# Glossary

#### **Agent**

An AI-powered interface (LLM-based) that interprets user instructions, decides what tools to use, and coordinates execution through MCP servers.

***

#### **Role**

A configuration profile that defines an agent’s personality, capabilities, tool access, and behavioral limits. Each session in PILSO runs under one role.

***

#### **Session**

A live runtime context where a role-based agent interacts with MCP tools and user prompts. Each session includes memory, logs, and a full execution history.

***

#### **MCP (Model Context Protocol)**

The protocol used in PILSO to standardize how agents interact with backend tools. MCP servers expose task-specific APIs for agents to call, such as compiling contracts or checking balances.

***

#### **MCP Server**

A standalone service that performs a specific task — such as signing, compiling, or resolving data. Each server exposes one or more **tools** to be called by agents.

***

#### **Tool**

A single callable function exposed by an MCP server. Examples: `erc20.transfer`, `solc.compile`, `metamask.sign`.

***

#### **LLM (Large Language Model)**

A language-based AI system (like Claude, GPT-4, or Deepseek) that processes natural language instructions and generates structured responses.

***

#### **Tool Call**

A structured request made by an agent to an MCP tool, typically with JSON-formatted arguments. Used to execute real-world actions like compiling contracts or preparing a transaction.

***

#### **Unsigned Transaction**

A raw transaction payload prepared by an MCP tool (e.g. `metamask-mcp`) that must be signed by the user’s wallet before it can be sent to the blockchain.

***

#### **Wallet-as-Signer**

A security model where all blockchain signing is done through the user’s own wallet (e.g. MetaMask). PILSO does not hold or access private keys.

***

#### **Prompt**

The natural-language input provided by the user to the agent. The prompt is interpreted by the LLM and translated into tool calls.

***

#### **Guardrails**

Instructions embedded in a role that tell the agent what *not* to do (e.g., “Never transfer tokens”, “Always confirm before signing”).

***

#### **Config File (`pilso.config.json`)**

Defines global system behavior for your local PILSO runtime — including LLM provider, tool endpoints, logging level, and auto-approval rules.

***

#### **Roles File (`pilso-roles.json`)**

Contains definitions for all available agent roles, including tool access, goals, and behavioral guardrails.

***

#### **CLI**

The command-line interface used to run and manage PILSO locally (`npx pilso`). All sessions, configs, tools, and logs are accessible from here.

***

#### **Session Log**

A JSON file generated per session that contains full prompt history, tool calls, responses, and any signing steps. Used for auditing or replaying agent behavior.

***

#### **$PILSO**

The native token of the PILSO OS ecosystem. Used for incentivizing infrastructure coordination, access control, and governance (not yet live).