Core concepts
Follow one Service conversation through workspaces, agents, sessions, threads, and runs.
Follow one conversation with an agent that reports project status. This example uses Service, the managed-agent runtime, through Console or the HTTP API. For embedded or interactive execution, see Choose your path.
Build the agent
An administrator gives you a role in a workspace, which contains your team's agents, other resources, and conversations. Your roles in the organization and workspace determine what you can use or change.
An agent combines a model, instructions, and tools. Example instructions: “Summarize project progress and cite the information you used.” Each save creates an immutable revision. A run uses the agent's default revision unless the message pins another one.
| Part | Role in the agent's work |
|---|---|
| Model | Interprets requests and produces responses or tool calls |
| Tool | Looks up a task, reads a file, or performs another configured action |
| Connection | Gives access to external tools through MCP or an app account |
| Environment | Provides the files and command execution used by tools |
| Memory | Holds project decisions the agent can read and update across conversations |
Start with a model and instructions. Before you ask the agent to look up real progress, add a tool or connection that reads your project data. Tools and connections explains the setup.
Start a conversation
Ask: “Summarize the project status and list the next two tasks.”
A session is one conversation; it groups threads. Each thread has its own history and inbox. Through the API, forking a run creates another thread in the same session to explore a different direction. A run advances a thread using the agent's model and tools.
You see output as it streams. A message sent while a run is active joins that run at its next model request when it is compatible; otherwise it waits in the inbox for a later run.
Answer a question or approve a tool call
If the agent asks which project you mean, answer in the conversation. If a tool call needs approval, approve or deny it. The conversation continues with your response. Applications handle these through Waits, approvals and questions.
Continue later
Ask “Which task should I do first?” in the same thread to continue from its history. An active run continues when you disconnect, and Service stores the thread's history for your return.
| Information | Purpose |
|---|---|
| Thread history | Continue this thread |
| Memory | Save and retrieve information across conversations |
| Environment files | Store documents and other working files for tools |
Mount a memory on the thread, or set it as an agent default, to share saved decisions across conversations. It can supply file context or recall relevant records automatically; enabled memory tools let the agent search and update it. The idle policy of the environment template determines how long an environment's files remain available.
Next, try an agent in Console or connect your application. See Agents, threads and runs for revision selection and execution details.