Glossary

Dashboard

Lets you manage, monitor and observe the entire AI stack in one web dashboard.

DeepFellow CLI (abbr: DFCLI)

The DeepFellow command line interface. Use it to install, configure, and manage DeepFellow Infra and Server.

DeepFellow Infra (abbr: DFInfra)

A self-hosted infrastructure stack purpose-built to run AI on your own hardware. It provides the compute foundation for LLMs, embedding models, MLAs, and MCP servers – giving you full ownership of the environment without relying on third-party cloud APIs.

DeepFellow Plugin (abbr: DFPlugin)

A hook that attaches to specific points in the server's request pipeline (e.g. chat completions, tool calls) to inspect or modify data as it flows through – before the request is processed, after a response is returned, or when an error occurs.

DeepFellow Server (abbr: DFServer)

The gateway to access MCP servers, Tools, DeepFellow Infra, Plugins, and Vector Stores. It also provides authorization and administration layers to the system, and is used for all the other application-related operations.

DeepFellow Threat Intelligence (abbr: DFTI)

A standalone component that connects to DeepFellow Server as a plugin to protect your AI system from modern, AI-specific attack vectors, like prompt injection, data poisoning and model inversion attacks.

Doc Chunker

Optional component that convert audio, video, PDFs, images, or Office documents into text chunks for vector stores.

MCP (Model Context Protocol)

An open-source standard for connecting AI applications to external systems.

Mesh

An arbitrary topology connecting multiple DeepFellow Infra instances in child-parent relations to gain efficiency via load-balancing.

Model

An underlying AI model (e.g. an LLM, embedding model, image-generation model, or speech model) that DeepFellow runs inference against. DeepFellow itself imposes no restriction on which models you use – you can install open-source models, connect proprietary ones, or bring your own.

Model Provider

The backend service that a model actually runs on or is sourced from – i.e., who or what is executing inference for that model. In DeepFellow Infra, each installed model is tied to a specific provider/service, which determines how it's installed, configured, and called.

NightShift

Plugin that runs automated, asynchronous backend processes to continually optimize your AI's intelligence and performance.

Organization

Top level of the DeepFellow hierarchy below the admin user. It's owned by exactly one user (owner_id) and typically represents a real-world entity like a company. An organization is the container for everything underneath it – projects, users/members, and organization-level API keys.

Project

One level below an organization. Projects are the actual working/operational unit – where models, custom endpoints, and usage limits get scoped.

Service

Models are organized under services. Each service is named after the backend, e.g. "ollama", or after the provider, e.g. "openai". Services group models from the same family.

Tool

A named, schema-defined capability, backed by an MCP server, that the model can choose to invoke mid-generation to fetch information or perform an action it couldn't do from its own training data alone.

Toolbox

A set of tools accessible from outside of your infrastructure using Model Context Protocol (MCP).

Vector Store

A searchable container of uploaded files where content is split into chunks and semantically indexed for retrieval. It's the storage/search layer that lets LLMs ground their answers in your own documents instead of just their training knowledge.