Skip to guide

Resource lifecycle guide

A place for your agent
to build and run.

Code in a sandbox. Deploy an app. Give it a database, a server, and an inbox. Manage it all through Orbio, paid in CREDIT.

One setup. Your agent’s resources.

01 / SETUP
  1. 1

    Connect with Orbio

    Use your existing server-side API key, or connect your MCP client and approve infrastructure access. No provider accounts or keys are needed.

  2. 2

    Discover your resources

    Call infra.status. Orbio automatically gives each API credential or approved MCP connection its own isolated agent. No product setup is needed.

  3. 3

    Create what you need

    Quote a resource and approve its cost. Your agent can manage only its own resources, paid from your available CREDIT. Each agent can have one inbox.

Use the same Orbio account and CREDIT balance as models and tools. Resources are isolated per credential or MCP connection. No manual product setup or provider OAuth is required.

Connect your agent

02 / CONNECT

Connect your agent’s MCP client

Add Orbio as a remote MCP server. Sign in and approve infrastructure access. Orbio creates a separate resource scope for this connection on its first infrastructure call.

Remote MCP server · Streamable HTTP
https://www.orbio.so/api/mcp

Already connected? Reconnect and approve the infra scope if it was not previously granted. For clients with custom headers, you can instead use Authorization: Bearer <ORBIO_API_KEY>.

Try this prompt
Check my assigned infrastructure and available tools.
List my resources, then quote a five-minute E2B workspace.
Show me the cost before creating it.

MCP names use underscores: infra_status, resource_list, workspace_quote. The agent only receives tools its permissions allow.

Choose what to build with

03 / RESOURCES

Each API credential or approved MCP connection gets its own isolated resources. Expand a provider for its workflow, permissions, and funding behavior.

Sandboxes / E2BWrite files, run code, and preview an app in a private workspace.
  1. Quote a runtime and create a workspace.
  2. Write files, start a command, then read its saved output.
  3. Pause to keep your code. Resume explicitly when you need it.

infra.read · workspace.manage · workspace.files · workspace.exec

Choose 15–3,600 seconds of runtime. Workspaces auto-pause at the funded boundary. Renewal is explicit; reconnecting does not extend runtime.

Powered by E2B ↗

Deployments / VercelUpload an app, configure its environment, and publish a deployment.
  1. Create a project and configure its framework and build settings.
  2. Set environment variables, then upload your source files.
  3. Read build status and logs. Use the production alias to visit your app.

infra.read · deployment.manage

Hosting uses a funded period, with explicit renewal. Creation requires on_expiry: "delete": the project and deployments can be deleted when funding expires. Keep a copy of your source. The first upload becomes production; later uploads default to preview.

Powered by Vercel ↗

Servers / Fly.ioRun a background service or HTTP server in an isolated app.
  1. Create an app with its own private network.
  2. Use an immutable image digest; create a volume first if you need persistent data.
  3. Create a Machine, configure HTTP and allocate an IP when you need public access.

infra.read · worker.manage

Server funding must be renewed explicitly. Stopping a Machine does not remove its disks or volumes. Image upload accepts prepared OCI/Docker artifacts; it does not build an image from source.

Powered by Fly.io ↗

Databases / SupabaseCreate a Postgres database, manage schema, and store files.
  1. Create a project in your chosen region and wait for verified bootstrap.
  2. Apply your schema, grants, and row-level security policies.
  3. Use the project URL and publishable key in your app.

infra.read · database.manage · database.read · database.write

Projects use a funded period and require on_expiry: "delete". Expiry can permanently delete the database. Renew before expiry and keep independent backups. Database passwords and service-role keys are never returned.

Powered by Supabase ↗

Email / AgentMailGive an agent its own inbox to read, draft, and send email.
  1. Read mail pricing, then explicitly create the agent’s inbox.
  2. Read messages or prepare a draft, including replies and attachments.
  3. Review recipients and content before the separate send action.

infra.read · mail.manage · mail.read · mail.draft · mail.send

One live inbox per agent. Creation includes a month for 2.30 CREDIT at the default 15% surcharge. Sending costs 0.05 CREDIT per action for 1–20 recipients; reads and drafts are free. Renew the inbox explicitly before its prepaid month ends.

Powered by AgentMail ↗

Stay in control

04 / SPENDING

Approve a cost before work starts

Calls need available CREDIT and room under your product ceiling. Set max_cost on each mutation. Allocation reserves a budget; a successful operation can still have a pending bill. The default infrastructure surcharge is 15%, with provider-specific terms.

Renew before the funded period ends

Hosting has a finite lifetime, not automatic renewal. Vercel, Fly.io, and Supabase accept 60–86,400 seconds per funding window. Vercel projects and databases require consent to deletion on expiry. Keep backups and check funded_until.

Reconnect to the same operation

Save the request, its idempotency_key, and the returned operation ID. Poll that operation to recover progress. If admission loses its response, use the identical saved request. Closing a client or stopping a local wait does not stop paid work.

Keep access with the right agent

Resources use stable Orbio IDs. Reconnecting with the same credential keeps those resources. A new API credential gets a separate agent; owners can recover old resources from the dashboard. Another agent cannot manage them. Revoking access blocks new calls, while the chosen revocation policy controls work already funded.

Full funding, recovery, and provider behavior ↗

Find a method

05 / REFERENCE

Search by action or resource. Schemas come from the same contracts as API and MCP. Check the live catalogue for availability; permissions still apply.

133 methods · showing 12

infra.statusREAD
mail.pricingREAD
infra.pricingREAD
resource.listREAD
resource.getREAD
resource.spendingREAD
operation.listREAD
operation.getREAD
operation.cancelACTION · DESTRUCTIVE
funding.listREAD
funding.getREAD
workspace.quoteREAD