TL;DR
- n8n Cloud reduces infrastructure work; compare the execution tier, project boundaries and subscription terms with the agency's client model.
- Elestio offers managed n8n hosting. Check which upgrades, backups, monitoring and recovery tasks are actually included.
- Railway provides deployment templates and managed infrastructure building blocks; the operator remains responsible for the configuration and application lifecycle.
- Render documents deployment with Postgres and a path to worker-based architecture. Its free database expiry and idle behaviour make free resources a poor production assumption.
- Hetzner provides infrastructure for a team prepared to operate the server, database, workers, security and recovery itself.
- Before hosting workflows for clients, check n8n's current licence and the exact access model. Separate client data, credentials and incident ownership regardless of the host.
What does an agency need from n8n hosting?
An agency running ten client workflows needs more than a container that starts. It needs a stable webhook endpoint, durable workflow and credential storage, a way to see failed runs and a plan for recovering one client without exposing another's data. The right host depends on who operates those pieces and what happens when the person who built them is unavailable.
Start with the service boundary. A vendor-hosted subscription, a managed third-party n8n deployment and a self-operated container on a cloud platform can all run workflows, but the responsibilities differ. Identify who patches n8n, manages the database and queue, responds to incidents, tests restores and communicates with the client.
This comparison concerns hosting after choosing n8n. The broader n8n vs Zapier vs Make guide addresses the platform decision. We have not deployed or load-tested these hosting options for this article.
n8n hosting options at a glance
| Option | Operating model | What the agency must verify | Cost boundary |
|---|---|---|---|
| n8n Cloud | Hosted by n8n | Execution tier, projects, access and retention | Subscription and usage allowances |
| Elestio | Third-party managed deployment | Scope of upgrades, backups, monitoring and support | Managed service plus n8n licence if applicable |
| Railway | Platform services with n8n deployment templates | Persistent storage, database, queue and recovery | Compute, database, storage and operator time |
| Render | Container service and managed infrastructure | Postgres, workers, webhook URL and resource plan | Service, database, queue and operations |
| Hetzner | Infrastructure operated by your team | Server security, backups, upgrades and availability | Compute, storage, networking, licence and staff |
A deployment template is a starting configuration, not a service-level guarantee. Check whether each route is permitted for your commercial and client-access model.
1. n8n Cloud: for vendor-hosted workflows
n8n Cloud is the direct hosted option. n8n's pricing describes workflow execution allowances and distinguishes hosted and self-hosted plans. Its Cloud support guidance identifies Pro-1 at 10,000 and Pro-2 at 50,000 monthly executions, with specific conditions for buying more. Check the current plan, billing frequency, concurrency and retention rather than applying an entry-tier price to 50,000 runs.
Cloud can fit an agency that wants n8n to operate the underlying service and can use the available project and access controls. It does not remove the agency's responsibility for workflow logic, client credentials, downstream API failures or a documented handoff. Ask how client data and user access are separated in the proposed account.
n8n's execution-limit guidance says a Pro-2 monthly subscriber can buy extra executions at €2, $2 or £2 per 1,000, plus applicable sales tax. Ten thousand additional executions would cost €20 per month before tax where euro pricing applies, on top of the Pro-2 subscription, if the desired increase is available in the preset tiers. The vendor says annual Pro-2 changes require support rather than self-service. Check the current base price and available tiers; an extra-capacity calculation is not the complete subscription price.
The exact commercial model matters. Review the Cloud subscription terms and discuss the agency's client-access arrangement with n8n. Do not assume a self-hosted licence explanation automatically applies to Cloud.
2. Elestio: for managed n8n hosting
Elestio offers managed n8n hosting and describes different deployment and billing options. It can be a candidate when the agency wants help operating the infrastructure but does not want to build every server-management process itself.
Request the precise responsibility matrix: n8n version upgrades, database backups, restore testing, security patches, monitoring, incident response, retention and customer support. “Managed” can cover infrastructure while leaving workflow errors and application-level recovery with the agency. Confirm whether the chosen configuration supports the database, worker and queue architecture you require.
Elestio's installation guide describes its included support level as including seven days of remote backup retention, proactive monitoring and email support with a three-day response time. Treat those as a stated support scope to verify in the contract, not as proof of a tested restore or a recovery-time guarantee. Its pricing varies by infrastructure and region, so quote the selected configuration rather than assuming a single advertised starting price covers a multi-worker deployment.
Elestio may fit a team that wants an operational partner. Compare the contract's support and recovery commitments with n8n Cloud and a platform-hosted deployment. Verify any required n8n licence separately; a hosting invoice is not automatically permission for every client-access model.
3. Railway: for a platform-managed infrastructure stack
Railway publishes n8n templates, including a simple deployment using a persistent volume. Other templates describe queue-mode architectures with separate workers, Postgres and Redis. Inspect the specific template and image version: marketplace templates can be maintained by third parties and are not all the same production configuration.
A platform can simplify deployment and managed services, but the agency still needs to configure persistence, secrets, webhooks, backups and upgrades. Confirm where the encryption key is stored, how the database is recovered and how workers obtain the same key. Do not use a nightly or preview image for a critical client workflow merely because a template advertises the newest version.
Railway is a candidate when the team wants platform services and has an operator who understands n8n. It may be a poor fit if nobody can investigate a failed deployment or restore the data. Price every service in the architecture, not just the main container.
4. Render: for a documented container and Postgres deployment
Render's n8n guide describes deploying the official n8n image with either Postgres or a persistent disk. Render recommends Postgres for the documented setup, including a route to multiple instances; a disk attached to one instance has different scaling constraints.
The guide states that free web services can spin down and free Postgres databases expire after 30 days. Those limits make the free example useful for evaluation but unsuitable as an unexamined production plan. Select paid resources and confirm backup, availability, webhook URL and recovery requirements for the actual workload.
Render is worth evaluating if the team prefers a documented container and managed-database approach. Your team still owns the n8n configuration, credentials, version and application-level failure handling. If queue mode is required, include the additional queue and worker services in the design and price.
5. Hetzner: for an operator-managed server
Hetzner provides cloud infrastructure on which a technically capable team can run its own n8n deployment. This route gives the operator control over the server, network and deployment choices, while placing a larger share of installation and maintenance work on that operator.
Specify the database, queue, backup location, monitoring, patching and recovery arrangements rather than comparing only a virtual-server sticker price. Decide how the deployment survives a server failure and who can restore encrypted credentials. One low-cost server can be appropriate for a non-critical pilot; it is not automatically a resilient multi-client production service.
Hetzner is a candidate when the agency has sysadmin capacity and wants control of its infrastructure. A managed service may be a better purchase when the team cannot support updates and incidents. Confirm regional hosting, data-processing terms and current infrastructure prices for the intended setup.
Can an agency host n8n for clients?
The deployment architecture and the software licence are separate decisions. n8n's licence FAQ says its free self-hosted Community edition can be used to build and run workflows for clients when only people in your organisation create or edit them and clients cannot change the workflow logic. It says a commercial licence is needed when people outside your organisation can build or configure workflows, including through an API or another interface, or when hosting n8n as a service. The FAQ says clients may connect their own accounts, but another example says they receive only outputs and have no access to n8n itself. Confirm how those statements apply to your connection and access design rather than assuming every client login is permitted. The FAQ treats n8n Cloud under its own subscription terms; this article does not determine your licence rights.
Isolation also has several layers. Separate projects or user permissions may help within one instance, but they do not automatically provide separate databases, encryption keys, network boundaries or incident blast radii. Map which users can see workflows, credentials and execution data, and which client can be paused or restored independently. Some features depend on edition or plan.
For higher-risk or contractually separate clients, compare a dedicated deployment per client with a shared deployment that has appropriately verified controls. Dedicated instances increase operating work and may change licence and hosting costs. Document who owns each domain, credential, backup and workflow at termination.
What do 50,000 monthly executions cost to operate?
Model ten client workflows producing 50,000 executions per month in total. Include the main application, workers, database, Redis or another required queue, persistent files, backups, monitoring and operator time. An execution is not necessarily one external API call: a single workflow can make several downstream requests that have their own limits and charges.
A synthetic monthly self-operated estimate might allocate $100 to the main service and workers, $80 to database and queue, $40 to backups and monitoring, and eight operator hours at $60 per hour. The total is $700 per month, or $0.014 per execution at 50,000 runs. Add any n8n licence, downstream API fees, incident response and initial implementation. These are planning inputs, not quotes from the listed hosts.
If an isolated setup requires five separate environments and each adds a hypothetical $60 per month in infrastructure, add $300, making the illustrative total $1,000, or $0.02 per execution at unchanged volume. This does not imply five environments are sufficient or necessary for ten clients; use the actual isolation design and contract.
For n8n Cloud, obtain the current qualifying 50,000-execution plan price and the required project and retention terms. Compare it with the self-operated total including a tested recovery process. A low infrastructure bill does not include the staff time needed to diagnose failed workflows or upgrade the application.
Distinguish execution capacity from execution history. n8n's pricing FAQ says saved runs are removed when the first applicable count, storage or retention limit is reached, without stopping workflows solely because that history limit was reached. If a client contract requires longer audit evidence, design an authorised export or logging approach and check its cost and data-retention terms instead of assuming all 50,000 runs remain searchable.
How do you test capacity and recovery?
Use synthetic client workflows before moving production. Include a scheduled CRM update, a webhook-driven lead event, a workflow with an attachment and one intentionally failing downstream API. Run a representative peak, not just the monthly average. Record queue age, completion time, error rate and the effect on other clients when one workflow becomes slow. This is a proposed test, not a benchmark we have performed.
For a queue-based architecture, follow n8n's scaling guidance for the supported database, Redis, workers and configuration. Verify that the main process and workers can access the required credential encryption key and that file handling works across workers. A template that launches containers does not establish that its binary-data and recovery settings match your workload.
n8n's queue-mode documentation says binary data uses database storage by default. Licensed external storage can use AWS S3 or Azure Blob Storage. Its binary-data guidance also permits filesystem mode in queue deployments when every main instance, worker and webhook processor handling binary data can access the same shared volume at the same configured path. A local disk available to only one container does not meet that condition. Check the selected version, edition, storage limits and backup design before choosing. In the attachment fixture, confirm every worker can access the file and the restored environment can recover it; a database restore alone does not demonstrate recovery of externally stored attachments.
Practise an isolated restore from the database, required file storage, configuration and encryption key. Confirm that credentials can be decrypted, webhooks point to the correct environment and scheduled workflows remain disabled until the test is safe. Rotate test secrets afterwards. Measure the recovery time and data gap, then document who can perform the procedure for each client.
During upgrades, restore a copy and run the same synthetic workflows before changing production. Define a rollback path and an alert for failed or delayed executions. The durable revenue workflows guide covers recovery concepts; the GTM engineering playbook helps assign the operator and client handoff responsibilities.
Which n8n hosting option should you choose?
Start with n8n Cloud when the hosted service and terms fit, Elestio when a third-party managed arrangement is attractive, Railway or Render when the agency can operate an application on managed infrastructure, and Hetzner when it owns the server and recovery process. Compare the same licence, isolation and availability requirements across all five.
For adjacent infrastructure decisions, see our cloud AI agent platform guide. Hosting n8n is one part of the workflow; it does not replace choosing and governing the services it calls.
FAQ
Is n8n Cloud the same as self-hosted n8n?
No. n8n operates Cloud under its subscription terms; a self-hosted deployment places more infrastructure responsibility on the operator and may use different licence and feature terms. Compare the exact editions and responsibilities.
Does queue mode solve every reliability problem?
No. Workers and a queue can help distribute execution, but database persistence, credentials, retries, monitoring, external API failures and recovery still need to be designed and tested.
Can one instance safely serve several clients?
That depends on the licence, edition, access controls, data sensitivity and contractual requirements. Test the intended boundaries and failure scenarios; separate projects do not by themselves prove complete infrastructure isolation.





