AI agent platforms let a company build assistants that read its own documents and apps and then act on them. They are billed in credits, runs or traces rather than plain seats.
This guide ranks thirteen on what the invoice becomes once agents run daily, where the platform can be hosted, and how far a team can build without engineers.
AuthorHannah ReiterSenior Analyst, Business Applications
Vendors can pay for visibility on this page. It never changes what an entry
says about a product, including the criticism, and we earn nothing when you click through to a
vendor. How that works.
In short
What AI agent platforms do
AI agent platforms provide the tooling to connect language models to company data and business apps, define what an agent may do, and run, monitor and govern those agents in production.
In this order: setup effort, what it really costs, how your data comes back out, whether
you can leave, and who each AI agent platform is built for. Why those five, and why there is no
score out of ten, is on the how we work page.
Prepaid credit packs or pay-as-you-go, published; needs an Azure subscription
—
Microsoft 365 organisations adding agents to Teams and SharePoint
Companies without Azure or whose data sits outside Microsoft
Country is where the vendor is headquartered or contracts from, which is a
different question from where your data is hosted. Where the two tell different stories, the
entry says so.
Company AI workspace with per-seat pricing and governance controls
Ranked #1 of 13 in Best AI Agent Platforms in 2026.
Self-hostablePublished pricingEurope
The buying question is how many people need agents versus how many need plain chat. Langdock prices the seat, with a Standard and a Max level, and treats workflows and API use as separate usage-based items with monthly run caps.
The 7-day trial carries a small credit allowance, so a pilot has to be planned. Its pricing page names managed, own-cloud and on-premises options, which matters if infrastructure policy is fixed.
What stands out
Per-seat price
Agent governance
On-prem option
Where it costs you
Workflow runs and steps are capped per tier, so heavy automation costs extra
The governance add-on becomes a paid per-user item after 1 January 2027
Right for
Companies that want a per-seat AI workspace with deployment choices
Wrong for
Teams wanting a free tier to run agents indefinitely
GermanyPer seat per month, published; 7-day trial; Enterprise quoted
Custom agents over company data, billed in credits per seat
Ranked #2 of 13 in Best AI Agent Platforms in 2026.
Free tierPublished pricingEurope
Dust treats the agent as a shared object: colleagues can browse each other's agents and see their instructions and tools, and several agents can work in one conversation. Pro and Max differ in how many credits a seat receives each month, and unused credits expire.
Business is limited to 100 people; larger groups move to Enterprise, which adds SCIM, audit logs and single-tenant deployment. The pricing page states US and EU data residency on every plan.
What stands out
Credits per seat
Data connectors
US or EU data
Where it costs you
Business has fewer connectors than Enterprise, which is unlimited
Monthly seat credits do not roll over, and the starter allowance is lifetime
Right for
Teams wanting agents grounded in connected company knowledge, paid per seat
Wrong for
Buyers needing a permanently free tier or self-hosted installs
FranceFree starter credits; Pro and Max per seat per month, published; Enterprise quoted
Activepieces is the least dependent on its vendor here, because the Community Edition is MIT-licensed and runs on your own Docker or Kubernetes setup.
On the cloud, credits scale with the model: a fast model step costs two, a smart one ten and a frontier one twenty, dropping to one if you bring your own provider key. The vendor's terms and privacy pages were unavailable when checked, so its country is not confirmed.
What stands out
Open source
Self-host or air-gapped
Credit based
Where it costs you
Enterprise features such as audit logs and SCIM need a commercial licence
AI steps cost more credits on frontier models than on fast ones
Right for
Technical teams that want to self-host automation and agents under MIT
Wrong for
Buyers who need a stated legal entity and country before signing
Not stated by the vendorFree plan with monthly credits; Plus and Team published; Enterprise quoted
Agent and workflow builder from LangGenius, in cloud or self-deployed
Ranked #4 of 13 in Best AI Agent Platforms in 2026.
Free tierPublished pricingNorth America
Dify offers three routes: the managed cloud, the Community Edition deployed with Docker, and Dify Enterprise with SSO, RBAC and support SLAs. The cloud plans scale by message credits, apps, members and knowledge storage, billed annually with monthly billing available.
The extra licence terms matter for agencies: running Dify as a multi-tenant service needs written permission, and the logo and copyright in the web console must stay in place.
What stands out
Self-deploy edition
Message credits
Workflow builder
Where it costs you
The licence is Apache 2.0 plus conditions on multi-tenant use and branding
Sandbox is one member and 200 message credits; Team caps members at 50
Right for
Developers who want a visual agent builder they can run themselves
Wrong for
Platform providers wanting to resell multi-tenant instances
United StatesFree Sandbox plan; Professional and Team published, billed annually; Enterprise quoted
Enterprise agent builder with a small hosted starter plan
Ranked #5 of 13 in Best AI Agent Platforms in 2026.
Free tierSelf-hostablePublished pricingNorth America
StackAI's pricing is two columns. The starter plan gives one seat, two projects and 500 runs per month with community Discord support. Enterprise raises every limit and adds dedicated engineers and on-prem or VPC options, plus SSO and role-based access.
Its terms state it does not train models on customer data. A team of five that wants a shared workspace has no listed price to budget against.
What stands out
Run based
On-prem or VPC
Enterprise focus
Where it costs you
No published tier between one free seat and a quoted Enterprise plan
On-prem and VPC deployment are Enterprise only; the free plan is cloud only
Right for
Larger organisations piloting agents before a quoted enterprise rollout
Wrong for
Small teams needing several seats at a listed price
United StatesFree plan with limited runs; Enterprise quoted
Agent workforce platform sold on a single enterprise plan
Ranked #6 of 13 in Best AI Agent Platforms in 2026.
Pricing on requestAsia-Pacific
Relevance AI is aimed at domain experts rather than engineers: agents are built without deep coding, then tested against quality benchmarks every version must pass before release. Approval gates hold high-stakes actions for human sign-off.
The pricing page offers unlimited agents, users and projects on its Enterprise plan with SSO, RBAC and audit logs, but no figures. Its enterprise page lists SOC 2 Type II and states it does not train on customer data.
What stands out
Enterprise only
Agent evaluations
1,000+ connectors
Where it costs you
Only an Enterprise plan is shown, with no prices and no mention of a trial
Terms allow hosting outside Australia unless agreed otherwise in writing
Right for
Operations teams that need evaluated, approval-gated agents at scale
Wrong for
Small teams wanting to start with a self-serve plan
AustraliaQuoted per organisation; Enterprise plan only on the pricing page
No-code assistants that work in email, meetings and Slack
Ranked #7 of 13 in Best AI Agent Platforms in 2026.
Free tierPublished pricingNorth America
Lindy bills for work done rather than seats: everyday tasks cost a few credits, deep work hundreds and complex builds up to a couple of thousand. Credits pool across the team, and top-ups are sold per thousand.
That makes monthly cost depend on how often agents run. The Team plan starts from a published price for a bundle of credits, and the legal terms identify Crivello Corp as the operator.
What stands out
Pooled credits
No-code
HIPAA on Enterprise
Where it costs you
Starter credits last seven days, so the free plan cannot run agents long term
HIPAA BAA, SSO and SCIM are limited to the quoted Enterprise plan
Right for
Small teams automating inbox, meeting and Slack tasks without code
Wrong for
Regulated teams on a self-serve plan needing HIPAA coverage
United StatesFree plan with credits; Team plan with pooled credits, published; Enterprise quoted
Agent builder that passes model costs through at provider rates
Ranked #8 of 13 in Best AI Agent Platforms in 2026.
Free tierSelf-hostablePublished pricingNorth America
MindStudio separates the platform fee from model usage. The Free plan is a plan fee of nothing plus usage, Individual is a monthly fee plus usage, and the page says model tokens are passed through at the same rates as the provider's own API.
Spend limits can be set per agent, which helps a team find a runaway workflow. Business is a custom plan that lists self-hosting options, SSO, custom SLAs and granular permissions.
What stands out
No token markup
Spend limits
Self-host option
Where it costs you
The Free plan allows one agent and 1,000 runs a month
Team features such as SSO and self-hosting need a quoted Business plan
Right for
Individuals and small teams who want model costs at provider rates
Wrong for
Teams needing published pricing for team-wide permissions or SSO
United StatesFree plan plus usage; Individual plan published; Business quoted
Multi-agent platform with a visual editor and a starter plan
Ranked #9 of 13 in Best AI Agent Platforms in 2026.
Free tierPublished pricingNorth America
CrewAI's page frames two choices. Basic gives the visual editor, an AI copilot and GitHub integration with 50 executions monthly, enough to prototype.
Enterprise adds governance (SSO, RBAC, workload identity, PII redaction), a choice of CrewAI cloud, private VPC or self-hosted infrastructure, a 45-day onboarding and forward-deployed engineering. Its terms name CrewAI, Inc. as a Delaware corporation under New York law; no street address was found.
What stands out
Visual editor
Private VPC
Execution limits
Where it costs you
Basic is limited to 50 workflow executions per month
Nothing is listed between Basic and the quoted Enterprise plan
Right for
Teams building multi-agent workflows who may later need a private VPC
Wrong for
Buyers wanting a published team price between free and enterprise
United StatesFree Basic plan with limited executions; Enterprise quoted
Builder for chat and voice agents, aimed at agencies and businesses
Ranked #10 of 13 in Best AI Agent Platforms in 2026.
Pricing on requestNorth America
Voiceflow builds conversational agents, deployed across chat and voice with real-time analytics and role-based permissions. Agencies see usage-based billing and multi-client workspaces with white-labelling; businesses see implementation support, either self-serve or fully managed, and a demo request.
It connects to several model providers. Its terms give Voiceflow, Inc. of Toronto, under Ontario law, so the contract is Canadian.
What stands out
Voice and chat
Agency billing
Multi-model
Where it costs you
The pricing page lists no tiers, seat counts or credit allowances
Businesses must book a demo to get a price
Right for
Agencies and support teams building chat and voice agents for clients
Wrong for
Buyers wanting to cost an internal knowledge agent from the page
CanadaUsage-based billing for agencies; quoted for businesses; trial without card
LangChain's platform for tracing, evaluating and deploying agents
Ranked #11 of 13 in Best AI Agent Platforms in 2026.
Free tierSelf-hostablePublished pricingNorth America
LangSmith sits beside the code rather than replacing it. The Developer plan costs nothing for one seat and a base trace allowance, Plus adds unlimited seats and a serverless deployment, and usage beyond that is billed in units worth one dollar.
Trace retention is 14 days at base. Self-hosted LangSmith keeps data in your VPC, runs on Kubernetes with ClickHouse, PostgreSQL and Redis, and is sold only as an Enterprise add-on.
What stands out
Developer tooling
Trace based
Self-hosted add-on
Where it costs you
Built for developers: you write the agent code, it does not build agents for you
Self-hosting is an Enterprise add-on needing Kubernetes and a sales licence key
Right for
Engineering teams that need traces, evaluation and deployment for their agents
Wrong for
Business teams that want to build agents without writing code
United StatesFree Developer plan; Plus per seat per month, published; Enterprise quoted
Company search and agents across 275 connected work apps
Ranked #12 of 13 in Best AI Agent Platforms in 2026.
Pricing on requestNorth America
Glean starts as search over a company's connected tools and layers agents and workflow automation on that index, which makes it a data-first buy rather than a builder-first one. Access follows existing permissions, and the vendor lists more than forty models.
Its pricing page shows no prices and offers a demo. Certifications listed include SOC 2 Type II, ISO 42001, HIPAA and GDPR, which a security review can check against its trust documents.
What stands out
Enterprise search
Permission aware
Sold by demo
Where it costs you
No prices are shown, so cost is unknown before a demo
Value depends on connecting many of the 275-plus supported work apps
Right for
Large organisations wanting search and agents across many connected apps
Wrong for
Small teams wanting to test a plan without a sales process
United StatesQuoted per organisation; demo on request
Microsoft's agent builder, metered in credits on an Azure subscription
Ranked #13 of 13 in Best AI Agent Platforms in 2026.
Published pricingNorth America
Copilot Studio is the cheapest route in for a company that already runs Microsoft 365, since agents reach Teams, SharePoint and OneDrive through built-in connections, plus more than 1,500 connectors and MCP servers.
Billing is either prepaid credit packs, with automatic pay-as-you-go when credits run out, or pure pay-as-you-go. Both sit on an Azure subscription, so the contract is with Microsoft Corporation and the exit runs through Azure.
What stands out
Microsoft 365 data
Credit packs
Azure required
Where it costs you
Both purchase routes require an active Azure subscription
Pay-as-you-go is consumption based, so monthly spend varies with use
Right for
Microsoft 365 organisations adding agents to Teams and SharePoint
Wrong for
Companies without Azure or whose data sits outside Microsoft
United StatesPrepaid credit packs or pay-as-you-go, published; needs an Azure subscription
AI agent platforms provide the tooling to connect language models to company data and business apps, define what an agent may do, and run, monitor and govern those agents in production. The differences that matter are rarely in the feature list, so this is
the order we would work through them.
01
Decide whether you need a published price
10 of the 13 tools here publish what they cost; the other 3 quote per organisation. The ones you can compare without a sales call: Langdock, Dust, Activepieces, Dify, StackAI, Lindy, MindStudio, CrewAI, LangSmith, Microsoft Copilot Studio.
02
Decide how much the jurisdiction matters
These 13 vendors are established in 6 countries across 4 regions (North America 9, Europe 2, Asia-Pacific 1, Elsewhere 1). That decides whose disclosure law applies to what the vendor holds, wherever the servers are.
03
Consider whether you want the source
1 of these are open source: Activepieces. Hosting one yourself trades a subscription for maintenance.
Credits, runs, traces or seats: what an AI agent platform bills
Agent platforms almost never charge a flat seat price, and the unit decides the invoice. Dust gives each seat a monthly credit allowance that expires unused, while Lindy pools credits across the team and sells top-ups per thousand. Activepieces charges more credits for a frontier-model step than a fast one and cuts the charge to one credit if you bring your own provider key.
Microsoft Copilot Studio sells prepaid credit packs on an Azure subscription, and LangSmith bills traces and then units worth one dollar. MindStudio passes model tokens through at provider rates. Relevance AI and Glean publish no prices at all. A pilot that looks cheap at ten runs a day can look different at ten thousand.
Ask each vendor what one real task costs in credits, runs or traces, and test it.
Check whether unused credits roll over; Dust's allowance does not.
Find out what happens at the limit: a stop, an overage rate or a plan jump.
Connecting an AI agent platform to company data
An agent is only as useful as the systems it can read. Glean starts from search across more than 275 connected work apps, and Dust limits connectors on Business while Enterprise is unlimited. Relevance AI lists more than a thousand native connectors plus MCP, Microsoft Copilot Studio lists over 1,500 connectors plus Teams, SharePoint and OneDrive, and Activepieces counts 760-plus apps.
A big connector count does not prove the one system you depend on is covered, nor how permissions carry over. Read the connector for your own CRM, ticketing and file store, and check whether the agent sees only what the asking user may see.
List the five systems the agent must read and check each connector by name.
Test with a restricted user to see whether permissions carry over.
Confirm that a missing connector can be filled with MCP or an API call.
Where AI agents run: cloud, VPC or your own servers
Hosting choices differ more than the marketing suggests. StackAI's free plan is cloud only and on-prem or VPC arrives with Enterprise, which is also where CrewAI offers a private VPC or self-hosted option. LangSmith self-hosting is an Enterprise add-on on Kubernetes with a licence key.
Activepieces and Dify can be run on your own servers, the first under MIT and the second under a modified Apache licence with a multi-tenant restriction. Langdock lists managed, own-cloud and on-premises options, and Dust's pricing page says every plan can choose US or EU data residency. Decide the hosting rule before the demo, not after it.
Ask which plan unlocks private hosting and whether that plan is quoted.
Read any licence conditions before self-deploying, as Dify's has them.
Request the data-residency option in writing for your plan.
Who builds the agents: engineers or business teams
Platforms split by who is expected to hold the keyboard. LangSmith assumes engineers writing agent code, with traces and evaluation around it. Dify and CrewAI offer visual builders. Lindy describes a no-code setup, MindStudio stresses ease of use, and Voiceflow is for teams building chat and voice agents.
Relevance AI says domain experts build while evaluations gate each version before release. The mismatch is the usual failure: a business team handed a developer tool stalls, and engineers handed a no-code tool hit its limits. Pick for the people who will maintain the agent in six months.
Name the person who will own each agent after launch.
Check for evaluation or approval controls before agents act on live systems.
Have the person who will build run the same task in two platforms.
What goes wrong most often when buying AI agent platforms
Costing a pilot at seat prices. Dust, Lindy and Activepieces all meter work in credits, so the bill follows how often agents run.
Assuming self-hosting comes with every plan. At StackAI, CrewAI and LangSmith it needs Enterprise, and at Dify the licence limits multi-tenant use.
Buying on connector counts. Glean, Relevance AI and Microsoft Copilot Studio each list large numbers, but only your own systems matter.
Starting a trial with no plan for it. Lindy's starter credits last seven days and Langdock's trial is seven days with a small credit allowance.
07
Frequently asked questions
8 answers
What are the best AI agent platforms in 2026?
Langdock leads our ranking of 13. Langdock sells a chat and agent workspace per user, with the Standard seat and a higher-usage Max seat both published, plus usage-based add-ons for workflows and API calls.
Its pricing page lists managed, own-cloud and on-premises deployment. The catch: workflow runs are capped per tier, and the agent governance add-on stops being free after 1 January 2027.
Which AI agent platforms publish their pricing?
10 of the 13, with the pricing model each one publishes:
Langdock: Per seat per month, published; 7-day trial; Enterprise quoted.
Dust: Free starter credits; Pro and Max per seat per month, published; Enterprise quoted.
Activepieces: Free plan with monthly credits; Plus and Team published; Enterprise quoted.
Dify: Free Sandbox plan; Professional and Team published, billed annually; Enterprise quoted.
StackAI: Free plan with limited runs; Enterprise quoted.
Lindy: Free plan with credits; Team plan with pooled credits, published; Enterprise quoted.
MindStudio: Free plan plus usage; Individual plan published; Business quoted.
CrewAI: Free Basic plan with limited executions; Enterprise quoted.
LangSmith: Free Developer plan; Plus per seat per month, published; Enterprise quoted.
Microsoft Copilot Studio: Prepaid credit packs or pay-as-you-go, published; needs an Azure subscription.
The other 3 quote per organisation.
Is there a free AI agent platform?
Dust, Activepieces, Dify, StackAI, Lindy, MindStudio, CrewAI, LangSmith offer a free tier or a free self-hosted edition.
Where are these AI agent platform vendors established?
In 6 countries across 4 regions: North America 9, Europe 2, Asia-Pacific 1, Elsewhere 1.
Langdock: Germany.
Dust: France.
Activepieces: Not stated by the vendor.
Dify: United States.
StackAI: United States.
Relevance AI: Australia.
Lindy: United States.
MindStudio: United States.
CrewAI: United States.
Voiceflow: Canada.
LangSmith: United States.
Glean: United States.
Microsoft Copilot Studio: United States.
Which AI agent platforms are open source?
Activepieces.
Which AI agent platforms can you host yourself?
Langdock, Activepieces, StackAI, MindStudio, LangSmith. The other 8 are hosted by the vendor only.
What should you use instead of Langdock?
Dust and Activepieces are the next two on this page. Dust is for teams wanting agents grounded in connected company knowledge, paid per seat; Activepieces is for Technical teams that want to self-host automation and agents under MIT.
Who should not buy Langdock?
Teams wanting a free tier to run agents indefinitely. Workflow runs and steps are capped per tier, so heavy automation costs extra.
If your AI agent platforms product belongs among these 13, tell us what it does and who it is for. Inclusion is an editorial call; what a listing is and is not is set out under software advice.
