The Glossary

Learn the terms. Then build the thing.

Plain-English definitions of the concepts behind AI agents, hardened Linux, and safe DeFi — no fluff, written by someone who ships.

AI Agents

Agent loop

The four-step cycle every AI agent repeats — perceive, reason, act, observe — until a goal is met or a stop condition fires.…

Tool calling

The mechanism that lets an LLM request an external function — with structured JSON arguments — which your code then executes and r…

stop_reason

The field in an LLM API response that tells you why the model stopped — tool_use means run a tool and continue; end_turn or max_to…

In-context memory

The conversation history passed to the model on every call — the agent's short-term memory, bounded by the context window (~200K t…

RAG

Retrieval-Augmented Generation — fetching relevant documents at query time and feeding them into the prompt so the model answers f…

Evals

Automated tests that measure whether a change made your agent better, not just different — a set of inputs with known-good expecta…

System prompt

The standing instruction you give a model on every call that sets its role, rules, and constraints — separate from the per-turn us…

Context window

The maximum amount of text — instructions, history, and tool results — a model can consider at once, measured in tokens (~200K for…

Multi-agent system

An architecture where several specialized agents coordinate — often a director delegating to workers — instead of one agent doing …

Orchestration

The layer that decides which agent or tool runs next, passes data between them, and handles failures — the 'conductor' of a multi-…

Observability

Logging and tracing everything an agent does — every step, tool call, token, and cost — so you can see what actually happened, not…

Vector store

A database that stores text as embeddings — numerical vectors — so you can retrieve the most semantically similar items to a query…

Prompt injection

An attack where untrusted content — a web page, document, or tool result — contains instructions that hijack the agent into doing …

Few-shot prompting

Giving the model a few example input/output pairs in the prompt so it learns the pattern you want, without any training.…

Function calling schema

The JSON schema that defines a tool's name, description, and expected inputs — what the model reads to decide when and how to call…

Fine-tuning

Further training a model on your own examples so it adopts a specific style, format, or behavior — as opposed to RAG, which feeds …

Tokens

The chunks of text a model reads and writes — roughly 3/4 of a word each — and the unit you're billed and rate-limited by.…

Chain-of-thought

Prompting a model to reason step by step before answering, which improves accuracy on multi-step problems.…

Guardrails

Constraints that keep an agent's actions within safe bounds — what tools it may call, what it may spend, and what requires human a…

Rate limiting

Capping how often an action can happen in a window — how many API calls, requests, or agent steps — to control cost and prevent ru…

Streaming

Receiving a model's output token by token as it's generated, instead of waiting for the whole response — how you get that typing e…

Multimodal

A model that understands more than text — images, audio, or documents — letting you feed it screenshots, photos, or scans alongsid…

Linux

Full-disk encryption

Encrypting an entire drive so its contents are unreadable without the key — on Linux, typically LUKS — protecting data if the mach…

UFW

Uncomplicated Firewall — Ubuntu's front-end to iptables that lets you set a default-deny inbound firewall with simple allow rules.…

DNS-over-TLS

Encrypting your DNS lookups over TLS so your ISP and network can't see which domains you resolve — configured via systemd-resolved…

SSH key auth

Logging into SSH with an Ed25519 key pair instead of a password — the first move in hardening remote access.…

AppArmor

A mandatory-access-control system that confines programs to only the files and capabilities they need, limiting blast radius if on…

fail2ban

A daemon that watches logs for repeated failed logins and bans the source IP, blunting brute-force attacks.…

LUKS

The Linux Unified Key Setup — the standard for full-disk encryption on Linux, encrypting a block device so it's unreadable without…

systemd-resolved

The systemd service that handles DNS resolution on modern Ubuntu — the piece you configure to enable DNS-over-TLS.…

Default-deny firewall

A firewall posture that blocks all inbound connections by default and only allows the specific ports you explicitly open.…

Sudo hardening

Tightening how elevated privileges are granted — least privilege, logging, and avoiding passwordless sudo — so a compromised accou…

Automatic security updates

Configuring the system to install security patches automatically so known vulnerabilities get closed without waiting for you.…

HTTPS

HTTP encrypted with TLS — the protocol that keeps your web traffic private and tamper-proof in transit, indicated by the padlock a…

Two-factor authentication

A second proof of identity beyond a password — an authenticator app, hardware key, or code — so a stolen password alone isn't enou…

Backup

A separate, recoverable copy of your important data, following the practice that a backup you haven't tested restoring isn't a bac…

Firewall

A filter that controls which network connections are allowed in and out of your machine, your first line of defense against networ…

Patch management

Keeping software up to date so known vulnerabilities get fixed — the unglamorous habit that prevents most real-world breaches.…

Least privilege

Giving each user, process, and service only the access it needs to do its job — nothing more — to limit damage if any one is compr…

Secrets management

Handling API keys, passwords, and tokens so they're stored encrypted, never committed to code, and rotatable when leaked.…

DeFi Safety

Token approval

Permission you grant a smart contract to move a specific token from your wallet — the mechanism behind most connect-wallet flows, …

Honeypot

A token you can buy but not sell — a scam contract with a hidden sell tax or transfer block that traps your money.…

Self-custody

Holding your own crypto keys via a wallet you control, rather than leaving funds on an exchange that holds them for you.…

Seed phrase

The 12-24 word master key that regenerates your wallet — whoever has it controls the funds, so it must never be typed into a websi…

Rug pull

When a token's creators abandon the project and drain its liquidity, leaving holders unable to sell.…

Impermanent loss

The temporary loss a liquidity provider suffers when the price ratio of their pooled tokens diverges from when they deposited.…

Phishing

An attack that tricks you into revealing your seed phrase, keys, or a signed transaction by impersonating a trusted site or suppor…

Liquidity pool

A smart contract holding reserves of two tokens that lets people trade against it, with prices set by the ratio of the reserves.…

Gas fees

The transaction fee you pay to the network to have a transaction processed and included on-chain.…

Slippage

The difference between the price you expect on a trade and the price you actually get, caused by the trade moving the pool as it e…

Smart contract

A program deployed on a blockchain that executes automatically when its conditions are met — the code behind every DeFi protocol a…

Wallet drainer

Malicious code — often a fake site or signed transaction — that tricks you into authorizing transfer of your tokens, emptying the …

Cold storage

Keeping crypto keys completely offline — on a hardware wallet or paper — so no connected device can compromise them.…

Front-running

When someone sees your pending transaction and inserts their own ahead of it to profit at your expense — common on public DEX memp…

Audit

A security review of a smart contract's code by specialists, looking for vulnerabilities before you trust it with funds.…

Mempool

The waiting area where pending transactions sit before a validator includes them in a block — visible to everyone, which enables f…

Private key

The secret cryptographic key that controls a wallet and authorizes spending — derived from and protected by your seed phrase.…

Hardware wallet

A physical device that stores your private keys offline and signs transactions without the key ever touching an internet-connected…

Sandwich attack

A type of front-running where a bot buys before your trade and sells after it, profiting from the price move your own trade causes…

Yield farming

Providing liquidity or staking assets to a protocol in exchange for rewards, usually paid in the protocol's token.…

Stablecoin

A token engineered to hold a steady value, usually pegged to a currency like the US dollar, used as the low-volatility leg of DeFi…