Documentation

Guide / Hosting client apps for agencies

Hosting client apps for agencies

Updated

An agency's hosting rarely starts as a plan. The first client app goes on whatever was quick. The fifth one goes somewhere else because the first host's limits ran out. By the tenth, one developer knows where everything lives, the backups are a cron job nobody has tested, and a client has asked what happens if their data goes wrong. This guide describes a structure that holds up as the client list grows. Each client gets a project. Each project has development, staging and production. Production changes that can't be undone wait for an approval. Your team, and the agents your team uses, work inside those boundaries.

The structure: organization, project, environment

Resilis organizes work in three layers:

  • Organization. The boundary for people, permissions and billing.
  • Project. A group of related apps, databases and storage. One project per client is the natural fit.
  • Environment. Every project starts with development, staging and production. Each environment has its own apps, databases and secrets.

For an agency, that gives you a predictable map:

Example

Your agency organization
├── client-a        (project)
│   ├── development  → app (develop branch) + PostgreSQL
│   ├── staging      → app (staging branch) + PostgreSQL   ← client review
│   └── production   → app (main branch) + PostgreSQL + custom domain
├── client-b        (project)
│   └── ...
└── internal-tools  (project)

Every developer, every agent session and every dashboard view starts from the same question: which client, which environment? That alone prevents the most common agency incident, which is a change meant for one client's staging landing in another client's production.

Use staging as the client's review environment

Many agencies call their review step UAT. Use the built-in staging environment for it rather than creating a custom one.

The reason is how protection works. Development and staging resources can be deleted or stopped directly by a person with the right role. Production resources need an emailed approval. Environments with custom names are treated as production: protection fails closed. A custom "uat" environment would make every cleanup wait for an owner's email.

So the pattern is: build in development, show the client in staging, release to production.

Who can see what

Organization membership has three roles:

Who can see what
RoleCan do
OwnerEverything, including approving protected production actions from the approval email
AdminCreate, change, deploy and delete; approve agent deletion requests in the dashboard
MemberRead projects, deployments and logs; restart deployments

Membership applies to the whole organization, not to one project. Plan your organizations with that in mind:

  • Your team works in one agency organization, with a project per client.
  • If a client needs to see their own dashboard, give them a separate organization that holds only their project. Do not invite a client into your agency organization, where they would see every other client's projects.
  • If a client pays their own hosting bill, a client-specific organization also keeps billing separate, because each organization has its own plan and invoices.

Plan limits that matter for agencies

Plan limits that matter for agencies
HobbyProEnterprise
Price$10 per month$50 per monthCustom
Projects510Unlimited
Team members310Unlimited
Custom domains510Unlimited
Database recoverySnapshotsSnapshots and point-in-time restoreRecovery, planned with you
Logs and metrics7 days30 days90 days

The plan fee becomes runtime credit. Apps, databases, disk, egress and object storage draw from it, and usage above the credit is billed at published per-second rates. Build minutes are separate: 200 on Hobby, 500 on Pro, then $0.02 per minute.

For an agency with production client data, Pro is the realistic floor. It is the plan with point-in-time restore. Past ten clients in one organization, talk to us about Enterprise.

Give the client a straight answer about backups

Sooner or later a client asks: "What's your backup story?" With this setup the answer is specific:

  • Every deploy can be rolled back. If a release breaks the site, the previous release is one action away.
  • The production database can be restored to a point in time (Pro and Enterprise, with continuous backups turned on). If a bad migration runs at 14:32, you can restore to 14:31, either in place or into a new database to inspect first.
  • Deleting or stopping anything in production waits for an owner's approval by email. A tired developer, or an over-eager agent, cannot remove a client's database with one click.

Say what it does not include, too. Rollback does not undo database migrations. There is no contractual uptime figure; support is by email, and the live status page is public. Resilis runs in Europe on infrastructure it operates.

Let your developers' agents in, with limits

Agencies adopting Claude Code or Codex face a specific risk: one developer's agent session can reach every client. Resilis keeps that bounded:

  • Each developer connects their own agent through the MCP server at https://mcp.resilis.io and signs in as themselves. The agent acts with that person's organization access.
  • Every agent tool call is recorded with the person, the agent client and the tool. If the record can't be written, the call does not run.
  • An agent can deploy, read logs, roll back and run migrations. It cannot delete anything directly. It can only prepare a deletion, and a person approves it. In production, that approval comes by email to an account owner.

A good working rule for client work: agents get development and staging freely, and production only for deploys and diagnosis. The platform enforces the destructive part. Your prompts and reviews handle the rest.

A checklist for each new client

  • 1. Create the project, named after the client.
  • 2. Connect the client's GitHub repository, or deploy from a developer's workspace through their agent.
  • 3. Create the development, staging and production apps on the develop, staging and main branches.
  • 4. Create one PostgreSQL database per environment. Bind each to its own app as DATABASE_URL.
  • 5. Turn on continuous backups for the production database and choose a retention window.
  • 6. Add the client's custom domain to the production app.
  • 7. Decide whether the client gets dashboard access, and if so, set up their own organization.
  • 8. Write down who on your team is an owner. Owners receive production approval emails.

When a client already has servers

Some clients want their app on infrastructure they own. Resilis can deploy to a client's Linux VPS or Kubernetes cluster through the Resilis connector, with the same workflow. Two limits apply: custom domains are not available on connected compute, and the client's infrastructure bill stays with their provider. Check the connected compute limits before promising this.

Frequently asked questions

Should each client be a separate project or a separate organization? A project per client inside your agency organization suits your own team. Use a separate organization when a client needs dashboard access or pays their own bill, because membership and billing apply to the whole organization.

Can a client see only their own app? Membership is organization-wide. Put the client's project in its own organization and invite them there as a member.

Can we call our review environment UAT? Use the built-in staging environment for client review. Custom-named environments are protected like production, which slows routine cleanup.

What happens if a developer deletes a client's production database by mistake? The deletion does not run. An account owner receives an approval email and must sign in and approve it first.

How many clients fit on one plan? Hobby allows 5 projects and Pro allows 10. Enterprise has no project limit.