Contents
- Verdict
- What is Stack AI?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with Stack AI
- What Stack AI actually costs - and why we cannot tell you
- The gotcha: there is no middle tier
- What breaks in production, and what is not published
- What it connects to that you already run
- Stack AI alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
Stack AI has moved firmly up-market and its pricing page now reflects that. As of August 2026 it publishes exactly two options: a free tier with 500 runs a month, one seat and two projects, and a custom-quoted Enterprise plan with dedicated infrastructure, on-premises or VPC deployment, SSO and SOC 2, HIPAA and GDPR compliance. There is no published self-serve paid tier in between. For a small business that means the honest answer is usually no: you can prototype for free, but the moment you outgrow 500 runs you are in a sales conversation sized for organisations with compliance officers. Where it genuinely fits is a regulated mid-market team that needs an LLM app inside its own network.
What is Stack AI?
Stack AI is a no-code platform for building LLM-powered applications and agents, including document assistants, chatbots, and internal tools. It provides a visual builder to connect models, data sources, and logic, with an emphasis on enterprise use cases. It is oriented toward teams that want to ship AI apps and agents without a full engineering effort.
Best for
Teams building LLM apps, document assistants, and internal AI tools without heavy engineering.
Not for
Solo operators and small teams on a self-serve budget. With no published paid tier between a 500-run free plan and an enterprise contract, there is no priced path for a business that needs more than a prototype but less than a compliance programme.
Strengths
- Visual builder aimed squarely at LLM applications and agents rather than app-to-app plumbing
- Connects models to your own data sources for retrieval-style workflows without writing a RAG pipeline by hand
- On-premises and VPC deployment on Enterprise, which very few no-code AI builders offer at all
- SOC 2, HIPAA and GDPR compliance listed on the Enterprise tier, which is the reason regulated buyers shortlist it
- Genuinely useful free tier at 500 runs a month for proving a concept before any spend
- Solution engineers included on Enterprise, which meaningfully shortens the build for a team without AI engineers
Limitations
- No published paid tier between free and Enterprise, so there is no self-serve upgrade path
- Enterprise pricing is not published at all, which makes budgeting impossible without a sales call
- 500 runs a month on the free tier is a prototype allowance, not an operating one
- Far fewer general-purpose connectors than mainstream automation tools, so it is a poor general automation platform
- One seat and two projects on free means you cannot even trial it as a team
- Procurement is a sales cycle, not a signup, which is weeks rather than an afternoon
Stack AI pricing
Stack AI publishes a free tier and a custom-quoted Enterprise tier only, with no self-serve paid plan in between as of August 2026. The free tier includes 500 runs a month, one seat, two projects and community support. Enterprise pricing is not published: the page lists custom run volume, custom seats, unlimited projects, dedicated infrastructure, on-premises and VPC deployment, SSO, access control and SOC 2, HIPAA and GDPR compliance, with pricing available only on request. Because the metered unit is runs rather than steps or seats, cost scales with usage volume rather than workflow complexity - but without a published rate, you cannot model that in advance.
| Plan | Price | What you get |
|---|---|---|
| Free | $0/mo | 500 runs/mo, 1 seat, 2 projects, community Discord support (as of Aug 2026) |
| Self-serve paid tier | None published | No plan of any kind between the 500-run free tier and a custom Enterprise quote as of Aug 2026 |
| Enterprise | Custom | Custom runs and seats, unlimited projects, dedicated infrastructure, on-prem and VPC deployment, solution engineers, SSO, access control, SOC 2 / HIPAA / GDPR |
| Alternatives, for reference | $20-$50/mo | The same 1,000-run workload costs about $49/mo on Dify Professional, $35/mo on Flowise Starter or EUR 20/mo on n8n Starter - the floor to judge a quote against |
Pricing reflects public plans as of August 6, 2026 and can change. Check Stack AI for the latest.
What people actually build with Stack AI
Stack AI is bought for a specific reason and it is rarely the features. It is bought when an LLM application has to run inside a boundary someone has promised a client or a regulator, and the buyer needs a contract and an attestation rather than a repository.
The builds that fit
An internal policy and procedure assistant inside a VPC. Staff ask questions of a document corpus that legally cannot leave the organisation's network, the model call is routed within the deployment boundary, and the SOC 2, HIPAA and GDPR attestations on the Enterprise tier are what make the project approvable at all.
A claims or case triage assistant that reads a submitted document, extracts the fields, checks them against internal criteria and routes the file with a recommendation. The deployed flow is a production API behind an organisation bearer token, so an existing system calls it rather than staff learning a new tool.
A knowledge tool for a professional services team over precedent documents, engagement letters or clinical protocols, where the answer must cite its source because someone will be asked to justify it.
A prototype on the free tier to prove a concept. 500 runs a month, one seat and two projects is enough to build the thing, point it at real documents and find out whether retrieval quality is good enough before any money moves. Use it exactly that way and it is genuinely useful.
Our reference workload of 1,000 triage items a month is, notably, not on this list, because it is twice the free allowance and there is no published plan to grow into.
What Stack AI is genuinely bad at
Being bought by a small business on a credit card. There is no self-serve paid tier, so the path from prototype to production runs through a sales cycle sized for organisations with a compliance function. That is a deliberate design, not an oversight, and it excludes a large share of the people who will read this page.
It is also poor at general app-to-app automation. Connecting a payment failure to a CRM update and a Slack message is a job for an automation platform, and Stack AI has a fraction of the connectors for it.
Bottom line: Stack AI is bought for on-premises deployment and compliance attestations, not for features, and at 1,000 runs a month you are past the free tier with no published plan to move to.
What Stack AI actually costs - and why we cannot tell you
We are going to be direct about the limits of this section, because pretending otherwise would be worse than useless. As of August 2026 Stack AI publishes exactly two things on its pricing page: a free tier at $0 with 500 runs a month, one seat, two projects and community Discord support, and an Enterprise tier whose price is 'custom' with custom runs, custom seats and unlimited projects. There is no self-serve paid plan in between and no published rate for anything. That means we cannot tell you what Stack AI costs at small-business volume, and neither can anybody else who has not been through its sales process. Any figure you see quoted elsewhere is either someone's specific negotiated contract or an invention.
What we can tell you is the shape of the meter, which still helps. The unit is runs, not steps and not seats, so cost scales with how often your workflow executes rather than how complicated you made it. That is structurally the friendlier of the two common models - the same distinction we lay out in execution vs task vs operation - and it means a Stack AI flow with fifteen nodes costs the same per run as one with three. Without a published per-run rate, though, you cannot convert that into a monthly number in advance.
The practical consequence for a small business: the free tier's 500 runs a month is a prototype allowance. At our reference volume of 1,000 triage items a month you are double it in month one, and the only available next step is a sales conversation scoped for organisations with a compliance function. If your budget for this is a credit card and a Tuesday afternoon, Stack AI is not currently sold to you.
| Option | Published price at 1,000 runs a month | Self-host possible? | Compliance position |
|---|---|---|---|
| Stack AI free | $0, but capped at 500 runs so this workload does not fit | Enterprise tier only, contractually | SOC 2, HIPAA, GDPR listed on Enterprise |
| Stack AI Enterprise | Not published; quote only | Yes, on-premises and VPC | The reason to buy it |
| Dify Professional | $590/yr, about $49/mo | Yes, but not multi-tenant without written authorisation | Yours to attest to |
| Flowise Starter | $35/mo | Yes, Apache 2.0 core | Yours to attest to |
| n8n Starter | EUR 20/mo for 2,500 executions | Yes, source-available community edition | Yours to attest to |
| Windmill free edition | $0, executions unlimited | Yes, AGPLv3 core | Yours to attest to |
The comparison Stack AI does not want you to run
Because the price is unpublished, the useful exercise is not what Stack AI costs but what the alternatives cost for the same 1,000 runs, so you know what number would make the quote reasonable. Dify's Professional plan is $590 a year, about $49 a month, and includes 5,000 message credits. Flowise Starter is $35 a month for 10,000 predictions. n8n Starter is EUR 20 a month for 2,500 executions. Both Dify and Flowise can also be self-hosted for the cost of a server, which is $10 to $60 a month depending on how much you index.
So the whole of the rest of this category prices the reference workload between roughly $20 and $50 a month. When the Stack AI quote arrives, that is the band it should be measured against, with an honest premium for the on-premises deployment, the solution engineers and the compliance attestations, which are real things that cost real money to provide. What you are buying is assurance, and assurance is worth paying for. Knowing the software floor is what stops you overpaying for it.
Bottom line: Everything comparable prices this workload between roughly $20 and $50 a month, so get the Stack AI quote before you build and judge the premium against that floor.
The gotcha: there is no middle tier
The surprise with Stack AI is not a hidden unit or a metering quirk - it is a missing rung on the ladder. Most platforms in this directory let you grow through a series of self-serve plans and only reach sales when you are genuinely large. Stack AI has removed that path. You prototype free, and then you negotiate. There is nothing wrong with that as a business model, and it is increasingly common among AI platforms selling into regulated buyers, but it changes what building on the free tier means. You are not starting small on a product you can scale into; you are building a prototype whose production version has an unknown price.
Plan for that explicitly. Before you invest weeks in a Stack AI build, get the Enterprise quote first, because the answer determines whether the prototype has anywhere to go. The reverse order - build, then discover the price - is how teams end up rebuilding on a different platform having spent the budget on the wrong one.
The related constraint is that the free tier gives you one seat and two projects. You cannot trial it as a team, and you cannot run a real evaluation across several candidate use cases at once. That is a deliberate funnel design, not an oversight, and it is worth knowing before you plan an evaluation.
Bottom line: Get the Enterprise quote before you build, not after, because the free tier is a funnel into a negotiation rather than a plan you can grow inside.
What breaks in production, and what is not published
Honest position: Stack AI does not publish the operational numbers that decide whether a platform survives production, and we are not going to guess them. Its API documentation states that the deployed API supports auto-scaling for a large volume of requests, but gives no concurrency ceiling, no requests-per-minute figure and no throttling behaviour. That is a vendor claim about elasticity, not a specification you can design against. Ask for the actual limits in writing during the sales conversation and get them into the contract, because 'auto-scaling' is not a number and you cannot build a retry policy around an adjective.
Similarly, we found no published data-retention policy for run logs or uploaded documents at either tier, and no published incident or status history to assess reliability from. Where a vendor publishes a status page, we would rather cite its actual incidents than generalise; here there was nothing public to cite as of August 2026. Confirm both in the live documentation and in your contract before you put customer data through it.
It is also worth naming what you cannot check here that you can check elsewhere in this category. For Dify, Flowise and Windmill you can read the repository and see the last commit date, the open issue count and the release notes for yourself: on 6 August 2026 those three had pushed code that day or the day before. Stack AI is closed, so project health is not something you can verify independently at all. That is not an accusation, it is a difference in what due diligence is available to you, and it should raise the weight you put on contractual commitments.
What is documented and does matter: the deployment path. A finished flow is deployed as a production API, protected by your organisation's bearer token, with the public API key issued from organisation settings. That is a conventional and sensible model. Rotating that token is your responsibility, and because it is organisation-scoped rather than per-flow, treat it with the care that implies.
The predictable failure mode for a platform of this shape is the one that afflicts every retrieval-backed LLM app: the documents behind it go stale, and the assistant confidently answers from last quarter's policy. That is not a Stack AI defect - it is the category's defining operational problem - but it means whoever owns the build also owns a re-indexing schedule and a routine for spot-checking answers against source documents.
Bottom line: Stack AI publishes no concurrency ceiling, no retention policy and no incident history, and being closed source means you cannot check project health yourself, so put the limits in the contract.
What it connects to that you already run
Stack AI is built for LLM applications rather than app-to-app plumbing, and its integration story reflects that. The connectors that matter are document and knowledge sources - the places your policies, contracts and manuals actually live - rather than a long catalogue of SaaS triggers. If your goal is to move a record from a form into a CRM, this is emphatically the wrong tool and a general automation platform will do it in an afternoon for a fraction of the effort.
If your goal is an internal assistant that answers questions from your own documents, the relevant question is whether it reads your repositories and writes to your systems, and the honest advice is to check that against your specific stack rather than trust a logo wall. The generic escape hatch exists as it does everywhere: the platform can call external APIs, so anything with a REST endpoint is reachable with a developer's involvement.
The more useful integration fact for a regulated buyer is the deployment side rather than the connector side. Stack AI markets on-premises and VPC deployment on the Enterprise tier, alongside SSO, access control and SOC 2, HIPAA and GDPR compliance. That combination - a no-code LLM app builder that can run inside your own network - is genuinely rare, and it is the actual reason this product gets shortlisted.
Bottom line: The integration that matters here is the deployment target, not the connector list: on-premises and VPC hosting is the feature people are actually buying.
Stack AI alternatives: what to pick instead, and when
Almost everyone who lands on a Stack AI pricing page and leaves disappointed wants one of two things it does not sell them: a priced plan, or an automation platform. Both are easy to fix.
If you wanted the data to stay on your infrastructure and you have someone technical, pick Dify and self-host it, or Flowise if the build is lighter and a developer owns it. Both give you the same physical guarantee - the documents never leave your network - for the cost of a server rather than a contract, and you can read their licences yourself. Read Dify's carefully, because its multi-tenant restriction matters if you are an agency. If the workload is really automation rather than an application, pick n8n, which self-hosts, has hundreds of connectors and prices per execution, or Activepieces for the same shape under an MIT licence. If the reasoning step is genuinely the product and you want it managed with published prices, Gumloop publishes its rates in a way Stack AI does not.
Where you should still pick Stack AI: when a named person has to be contractually accountable for the compliance posture, when solution engineers are part of the value, and when your procurement process expects a vendor rather than a repository. Those are real requirements, and a free self-hosted project cannot satisfy them at any price.
| Tool | Billing unit | Monthly cost at the reference workload | Who it suits | Does better than Stack AI |
|---|---|---|---|---|
| Stack AI | Run, at an unpublished rate above the free tier | Not published; the free tier stops at 500 runs | Regulated buyers needing a contract and attestations | On-premises deployment with compliance accountability |
| Dify | Message credit, per workspace per year | About $49/mo Professional, or $30-$60 of server self-hosted | Technical teams wanting the data to stay put | A published price and a licence you can read |
| Flowise | Prediction (one flow run) | $35/mo Starter, or roughly $10-$20 of server | Developers building a lighter assistant | Cheapest route to a self-hosted LLM app |
| n8n | Execution (whole workflow run) | EUR 20/mo Starter covers 2,500 executions | Businesses whose problem is automation, not apps | Connectors, scheduling and self-hosting together |
| Activepieces | Credit: 1 per flow run, steps free | $20/mo Plus, or free self-hosted with no run cap | Teams wanting a permissive licence | MIT licensing with no restrictions to read around |
| Gumloop | Credit, $0.005 each, published rate table | $37/mo Pro | Cloud-tolerant teams doing AI-heavy document work | A published rate card you can budget from |
Bottom line: If you wanted Stack AI for data control rather than a contract, a self-hosted Dify or Flowise gives you the same physical guarantee for the price of a server.
Migration and lock-in: can you get your work out?
Better than we expected, with a caveat. Stack AI documents exporting a project from the dashboard as a JSON file, described as the workflow, which can be saved locally and imported again. That is a real artefact and a real backup, and it puts Stack AI ahead of the several platforms in this directory that offer no export at all. Take those exports on a schedule rather than assuming the account will always be there.
The caveat is what the documentation does not say. It does not enumerate what the JSON excludes - whether credentials, uploaded knowledge bases, indexed vectors and deployment settings travel with it, or whether you are exporting the graph and leaving the substance behind. Based on how every comparable platform works, expect credentials and uploaded corpora to stay behind, but we could not confirm that from published documentation as of August 2026, so verify it by doing a test export and import into a second project before you rely on it.
The structural point is the one to hold on to, and it is the difference between this entry and the open-source-plus-cloud tools around it. There is no source code, no licence file and no community edition, so on-premises deployment is an Enterprise entitlement rather than an inherent property: the ability to run the software on your own hardware exists while you are paying for the tier that includes it, and ends when you stop. Compare that with Dify, Flowise or Windmill, where the right to keep running the software is a licence grant nobody can withdraw, subject only to the restrictions written into those licences. Because the Enterprise price is unpublished, your renewal leverage depends entirely on how hard the flow would be to rebuild elsewhere. Keeping the source documents, the prompts and the business rules in your own systems, rather than only inside Stack AI, is what preserves it.
Bottom line: Stack AI has no community edition, so on-premises deployment is a contractual entitlement that ends when you stop paying, unlike a licence grant nobody can withdraw.
Stack AI FAQ
How much does Stack AI cost?
As of August 2026 the published pricing page shows only two options: a free tier at $0 a month with 500 runs, one seat and two projects, and an Enterprise tier at custom pricing. No dollar figure is published for Enterprise, and there is no self-serve paid plan between the two. That means the practical answer for anyone budgeting is that you can evaluate it for nothing and you cannot forecast the production cost without contacting their sales team.
Is 500 free runs enough to run something real?
It is enough to prove a concept and not enough to operate. 500 runs a month is roughly 16 a day, which covers an internal document assistant a handful of people use occasionally. Anything customer-facing, anything triggered by inbound volume, or anything processing a queue will exhaust it in days. Treat the free tier honestly as an evaluation allowance: build the thing, confirm it works on your own documents, then have the pricing conversation with real usage numbers in hand.
Why would a business pick Stack AI over a cheaper tool?
Almost always compliance and deployment. The Enterprise tier lists on-premises and VPC deployment alongside SOC 2, HIPAA and GDPR compliance, which means the model workflow can run inside infrastructure you control. For a healthcare practice, a law firm or a financial services business that has promised clients their records do not leave a defined boundary, that is not a nice-to-have - it is the only thing that makes the project possible, and cheaper cloud-only tools are disqualified regardless of features.
Who is Stack AI wrong for?
Small businesses without a compliance driver, and anyone wanting general app-to-app automation. If you want to connect Shopify to QuickBooks and post to Slack, this is the wrong category of product entirely and has a fraction of the connectors. And if you are a five-person business that needs more than 500 runs but has no appetite for an enterprise procurement cycle, the pricing structure itself excludes you - there is simply no plan shaped for you.
What integrations does Stack AI offer?
Its strength is connecting models to data sources - document stores, databases and knowledge bases - for retrieval-style applications, rather than a broad catalogue of business SaaS connectors. If your requirement is 'let staff ask questions of our policy documents', that is exactly the shape it is built for. If your requirement is 'when a Stripe payment fails, update the CRM and notify the account manager', you want a general automation platform instead, and Stack AI will frustrate you.
How locked in would we be?
Structurally, quite. Applications built in the visual builder do not export to any other platform, so a move means rebuilding. The partial mitigation is that the expensive intellectual work in a RAG application is usually the prompts, the document preparation and the evaluation set, and all three are portable artefacts you can keep outside the platform. Insist on holding those independently from the start and a future migration is a rebuild of the wiring rather than the thinking.
Is Stack AI an agent platform or an app builder?
Primarily an app builder that also does agents. The centre of gravity is shipping an LLM-powered application - a document assistant, an internal chatbot, a knowledge tool - with a defined interface and users. Agent capabilities exist within that frame. If what you actually want is unattended work happening on a schedule with no human interface at all, a workflow automation platform is a more natural shape, and you will spend less time fighting the product's assumptions.
Stack AI or Dify for a self-hosted AI app?
Dify is open source and free to self-host, which makes it dramatically cheaper for a technical team that wants control - you own the deployment and pay nothing for the platform. Stack AI's on-premises option comes with an enterprise contract, support, solution engineers and compliance attestations attached. The question is whether you are buying software or buying assurance. Regulated buyers who need someone contractually accountable pick Stack AI; technical teams who just need the data to stay put pick Dify.
Can you build a document assistant for us instead?
Often that is the cheaper route, and we will say so plainly if it is. Much of what buyers approach enterprise AI app platforms for is achievable with a well-built retrieval workflow on infrastructure you already have. We scope it on a free automation audit, tell you honestly whether you need an enterprise platform or not, and if you do not, build it and hand it over with documentation you own plus a recurring report on what it did.
Looking at alternatives? Stack AI is a fit when the deliverable is an LLM application - a document assistant, an internal knowledge tool - and the buyer has compliance requirements that rule out ordinary cloud SaaS. That is a real and well-served niche, and the on-premises option is a genuine differentiator. It is not a fit for a small business that wants recurring tasks done, because there is no priced route from prototype to production. Autonoly is less about building an app and more about getting scheduled work done, so if you want agents running digital work rather than an AI product to ship, that is the closer match. See the Autonoly review.
