Choose your path
Start with the outcome you want. Agent Studio and Sapiom MCP are two interfaces for the same agent-authoring lifecycle. Router is a separate product that can be used directly or during an agent run. The dashboard is where you operate cloud-owned state.
Choose by outcome
Section titled “Choose by outcome”Build an agent with Studio or MCP
Section titled “Build an agent with Studio or MCP”Agent Studio
Section titled “Agent Studio”Choose Agent Studio when you want the integrated experience. Install the desktop app or launch the command-line host, choose Claude Code or Codex for a coding session, inspect the project on Canvas, test with local stubs, deploy, and inspect production runs.
Studio manages its Sapiom project connection for the session. You ordinarily do not configure MCP aliases or invoke raw tool names yourself.
Sapiom MCP
Section titled “Sapiom MCP”Choose Sapiom MCP when you want to stay in Claude Code or Codex. The local @sapiom/mcp server gives your coding agent project-aware tools to scaffold, check, test, link, deploy, run, inspect, signal, and schedule an agent.
Both paths work on the same TypeScript agent project and deploy it to the same Sapiom runtime. You can move between them without converting the project.
Use Router directly or in an agent
Section titled “Use Router directly or in an agent”Router is the named Sapiom product for LLM traffic. Use its provider-compatible API directly when your application needs model inference without a Sapiom agent graph. Direct calls require a Sapiom account and are recorded as Router activity.
Use Router during an agent run when model inference is one step in a larger automation. In that path, the runtime supplies the credential and Sapiom can associate the Router request with the agent run and step.
These are both supported Router paths. Router is not a general route for invoking browser, compute, data, messaging, or other Sapiom capabilities outside an agent.
Capabilities run with agents
Section titled “Capabilities run with agents”Capabilities are managed operations that deployed agent steps call through ctx.sapiom. The Sapiom runtime supplies the execution-scoped credential and records capability activity with the run.
During local development, Local Run executes the real step code but replaces Sapiom capability calls with committed or explicitly supplied stubs. This lets you test control flow without creating capability spend.
Do not initialize a standalone capability client with an account API key. Put the operation in an agent step, test it locally with a stub, and deploy the agent when you need the real capability.
Use the dashboard for cloud state
Section titled “Use the dashboard for cloud state”The Agents dashboard shows organization-owned agents, versions, builds, production runs, alerts, secrets, and settings. It operates deployed state; it is not a browser editor for the local source project.
| If you want to… | Start with… |
|---|---|
| Create or change an agent in an integrated workspace | Agent Studio |
| Create or change an agent from Claude Code or Codex | Sapiom MCP |
| Test an agent without real capability use | Local Run through Studio or Sapiom MCP |
| Create, edit, test, deploy, run, or inspect an agent | Sign in to Sapiom, then use Studio or Sapiom MCP |
| Send provider-compatible LLM requests from an application | Router directly |
| Include LLM inference in a multi-step automation | Router during an agent run |
| Use a Sapiom capability | A step in a Sapiom-managed agent run |
| Inspect organization-wide deployed state | Agents dashboard |
Next, see How Sapiom works, or build your first agent.
© 2026 Sapiom, Inc.