GeniSpace glossary
Definitions in this glossary describe how terms are used in the GeniSpace product.
Agent
An AI runtime that can understand a request, use approved tools, evaluate results, and answer with evidence. GeniSpace exposes four product forms: conversational agents, task agents, GeniAssistant, and built-in agents.
Agent mode
The Chat mode that permits an agent to call authorized tools and perform actions. Ask mode answers questions without action-tool execution.
API Playground
The interactive Dataset page for testing insert, update, query, count, delete, vector search, and full-text search requests before using generated client code.
Assistant (GeniAssistant)
The platform assistant that helps users create and manage platform resources through authorized tools. It is distinct from a user-created conversational agent.
Built-in agent
An agent embedded in a product surface for a bounded purpose, such as Workbench Copilot, Knowledge Copilot, or GeniBot.
Checkpoint
Persisted runtime state that lets an agent or task continue a thread, resume after user input, or recover from an interruption. It is not the same as the visible chat transcript.
Compaction
Summarizing older model context to free context capacity while retaining goals, decisions, constraints, and important tool evidence. /compact starts manual compaction; automatic compaction can run near the context limit.
Context
The information available to the model for the current turn: recent messages, compacted summaries, instructions, selected memory, retrieved knowledge, attachments, and tool evidence. Context is finite and differs from the complete stored conversation history.
Conversational agent
An interactive agent optimized for multi-turn chat, streaming responses, tools, user confirmation, memory, knowledge retrieval, and context compaction.
Dataset
A schema-defined collection of business records managed by Dataset service. It supports structured queries and, when configured, full-text and vector search. A Dataset is separate from an Agent knowledge base.
Datasource
A connection to an external or managed business data system. Workbench components can bind directly to a Datasource or to a Dataset, depending on the application design.
Embedding
A numeric representation used for semantic similarity. A Dataset uses its configured embedding model consistently for writing and querying vectors. Multimodal embedding is selected explicitly when image or video semantics are required.
Embedded agent application
A bounded Agent experience hosted by another product. The host backend keeps the long-lived application credential and maps each external user to a stable user code; the browser receives only a short-lived application-scoped session.
Execution process
The operational timeline shown for an agent run, including decisions, tool calls, user-input requests, validation, and results. It is not a disclosure of the model's private reasoning.
Full-text search
Search based primarily on text terms and the text index. It is useful for names, identifiers, phrases, and literal evidence, and complements vector search.
GeniApp
A business application installed into a Space. GeniApps can combine pages, datasets, workflows, agents, permissions, and navigation, and are managed through Console.
Industry solution
An installable, localized solution package for a business scenario. It can provision applications, sample data, datasets, agents, workflows, dashboards, and tasks. Existing installations can receive an upgrade-safe draft for review and publication.
Knowledge base
A governed collection of processed documents used for retrieval-augmented answers. Knowledge retrieval returns supporting passages; it is separate from querying structured Dataset records.
Knowledge base — Standard and Advanced
Standard provides document organization, processing, search, Agent binding, and eBook. Advanced adds stable document identity, governed revisions, lifecycle, resource policies, audit, and controlled compilations where entitled.
Memory
Durable user or agent information selected for reuse across turns or sessions. Memory is managed independently from chat history and context compaction.
MCP
Model Context Protocol, a standard way to expose external tools and resources to an agent. Availability and authorization are still controlled by the platform and Space.
Operator
A reusable executable capability used in workflows or exposed as an agent tool. Operators have defined inputs, outputs, authorization requirements, and execution results.
RAG
Retrieval-augmented generation. The system retrieves relevant evidence from a knowledge source and supplies it to the model before it produces an answer.
Semantic search / vector search
Search that embeds the query and compares it with stored vectors to find semantically related records. It helps with concepts and cross-language matching but does not itself prove that a record satisfies every business constraint.
Space
The tenant and collaboration boundary for applications, agents, tasks, datasets, knowledge bases, permissions, subscriptions, quota, and usage. Always verify the active Space before creating or searching resources.
Structured query
An exact query over typed Dataset fields, filters, IDs, sorting, and pagination. Use it for known field values, counts, and deterministic conditions.
Task agent
An agent runtime optimized for a structured or background task with a defined input and output. It does not automatically inherit every interactive-chat behavior.
Tool
An authorized capability an agent can invoke. Tools can be platform-native, application-local, operator-backed, plugin-provided, or exposed through MCP.
Top-K
The maximum number of nearest search results returned. Top-K means “the K closest available records,” not “K records that satisfy the user's condition.” Results still need evidence-based qualification.
Version
A saved revision of a managed resource. In Workbench, saving creates or updates a draft and publishing makes the reviewed version live. Version history supports comparison and restoration.
Workflow
A visual automation composed of triggers, operators, control nodes, and data mappings. Workflows can run manually, on a schedule, or in response to an event.
Workbench
The low-code application editor and runtime for pages, navigation, responsive layouts, business components, data binding, actions, versions, publishing, and Copilot-assisted draft changes.