Everything about Fractal, answered.

Fractal is a platform where you and your AI agents build software — interfaces, processes, integrations and AI agents — and Fractal runs, hosts and scales it. 1 connected platform instead of a stack of disconnected tools: front-end, back-end, workflow engine, queues and hosting.

For anyone who builds a product with AI agents — or in a visual editor — and needs it to run in production: founders and builders, engineering and product teams, and companies whose processes can’t stop. 2 banks with 1.2M customers run on Fractal today.

A whole product: its interfaces, the processes behind them, its integrations and its AI agents. SaaS apps, online stores, marketplaces, CRMs, booking apps, back offices, AI support desks — and banking, fintech, insurance, logistics, education and healthcare products.

No. Everything can be built in the visual editor — interfaces, processes and integrations — by you or by your agent. Where you need code, a Function node runs JavaScript.

Join the waitlist: Fractal opens to teams in waves. The Free plan stays free forever for early birds, with no credit card.

Over MCP. Connect your agent to Fractal’s MCP server, and it signs in itself — no token to paste. It builds interfaces, processes and agents, runs them and reads the results. You see everything it builds in the visual editor.

Claude Code, Cursor, Codex, the Claude app — or any MCP client.

Anthropic’s, OpenAI’s and Google’s. Connect a provider once and reuse it; which model runs is set apart from the process, so you can swap the model and keep the process.

Yes. Everything an agent builds, you can build yourself in the visual editor — the same editor where you see every step your agent takes.

It runs live the moment it’s saved — nothing to build, deploy or release. A change goes from request to production in 1 hour.

On a separate branch: try it in View mode and through its endpoints, then merge it. Branches come from the Startup plan up.

Yes. Every change is a version, with its author and time. Roll back to any earlier version — and see what the rollback will change before you run it.

Retries and waiting are built into the engine. When a task does stop, you find it, see what it carried and the node where it stopped — and a process can route each failure to its own next step.

As long as it needs: tasks survive restarts and deploys, and resume the moment the next event arrives. On Fractal Cloud a task can wait up to 30 days; on your own servers, longer.

Up to 200,000 actions per second on a load test, each under 5 ms, and 20,000 requests per second, each starting a whole process — run end to end in under 100 ms (p99). For comparison, as each publishes it: Temporal Cloud, 150,000+ actions per second at peak; n8n, 220 executions per second on 1 instance, load-tested up to 2,500 on 7 servers.

Sources: Temporal’s Series D announcement · n8n’s benchmark

Every part runs as a service of its own and scales out as its own load grows — more of the busy part, and only that one. A rush on one part never slows the rest.

99.99% uptime. Fractal runs across availability zones and data centers: if a zone or a whole data center goes down, another takes over, with your data.

Yes. 2 banks with 1.2M customers run on Fractal in production. In 1 of them, a new credit product went from request to live in 3 days.

Every call is checked for its token and permissions. Every credential is encrypted with its own AES-256-GCM key. API calls can be secured with client certificates (mTLS) and PSD2 signing (QSEAL). JavaScript runs in isolated containers, and every change is recorded with its author and time.

Yes. 5 roles, from run-only to owner, given at any level — a company, a project, a process — and inherited by everything inside. Run-only access lets a person or a system run a process without seeing its logic.

Yes — single sign-on (SSO) from the Enterprise plan up: your team signs in through Okta, Azure AD or any OIDC or SAML provider. And two-factor sign-in (2FA) is coming in Q4 2026, with a second factor at every sign-in.

Not yet: SOC 2 and ISO 27001 are planned, as is a DORA exit plan. Security documentation is available on request.

On Fractal Cloud, or on your own hosting — in your cloud or your own data center. Self-hosted Docker (Open Source) is free: 1 container, up to 50 requests per second, coming soon to GitHub. Self-hosted K8s (Enterprise) runs on your Kubernetes, with scaling and failover.

In your own database: the Data node connects your processes to it. A database in Fractal Cloud is planned.

Everything exports — as Fractal objects, not as code. It runs only on Fractal: in Fractal Cloud or on your own servers.

Yes. Web and mobile interfaces — iOS and Android — are built in the same editor, connected to your data and to the process behind every button.

Yes. Your customers and your employees sign in to the interfaces you build on Fractal.

Yes — through your payment provider’s API, which a process calls like any other integration.

No. Bring your prototype. We look at it and tell you which flow we move first. We help you move it to Fractal for free — the process, the interfaces and your database, out of Lovable Cloud and into your own.

Into your own database. The Data node works with your own database, so it stays yours. We help you move 1 of your processes to Fractal for free — its interfaces and its data included.

No. Keep your site. Bring 1 form from it. We help you move the process behind it to Fractal for free — its interfaces and your database included. Your customers sign in to their account on Fractal — in place of the User Accounts Webflow switched off on 29 January 2026.

Sources: Webflow’s changelog, 29 January 2026

Fractal is the platform even banks run on. n8n’s own benchmark: 220 executions per second on 1 instance, load-tested up to 2,500 on 7 servers — and under load it takes over 100 seconds to respond. On Fractal, a whole process runs end to end in under 100 ms (p99), up to 200,000 actions per second on a load test, with retries and waiting built in. And you don’t move everything at once: start with a free migration assessment, and a free PoC licence lets you run Fractal in parallel with n8n.

Sources: n8n’s benchmark

You don’t rebuild them yourself. Bring the process your Zaps — or your Make scenarios — hold together. We move it to Fractal for free, its interfaces and your database included. Running Zapier across a company? Start with a free migration assessment: you get a report on your processes and a migration plan, and a free PoC licence lets you run Fractal in parallel with Zapier.

You don’t rebuild them alone. Start with a free migration assessment of the processes your team runs from Retool, consoles included: you get a report on your processes and a migration plan. Then we help with the whole migration, staff interfaces included. A free PoC licence lets you run Fractal in parallel with Retool: tasks already running finish in Retool, new tasks start on Fractal.

Yes. A free PoC licence lets you run Fractal in parallel with your current platform. While both run, tasks already running finish on your current platform, and new tasks start on Fractal. Start with a free migration assessment: you get a report on your processes and a migration plan. Then we help with the whole migration, staff interfaces included.

We build your processes on Fractal — and, if you need it, AI can generate BPMN documentation from them. Start with a free migration assessment: you get a report on your processes and a migration plan, and a free PoC of the first process runs alongside it. While both run, tasks already running finish on Camunda; new tasks start on Fractal.

With a free migration assessment: you get a report on your processes and a migration plan. How long it takes depends on your project. A free PoC of the first process runs alongside it — we build it for you or with your team, or we train your team, free. Then we help with the whole migration, staff interfaces included.

Yes: 1 use case in 1 week, free. The week starts when we get access to your systems or when we agree on the use case. The PoC runs on your own servers or on a trial plan on Fractal Cloud — as we agree.

Yes. Talk to us about moving 1 of your client projects to Fractal for free.

1 node run. When a process runs, each node it passes through is 1 action, counted across every process in the company. The count starts again each month — on yearly plans too.

Yes. Free is a working plan, not a trial: build, deploy and run a complete process on it, with no card.

You see it coming. The editor’s side panel shows the month’s actions and storage as rings: orange at 75%, red at 90%, with how much is used and how to avoid a stop. A task already running finishes. New tasks past the month’s actions fail with an error your process can handle, and requests past the per-second limit get HTTP 429. Upgrade at any time, and the new limits apply as soon as the payment completes.

Per company: each company has its own plan, usage and invoices. Payment runs through Stripe’s hosted checkout — Fractal never sees or stores your card.

Upgrade at any time. Moving down to Free is always open. A plan whose storage is below what you already use can’t be chosen until you free some up.

Yearly billing saves 34% on Personal plans and gives 2 months free on Business plans. Actions still reset each month.

Personal plans are for individual developers and small projects. Business plans are for teams going to production: higher throughput, more storage, larger limits, a built-in ledger and fast counters, and version control — and from the Enterprise plan up, single sign-on (SSO) and a Slack channel with the team. Every paid plan has a 99.9% SLA; Custom, 99.99%.

Yes. Fractal is cloud-agnostic: it runs in any cloud or your own data center. Self-hosted Docker (Open Source) is 1 container on your server, up to 50 requests per second; Self-hosted K8s (Enterprise) runs on your Kubernetes with scaling and failover, beyond that, its limits set with you.

Banks run on Fractal in production. Every call is checked, every secret encrypted, every change on record — and security documentation is available on request.

The files your company keeps and the data of its tasks that are saved, together. A plan can’t be moved to while you already use more storage than it holds.

How many requests per second your company’s synchronous endpoints take; past it, they answer HTTP 429. Work that runs through queues isn’t limited at entry — it drains at the engine’s own speed.

Tasks are kept 30 days on every plan. Object history is kept from 1 day on Free up to 180 days on Enterprise, then deleted.

Yes, from Hobby up. Bind your domain to a project, prove you own it with a DNS record, and its published interfaces and API endpoints answer there.

Branches, merges and rollbacks for everything you build — processes, interfaces and the rest — from the Startup plan up.

The top level of everything you build: a company holds your projects, and each project its folders. It has its own team, its own plan and limits of its own, so a busy company never borrows from a quiet one.