Skip to main content

Glossary

A complete dictionary of terms you'll encounter in ClawDesk and AI in general. Written in plain English.


A

A2A (Agent-to-Agent)

A communication protocol that lets AI agents find each other and exchange messages. Think of it as a phone directory and messaging system for AI agents. Agents can discover what other agents are capable of and ask them for help.

ACP (Agent Communication Protocol)

The specific messaging format that agents use to talk to each other in ClawDesk.

Agent

An AI assistant configured with a specific role, personality, and set of tools. Like a specialist employee — a "Coding Agent" has instructions to focus on programming, while a "Writing Agent" focuses on content creation. Learn more →

Agent Flow

A predefined configuration that connects ClawDesk to external AI tools (like Claude Code, Codex, or Cursor). Flows define how messages are processed and by which system.

API (Application Programming Interface)

A way for software programs to talk to each other. When ClawDesk sends your message to Claude, it uses Anthropic's API. You don't need to understand APIs to use ClawDesk — the app handles it for you.

API Key

A unique code (like a password) that identifies you to an AI service. Example: sk-ant-api03-xyzABC123.... You get one by creating an account with the AI provider. How to get one →

Audit Trail

A tamper-proof log of everything that happens in ClawDesk. Uses a hash chain so nobody can secretly modify the records.

Automation

A task that runs on its own without manual intervention. Example: "Every morning at 8 AM, summarize my unread emails."


B

Backend

The "engine" of ClawDesk that runs behind the scenes — processes messages, manages agents, talks to AI providers. Built entirely in Rust.

Bedrock (AWS)

Amazon's enterprise AI service. Provides access to multiple AI models through your AWS account.


C

Channel

A connection between ClawDesk and an external messaging app (Telegram, Discord, Slack, etc.). Each channel routes messages to a specific agent. Set up channels →

CLI (Command Line Interface)

A text-based way to control ClawDesk by typing commands in a terminal. Example: clawdesk-cli message "Hello". The alternative to using the graphical desktop app.

Claude

An AI model made by Anthropic. Known for being helpful, harmless, and honest. One of the most popular AI providers used with ClawDesk.

Context

The information an AI model considers when generating a response. This includes your current message, conversation history, and the system prompt. More context = the AI understands your situation better, but costs more tokens.

Context Window

The maximum amount of text an AI model can process at once. Measured in tokens:

  • Claude: ~200,000 tokens
  • GPT-4: ~128,000 tokens
  • Local models: varies (2,000-128,000 tokens)

Crate

A Rust programming term for a module or package of code. ClawDesk is built from 45 crates that each handle a specific function. Non-developers don't need to worry about crates.

CRUD

Create, Read, Update, Delete — the four basic operations for managing data.


D

DAG (Directed Acyclic Graph)

A way of organizing tasks where each step leads to the next without going in circles. In ClawDesk, agent pipelines use DAGs to define multi-step workflows. Think of it as a flowchart.

Dashboard

The Overview page in ClawDesk that shows system status, active agents, connected channels, and recent activity at a glance.


E

Embedding

A way to convert text into numbers that a computer can compare and search. Used by ClawDesk's memory system to find relevant past conversations.

Extension

An add-on that give ClawDesk new capabilities. Browse and install from the Extensions page.


F

Failover

When one AI provider fails (rate limit, error, outage), ClawDesk automatically switches to another provider. This happens seamlessly — you don't notice it.

Frontend

The visual part of ClawDesk you interact with — the buttons, text fields, and pages. Built with React and TypeScript.


G

Gateway

ClawDesk's internal server that receives messages from channels, routes them to agents, and sends responses back. It's what makes everything connect together.

Gemini

Google's AI model. Good at processing long documents and multimodal content (text + images).

GGUF

A file format for local AI models. When you download a model to run locally, it's typically in GGUF format. You don't need to know this — ClawDesk handles it automatically.

GPT (Generative Pre-trained Transformer)

The AI model technology behind ChatGPT, made by OpenAI. GPT-4o is the latest version.


H

Hash Chain

A security technique used in ClawDesk's audit trail. Each log entry contains a code derived from the previous entry, creating an unbreakable chain. If anyone tries to tamper with the logs, the chain breaks.

Hexagonal Architecture

The design pattern ClawDesk uses internally. It separates the core logic from external connections (providers, channels, storage) so they can be swapped independently. Developers care about this; users don't need to.

Hot-Reload

When ClawDesk applies configuration changes without restarting. Edit your config file, and changes take effect immediately.


I

IPC (Inter-Process Communication)

How the ClawDesk frontend (what you see) communicates with the backend (the engine). Happens incredibly fast (~15 microseconds per call).


L

LLM (Large Language Model)

The AI technology that powers conversational AI. Examples: Claude, GPT-4, Llama, Gemini. These models are trained on vast amounts of text and can generate human-like responses.

Local Model

An AI model that runs directly on your computer, without needing internet. Set up local models →


M

MCP (Model Context Protocol)

A standard way to connect AI models to external tools and data sources." Think of it as USB for AI — a standard plug that works with many different tools. Learn more →

Memory

ClawDesk's system for remembering information across conversations. Uses both keyword search and AI-powered semantic search to find relevant past information.

Model

An AI "brain" — the trained software that generates responses. Different models have different strengths. Examples: Claude claude-sonnet-4-20250514, GPT-4o, Llama 3.1.


O

Ollama

Free software for running AI models on your computer. ClawDesk integrates with Ollama to give you easy access to local models. Set up Ollama →

Open Source

Software whose source code is publicly available for anyone to inspect, modify, and distribute. ClawDesk is open source, meaning you can verify exactly what it does with your data.

OpenRouter

A service that provides access to many AI models through a single API. Pay per use, with access to models from multiple providers.


P

Pipeline

A sequence of processing steps. In ClawDesk, an agent pipeline might be: receive message → search memory → add context → call AI → check safety → send response.

Plugin

An extension that adds new functionality to ClawDesk. Plugins can add tools, integrations, or custom processing to agents.

Prompt

The text you send to an AI model. Can be a question, instruction, or context.

Provider

A company or service that runs AI models. Examples: Anthropic (Claude), OpenAI (GPT), Google (Gemini), Meta (Llama via Ollama). Compare providers →


R

RAG (Retrieval-Augmented Generation)

A technique where the AI retrieves relevant documents before generating a response. Makes the AI more knowledgeable about specific topics by giving it access to your documents.

Rate Limit

A restriction on how many messages you can send in a given time period. Cloud AI providers enforce rate limits to manage demand. ClawDesk can also set rate limits on channels to prevent abuse.

Rust

The programming language ClawDesk's backend is built with. Known for being extremely fast, safe, and reliable. This is why ClawDesk uses only 20-40 MB of memory.


S

Skill

A specific ability given to an agent. Built-in skills include web search, file reading, code execution, and calculator. You can also create custom skills. Skills are like giving your AI employee the right tools for their job.

SochDB

ClawDesk's built-in database engine. Stores conversations, agent configs, and other data locally. Supports both regular data queries and vector search (for AI-powered memory). You don't need to manage it — it works automatically.

Streaming

When AI responses appear word by word as they're generated, rather than all at once. Makes the experience feel more interactive and responsive.

System Prompt

Hidden instructions given to an AI agent that define its role and behavior. The user doesn't see these, but they shape every response. Example: "You are a friendly customer support agent. Always be empathetic and provide step-by-step solutions."


T

Tauri

The framework used to build ClawDesk's desktop GUI. Tauri creates small, fast, native applications (2-8 MB) compared to alternatives like Electron (100+ MB). This is why ClawDesk starts instantly and uses little memory. Note: Tauri is only used for the desktop GUI — ClawDesk's CLI, tmux workspace, TUI, gateway, and daemon modes don't require Tauri at all.

Temperature

A setting that controls how creative (random) vs. precise (predictable) an AI's responses are. Scale of 0 to 1:

  • 0.0-0.3: Very precise, consistent (good for code, facts)
  • 0.4-0.7: Balanced (good for most conversations)
  • 0.8-1.0: Very creative, varied (good for brainstorming, stories)

Token

The smallest unit of text that AI models process. Roughly:

  • 1 token ≈ 4 characters or ¾ of a word
  • 1,000 tokens ≈ 750 words
  • API providers charge per token

Tool

A capability that lets an AI do something beyond text generation. Examples: searching the web, running code, reading files. Tools extend what agents can do.

TUI (Terminal User Interface)

A text-based visual interface that runs in the terminal (command line). ClawDesk has a TUI built with ratatui — featuring 10 screens, Vim keybindings, 4 themes, and full session multiplexing. Perfect for SSH sessions, cloud VMs, and Raspberry Pi deployments where a graphical desktop isn't available.

tmux Workspace

A terminal multiplexer layout built into ClawDesk's CLI. Creates 10 terminal windows that mirror the Tauri desktop app — Dashboard, Chat, Agents, Skills, Security, and more. Navigate with Ctrl-B + 0..9. Ideal for remote servers, SSH sessions, and cloud VMs. Launch with clawdesk tmux launch.


V

A way to find similar content by meaning, not just exact word matches. When ClawDesk's memory system searches for relevant past conversations, it uses vector search to find messages that are semantically similar to the current topic, even if they use different words.


W

WebSocket

A technology for real-time communication between your browser and a server. ClawDesk uses WebSockets to stream AI responses in real time, so you see words appearing as the AI generates them.

Workflow

A series of automated steps. Example: "When a message arrives → classify it → route to the right agent → send the response." Workflows in ClawDesk are built through the Automations page.