Sherlock definitions of terms on CloudFerro Cloud
The console uses similar names for resources that do different jobs. Knowing the difference helps you decide whether to change an assistant, its search agent, or the documents used to answer a question.
The terms below follow the screens in the introduction. They also distinguish browser testing from API requests made by your own code.
Term |
Meaning |
|---|---|
Organization |
The account structure through which users access Sherlock and create projects. |
Project |
The selected workspace for resources such as assistants, agents, knowledge bases, and API keys. Its name and ID appear in Overview. |
Model |
The model selected to generate a response, process an image, or create text embeddings. Use a model suited to the request type. |
Prompt |
The question or task sent to a model or assistant. |
Instructions |
The configured guidance for an assistant, such as its purpose and the style of its answers. |
AI Assistant |
A saved configuration with a model, instructions, and any connected search agents. Configure it under Build and test it under Test Playground. |
Knowledge base |
A resource containing documents that search agents use to provide context for assistants. |
AI Agent / search agent |
In the documented creation form, a resource connected to one knowledge base and an embedding model. Its settings control document retrieval. |
Embedding |
A numerical representation of text used for semantic search: finding related meaning rather than only matching words. |
Retrieved results |
The number of documents retrieved per query by a search agent, as described in its configuration form. |
Conversation history limit |
The number of previous messages used as additional context for retrieval by a search agent. |
Conversation |
An exchange of prompts and answers in the Playground. The assistant view provides a Conversations list. |
Token |
A unit of text processed by a model. Consumption distinguishes input tokens from output tokens. |
API key |
A project credential used to authenticate programmatic requests. It is a secret, unlike a project or resource ID. |
Model ID |
The identifier passed in an API request to select a model. Obtain available identifiers from the models endpoint. |
OCR |
Optical character recognition: extracting text from documents or images. Sherlock provides a dedicated OCR page in Test Playground. |
How document-based answers fit together
A knowledge base supplies documents. A search agent uses its embedding model and retrieval settings to find relevant information. An assistant uses a generative model and instructions to respond, with connected agents providing document context. This pattern is commonly called retrieval-augmented generation (RAG).
Creating an empty knowledge base does not add any source documents. An agent’s Active status also does not mean that files have finished processing. Check the file counts and processing status before expecting document-based answers.
What to do next
Prepare account access with Set up Sherlock AI.
Use Projects and Overview to create or select the workspace for the examples.