Open-source AI agent platform pricing: what you actually pay

What you are actually paying for

Every AI agent platform bill has three parts. The software is what you pay to use the product, usually per seat. The hosting is the compute your agents run on, billed as credits, executions, or your own server bill. The model spend is the token cost of the AI itself, billed by the model provider or rolled into credits.

Vendors bundle these differently, and the bundle is where cost surprises live. A per-seat platform charges for people whether they use agents all day or once a week. An execution-priced platform charges each time a workflow runs, so the bill follows volume rather than headcount. An open-source platform unbundles all three: run the software on your own machines, buy compute at cost, and pay model providers directly with your own API keys.

Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own, with the agents working on real cloud computers. It ships in both shapes at once: managed cloud with per-seat pricing, and a self-hosted install with no software bill. (The Kortix overview covers the platform itself if you want the features before the finances.)

Kortix pricing, as written

Kortix's plans, as written on Kortix pricing in October 2026:

• Free, $0. 200 credits per month for sandbox compute, 1 project, bring your own API key (BYOK) for any premium model, connect your ChatGPT subscription.
• Team, $40 per seat per month. Everything in Free, plus 2,500 credits per month per seat, pooled (about 125 default Agent Computer hours when used only for compute). Optional managed models use pooled credits. BYOK and ChatGPT subscription still supported. Up to 200 projects and 100 seats, top up credits anytime, support via email.
• Enterprise, custom. SAML SSO and SCIM directory sync, advanced RBAC and audit logs, cloud, VPC, or on-prem deployment, plus SLA, DPA and dedicated support.
• Self-host, free. The same platform on a laptop, a VPS, your own VPC, or your own on-prem network. No seats and no credits; you pay your own infrastructure and your own model bills.

The model spend sits outside every plan if you want it to. Any model provider with your own keys. Or the ChatGPT plan you already pay for. The provider invoices you directly, so the platform subscription and the token bill stay separable. Only the optional managed models draw on pooled credits.

Where self-hosting changes the math

Self-hosting removes the software bill and turns hosting into your own line item. The whole install is three commands:

curl -fsSL https://kortix.com/install | bashkortix initkortix self-host start

From there the cost structure is two numbers you control: what your own compute costs, and what your model providers charge. With your own keys, the token bill is the same in either shape, because the same provider invoices you either way. The credits belong to Kortix Cloud; on your own machines there is no plan meter at all.

The trade is straightforward. Cloud buys you hosting you never operate, with pooled credits and email support. Self-host buys you the same software for $0, plus the operations that come with running it. A five-person team on the Team plan pays $200 a month with credits pooled across the seats; the same team self-hosting pays its own server bill and its model invoices, and runs entirely on infrastructure it owns.

When does each shape win? Teams that want agents working this week choose the cloud: the first session needs nothing installed. Teams with many seats, heavy usage, or a network boundary to hold choose self-host. The self-hosting guide on this site walks the install step by step.

How the rivals price the same work

Prices in this section were checked on each vendor's own pricing page in October 2026.

Dust charges per seat, with a credit allowance attached to each seat. Pro costs $24 per seat per month billed yearly, or $30 monthly, and carries 8,000 credits a month; Max is $120 yearly or $150 monthly with 40,000 credits. A free seat gets 500 credits once, and monthly allowances reset without rolling over. Enterprise is custom, adding single-tenant deployment and workspace-pooled credits.

n8n charges per workflow execution instead of per person. Starter is $20 a month and Pro is $50, both billed annually; Business is $800 and runs self-hosted. No cloud plan caps users, so the meter is runs: 2,500 executions a month on Starter, 10,000 on Pro, 40,000 on Business. A self-hosted Community Edition sits outside the paid plans entirely.

CrewAI publishes one free tier and one custom tier. The free plan includes the visual editor and 50 workflow executions a month; everything past that is an Enterprise quote with SSO, RBAC, and deployment on CrewAI's cloud, your VPC, or your infrastructure. No seat price is published.

Two more names fill out the field. Dify splits the same way: a free Community edition you deploy yourself, a free Sandbox tier with 200 message credits, and paid cloud workspaces plus a custom Enterprise tier. Budibase self-hosts free on its open-source plan, and its cloud starts at $19 a month for one creator with 5,000 actions.

The pattern across all of them: per-seat platforms fold software, hosting and model credits into one price per person; execution-priced platforms meter volume; an open-source platform drops the software bill to zero and leaves the compute and the tokens on your side of the ledger.

Which shape fits your team

Start on Kortix Free when the question is what agents can do: $0 and 200 credits a month answers it without a purchase order. Put a team on Kortix Cloud at $40 per seat per month when agents should run daily and nobody wants to operate servers. Self-host when seats multiply, usage runs hot, or the data must stay on your own network: the software is the same repo either way, and Kortix on GitHub is where you can read it first. The comparison chart on this site lines up the platforms feature by feature.

Ready to put your own numbers in? Try Kortix.