAI pricing, StackAI homepage, StackAI security, StackAI docs home, StackAI API docs, StackAI MCP docs (read 2026-10-06).