Overview

Difference from Agent Applications

Dimension	Agent Applications	Managed Agents
Execution Mode	Stateless invocation; context maintained on the application side	Session state maintained on the server side; supports interruption and resumption
Execution Environment	Shared runtime	Isolated sandbox — cloud-based container
Event Model	Response-level streaming output	Session-level SSE event stream; event history persisted
Typical Use Cases	Q&A, chat, lightweight tasks	Long-running tasks such as multi-step tool calls, code execution, and file processing

Concept	Description
Agent	A configuration combining model, system prompt, tools, MCP services, and Skills. Once created, it is referenced by ID and can be reused across multiple sessions.
Environment	Sandbox configuration for session execution — a cloud-based container hosted by Bailian.
Session	A single execution instance of an agent within a specified environment, performing tasks and generating outputs.
Event	Messages exchanged between the application and the agent, including user messages, tool call results, and state changes.
Memory Store	A persistent mounted resource that the agent reads and writes with file tools; its content persists across sessions.
Scheduled task	A resource that binds an agent to a trigger configuration — scheduled or manual — with each trigger producing a run record.

Agents interact with the execution environment via the following tools:

Command Execution: Run shell commands inside the sandbox
File Operations: Upload files to mount into the sandbox or download from URLs; supports reading, writing, editing, finding, and searching
Memory: Mount memory stores and read/write working notes across sessions
MCP Services: Integrate external tooling services
Skills: Mount preconfigured tool combinations that encapsulate end-to-end task workflows

Create an Agent: Configure model, system prompt, tools, and MCP services. After creation, reference it by ID.
Configure the Execution Environment: Create a cloud-based container sandbox and optionally configure pre-installed packages.
Initiate a Session: Create a session instance referencing both the agent and environment configuration.
Send Events and Receive Responses: Send user messages; the agent autonomously executes tool calls and pushes results via an SSE event stream. Event history is persisted on the server.
Intervene or Interrupt: Send new events during execution to guide direction, or interrupt the current task.

Long-Running Tasks: Tasks involving multi-step tool calls lasting minutes to hours.
Code Execution: Run scripts, install dependencies, and debug programs inside the sandbox.
File Processing: Read, write, search, and edit files within the sandbox.
Stateful Sessions: Maintain context and filesystem state across multiple interaction rounds.
Rapid Development / Low Operational Overhead: Focus on agent logic instead of building and maintaining infrastructure like agent loops, sandbox orchestration, or tool execution systems from scratch.