Skip to main content
Find setup guidance for AgentScope development, integrations, and service deployment.
No. AgentScope 2.0 changes the agent, message, event, tool, and state APIs. Existing applications need code changes; upgrading the package alone does not migrate them.Start new projects with 2.0. For existing applications, follow the 1.x to 2.x changes or keep using documentation matching the installed 1.x version.
Yes. Sandboxed workspace backends include Docker, Bubblewrap, Apple Container, Kubernetes, E2B, Daytona, and OpenSandbox, among others. LocalWorkspace runs on the host and does not provide sandbox isolation.Workspaces manage the execution environment, MCP server lifecycles, skills, and context offloading. See Workspaces to choose a backend and Workspace Management to allocate workspaces across users, agents, and sessions.
Use the built-in Agent Service and its matching web UI. Run the backend example and web UI example from the same checkout to try the application.Follow the Agent Service setup guide for prerequisites and startup commands. You can also build your own frontend using the service’s HTTP APIs and event streams.
Yes. RAG provides document parsing, chunking, embedding, and vector retrieval. RAG Service adds document management and indexing for hosted applications.Long-term memory supports file-based agentic memory, ReMe, and Mem0 through middleware. Use RAG to retrieve knowledge and long-term memory to retain information across sessions; both can be used together.
No. AgentScope supports model APIs including OpenAI, Anthropic, Gemini, DashScope, and Ollama. Choose a model class and matching credentials; available features depend on the selected model and API.See Model Configuration and Chat Models.
Yes. MCP (Model Context Protocol) connects agents to external tools through STDIO or HTTP. Skills provide instructions in SKILL.md files that agents read and follow using their available tools.Register them through the toolkit or manage them through a workspace. See MCP and Skills for setup.
You can compose agents in Python or use Agent Team for service-based collaboration. A leader creates workers, delegates tasks, and exchanges messages with them; each worker runs in its own session.Use teams when a task benefits from delegation or separate agent contexts. A single agent is sufficient when one model and toolkit can complete the task.
Use Agent Service under agentscope.app to manage user-owned resources and persistent sessions. Storage backends include Redis and SQLAlchemy databases such as PostgreSQL, MySQL, and SQLite; workspace allocation can be configured per user, agent, or session.Integrate your own authentication to establish user identity. See Agent Service and Workspace Management for configuration.
Yes, from version 2.0.8. A2AAgent connects to remote A2A agents and translates their responses into AgentScope messages and events; the remote service controls its own models and tools.See the A2A guide and client/server example.
Configure the permission system to allow, deny, or request confirmation for tool calls. When confirmation is required, present the request to the user and return their decision through the human-in-the-loop events.To stop an ongoing run, use the separate interruption flow.
Install the official agentscope-skill to provide AgentScope guidance, API references, and examples. Give the assistant your installed SDK version and use the documentation index to find matching pages.Ask it to verify API names against that version and run the resulting example.
Yes. AgentScope also has TypeScript and Java implementations, each with its own repository and release schedule.Use the selected implementation’s documentation and examples. Their APIs are not interchangeable; this documentation describes the Python SDK.