Glossary

Workspace​

The smallest billing unit for a subscription plan on the Asgard AI SaaS platform.

Project​

Under a workspace, you can create as many projects as your subscription plan allows. Resources inside a project, such as knowledge bases, settings and apps, are shared.

Collection​

A collection of workflows. Think of it as a workflow set.

Workflow​

A flow built by connecting processors together, with a start and an end.

Processor​

The smallest unit of processing. Flow types are Entry, Exit and Router; message types are Push Message and Listen Message; action types are Update Context and Execute Script; model types are Generate Embedding, LLM Completion and Stream LLM Completion; query types are SQL and Retrieve Knowledge; the API type is HTTP Request; and Validate Request and Response belong to the Automation Tool.

Environment​

The environment used for version control. A collection is created under the main environment by default.

Knowledge Base Storage​

Segments​

Loaders​

Scheduled, automatic knowledge ingestion.

Workspace Owner​

Entry​

Where a workflow starts.

Exit​

Where a workflow ends, or the handover point to another workflow.

Push Message​

Sends a message straight out as the response.

Listen Message​

Waits for message input.

Router​

Decides which path the workflow takes, based on If, Else If and Else conditions.

Update Context​

Updates content and initialises variables.

Execute Script​

Runs a custom script.

Generate Embedding​

Turns text into a vector embedding.

LLM Completion​

Calls a large language model and produces structured output, to support a decision in the flow or to generate natural language.

Stream LLM Completion​

Calls a large language model and produces streamed natural-language output.

SQL​

Queries a database.

Retrieve Knowledge​

Searches a knowledge base in natural language.

HTTP Request​

Sends a request over HTTP, for calling an external API, a webhook and so on.

Validate Request​

Defines the input format of a tool.

Response​

Returns the response.

Indexer​

Index processing.

Completion Model​

A completion model is one way of using an LLM: it takes a prompt and generates the text that follows it.

Embedding Model​

An embedding model is another way of using an LLM: it converts text into numeric vectors so that a computer can compare, search and classify it.