Glossary

Definitions of LibreFang-related terms and concepts.

A

Agent

A software entity that runs autonomously, capable of executing tasks, making decisions, and interacting with users or other systems.

API (Application Programming Interface)

Application programming interface that allows external programs to communicate with LibreFang.

A2A (Agent to Agent)

Protocol standard for Agent-to-Agent communication.

B

Bearer Token

Token format used for API authentication.

Benchmark

Performance testing used to evaluate system speed, throughput, and response time.

C

Capability

The scope of operations an Agent is authorized to perform, such as using specific tools or accessing specific resources.

Channel

Message platform adapter that allows Agents to communicate with users through different platforms.

CLI (Command Line Interface)

Command-line interface for interacting with LibreFang through the terminal.

Compaction

Session compression that reduces historical messages to control token usage.

D

Daemon

LibreFang daemon running in the background.

DAG (Directed Acyclic Graph)

Directed acyclic graph, the execution model for workflows.

E

Ed25519

Asymmetric encryption algorithm used for signing and verification.

Environment Variable

Key-value pair used to store configuration information.

F

FangHub

LibreFang skill marketplace.

Fallback Provider

Backup LLM provider used when the primary provider fails.

G

GCRA (Generic Cell Rate Algorithm)

Generic cell rate algorithm used for rate limiting.

Groq

One of the LLM providers, known for fast inference.

H

Hand

LibreFang's autonomous capability package that can run independently and execute tasks.

HTTP Headers

Metadata of HTTP requests/responses.

K

Kernel

LibreFang's core component responsible for coordinating all subsystems.

Knowledge Graph

Structured data that stores entities and relationships.

L

LLM (Large Language Model)

Large language model such as Claude, GPT, etc.

LLM Driver

Adapter layer that LibreFang uses to interact with different LLM providers.

M

MCP (Model Context Protocol)

Model Context Protocol, a standardized way to connect LLMs with external tools.

Memory

LibreFang's persistent storage system including sessions, vector search, and knowledge graphs.

Merkle Hash Chain

Merkle hash chain, a data structure used for audit trails.

Model Catalog

Model catalog containing all available models and provider information.

O

OFP (OpenFang Protocol)

LibreFang's peer-to-peer communication protocol.

OpenAI Compatible API

Interface compatible with OpenAI API format.

P

Prompt Injection

A security threat type that manipulates AI behavior through malicious input.

Provider

LLM provider such as Anthropic, OpenAI, Groq, etc.

R

RBAC (Role-Based Access Control)

Role-based access control.

REST API

Representational State Transfer API, standard architectural style for web services.

S

SSE (Server-Sent Events)

Server-sent events, used for one-way streaming from server to client.

Skill

Pluggable toolkit that extends Agent capabilities.

SSRF (Server-Side Request Forgery)

Server-side request forgery, a type of security threat.

Stable Mode

Agent running mode with fixed model and skill registry.

System Prompt

System prompt that defines Agent behavior and persona.

T

Taint Tracking

Security mechanism that tracks the flow of sensitive information.

TOML

Tom's Obvious, Minimal Language, configuration file format.

Tool

Function that an Agent can invoke, such as file reading, network requests, etc.

Trigger

Event source that triggers workflow execution.

V

Vector Embedding

Vector embedding that converts text to numerical vectors for semantic search.

Vault

Credential vault for securely storing sensitive information.

W

WASM (WebAssembly)

WebAssembly, a secure sandbox runtime environment.

WebSocket

Full-duplex communication protocol supporting real-time bidirectional interaction.