Glossary
Terms used across this documentation, aligned for readers from different roles.
- Agent — A packaged capability for a class of tasks or roles, combining model, instructions, skills, and tools; can be instantiated many times.
- Template — A preset or reusable agent blueprint for quickly building similar capabilities.
- Instance — A runnable object created from a template or configuration, with its own state (e.g. running, stopped).
- Skill — A packaged step or domain unit that multiple agents can mount.
- Tool — An extension for talking to external systems or channels, e.g. messaging, search, or business APIs.
- Resource — Uploaded or connected documents and data for Q&A, screening, or RAG.
- Task — An asynchronous or scheduled run record for tracking duration and results. The left-hand task list is used to run, pause, resume, stop, and view history.
- Agentic assistant — The built-in general chat agent for everyday Q&A, drafting, search, analysis, and collaboration, and for guiding natural-language creation or changes to dedicated agents and related configuration.
- Community template — A verified agent published to Create Agent and the Resource Plaza so others can select and launch it directly.
- Public access URL — A dedicated URL or share link for a public agent. Recipients open it and start using the agent without signing in first.
- Sleep and wake — An agent can sleep to stop conversation, then wake when needed.
- LLM Wiki — Source materials organized into a reviewable, publishable, maintainable Wiki, not only a retrieval knowledge base.
- OpenClaw — Module to connect Feishu, WeChat, DingTalk, and other collaboration and reach channels; what you see in the console is authoritative.
- MCP — Model Context Protocol–style connections for extending the tool ecosystem; configuration follows product documentation.
- Long-term memory — User-authorized preferences or stable information that can persist across sessions. Currently supported only for the Agentic assistant; other agents do not support it yet.