Open
Open
Open
Open
The problem

Agents hold the keys to everything — so where they run matters

An AI agent is an LLM given a goal, a set of tools it can call, and memory — and to do its job it holds credentials for all of them: your OpenAI or Anthropic key, OAuth tokens for the apps it touches, database passwords, webhook secrets. Run that on a shared multi-tenant platform and every one of those secrets sits on a machine you share with strangers. For an AI developer building per client, that's not a detail — it's the whole risk.

A Langflow environment gives each build its own sealed ground. One tenant, one instance, one database: your flows, your stored credentials, your execution history on hardware we own, under EU jurisdiction, isolated from every other environment by default-deny networking. Your agents' keys never share a box.
high-performance-cloud-compute.jpg
What every environment includes

The full agent builder, provisioned and locked down

✓ Langflow installed and ready — current release, provisioned and migrated, with a private admin login seeded on day one (no open-by-default instance). Log in and start building.
✓ Your own PostgreSQL database — flows, components, stored credentials and run history in a database that is yours alone, not a shared table.
✓ Login on from the first minute — auto-login is off; the instance is private to your account out of the box, not exposed until you lock it down.
✓ Call any LLM you like — point flows at OpenAI, Anthropic, Google, or your own self-hosted model; the key stays in your environment, the inference happens at the provider.
✓ Egress for the tools your agents use — reach the APIs and services your flows call, with private networks blocked (SSRF hygiene) and other tenants unreachable.
✓ Nightly backups included — database plus stored flows, on separate platform storage.
✓ Everything every environment gets — sealed networking, dedicated IP, automatic TLS, reserved compute.
No GPU needed, by design

Orchestration here, inference wherever you want it

Langflow builds and runs the agent — the reasoning loop, the tool calls, the RAG retrieval — but it never runs the model itself. Inference happens at your LLM provider (their GPUs, your API key) or at a self-hosted model you point it at. So an environment is normal CPU and RAM, priced like any other, with none of the cost or scarcity of GPU hosting. The heavy compute stays where it belongs; you pay for the orchestrator.
✓ Bring your own provider — OpenAI, Anthropic, Google, Mistral, or an OpenAI-compatible endpoint.
✓ Or your own model — point a flow at your self-hosted Ollama/vLLM box; Langflow just calls it.
✓ CPU/RAM pricing — no GPU premium, because no GPU runs here.
For AI developers and the agencies who ship agents

We run the substrate. You build the agents.

We guarantee the environment: the app pods, the database, the network, uptime and provisioning. You build in Langflow — your flows, components, prompts, tools and credentials. It's stock, current Langflow with the full builder and the flow API, so everything in the docs and every component works as published.

Export a flow as JSON and move it; call your flows over the API from your own app. Deploy one environment per client, credentials and data cleanly separated — the same repeat-deploy shape agencies already run for hosting.
Isolation is a security feature

The strongest sovereignty story we have

✓ Credentials that never share a box — every API key, OAuth token and secret your flows hold lives in your own database, in your own sealed environment. A neighbouring tenant cannot reach it, by network policy, not by promise.
✓ Your clients' data stays put — the documents your RAG flows index and the payloads your agents process are in your environment only, under EU jurisdiction.
✓ Sovereign EU ground — on hardware we own, in Sofia or Amsterdam, no US parent, no CLOUD Act reach. For agents handling European data and European clients' secrets, that's a real answer, not a region label.
Reserved resources, no run caps, no per-execution meter — you pay for the environment you reserve, not for how many times your agents run.

Under the hood

Compute
Dedicated Langflow (Python) pods, resources reserved in-cluster (request = limit); Shared or Dedicated physical CPU clusters — your choice. No GPU — inference is external
Database
Your own PostgreSQL pod on replicated Ceph block storage — flows, components, credentials, run history
Auth
Private by default: auto-login off, an admin account seeded at provision; the instance is never open-by-default
Egress
Outbound to LLM providers and your flows' tool APIs; private-range destinations blocked (SSRF hygiene); other tenants unreachable
Network
Default-deny NetworkPolicies: environments cannot reach each other, period
Ingress
Dedicated IP per environment, TLS at our EU edge for your Langflow domain, HTTP/2
Backups
Nightly database plus stored flows, on separate platform storage
Releases
Every component version-pinned down to image digests; upgrades staged on test environments first
Jurisdiction
EU company, own network (AS202309), own hardware in Sofia & Amsterdam, own keys

Plans

Langflow is orchestration — an app and a database, CPU/RAM light, no GPU. A Standard environment runs real agent workloads comfortably; larger plans add headroom for heavier flows, bigger RAG stores and more concurrency. Per environment, per month, excl. VAT. Dedicated IP and nightly backups included in every plan.
Plan
Resources
Sofia — Shared CPU ★
Sofia — Dedicated CPU
Amsterdam — Shared CPU ★
Amsterdam — Dedicated CPU
Standard
4 vCPU · 16 GB RAM · 100 GB SSD
€45
€105
€55
€130
Pro
8 vCPU · 32 GB RAM · 200 GB SSD
€90
€210
€110
€260
Enterprise
16 vCPU · 64 GB RAM · 400 GB SSD
€180
€420
€220
€520
Plan
Resources
Standard
4 vCPU · 16 GB RAM · 100 GB SSD
Pro
8 vCPU · 32 GB RAM · 200 GB SSD
Enterprise
16 vCPU · 64 GB RAM · 400 GB SSD
Plan
Sofia — Shared CPU ★
Standard
€45
Pro
€90
Enterprise
€180
Plan
Sofia — Dedicated CPU
Standard
€105
Pro
€210
Enterprise
€420
Plan
Amsterdam — Shared CPU ★
Standard
€55
Pro
€110
Enterprise
€220
Plan
Amsterdam — Dedicated CPU
Standard
€130
Pro
€260
Enterprise
€520
★ Shared CPU is our recommended option — same isolation and features in both modes; the price difference is the physical CPU guarantee level alone. Disk grows online; plan upgrades are applied in place. Volume pricing for agencies running one environment per client on request.

Frequently asked

Do I need a GPU?
No. Langflow orchestrates — it calls an LLM, it doesn't run one. Inference happens at your provider (their GPUs, your key) or at a self-hosted model you point to. Your environment is normal CPU and RAM, priced accordingly.
Which LLM providers can I use?
Any Langflow supports — OpenAI, Anthropic, Google, Mistral, OpenAI-compatible endpoints — or your own self-hosted model. The key lives in your environment; the call goes out to the provider.
Is my instance open to the internet?
No — auto-login is off and an admin account is seeded at provision, so the instance is private to you from the first minute, not open-by-default. You control access.
Where do my API keys and credentials live?
In your own PostgreSQL database, inside your own sealed environment. Default-deny networking means no other tenant can reach them — the isolation is architectural, not a policy line.
Can I run one per client?
Yes — that's the model. One environment per client, each with its own instance, database, credentials and dedicated IP, fully isolated. The same repeat-deploy shape agencies already use for hosting.
Do you support or build my flows?
No — same boundary as everywhere on the platform. We guarantee the environment: database, network, uptime, provisioning. Your flows, components, prompts and credentials are yours.
Where is my data?
Sofia or Amsterdam — your choice — on hardware owned by Radibase, an EU company on its own network. No US parent, no CLOUD Act exposure.

Give your agents their own ground

Get your environment