Glossary

Technical terms used across Almost Autonomous, explained in plain language.


A

API — Application Programming Interface. A structured way for programs to communicate with each other.

B

Blast radius — How far the damage spreads when something goes wrong. A contained blast radius means failures stay local.

Broker — A service that sits between a client and multiple backends, routing requests to the right destination.

C

Cerbo GX — The system controller in a Victron installation — runs Venus OS, hosts Node-RED, and bridges all components via D-Bus and MQTT.

Chunker — A function that splits documents into smaller pieces for storage and search.

Chunks — Pieces of text (typically 200–500 characters) stored as vectors for similarity search.

Collection — A named group of vector embeddings in a database, searchable as one unit.

Container — An isolated environment that runs a program with its own filesystem, network, and permissions — like a sealed room.

Context — The maximum amount of text an AI can hold in working memory. When exceeded, older parts get compressed.

Cron — A scheduled task that runs automatically at set times on a Unix system — like a timer that triggers scripts.

D

D-Bus — An inter-process communication system in Linux that Venus OS uses to coordinate all Victron services and device control.

Docker — A platform for running containers — packages an application with its dependencies into an isolated, portable unit.

F

Function-calling — An AI model's ability to convert natural language requests into structured tool calls with correct parameters.

G

GPU — Graphics Processing Unit. Used for parallel computation in AI model inference and training.

I

Ingest — The process of loading, processing, and storing documents into a knowledge base.

J

J1939 CAN bus — A protocol used in heavy-duty vehicles and engines for standardised communication between electronic control units.

K

Knowledge base — A searchable database of documents that an AI agent can query for relevant context. See also: RAG.

L

Landlock — A Linux kernel feature that restricts which files and directories a process can access — filesystem sandboxing.

Lexical — Ghost's internal content storage format — a JSON tree of content nodes, replacing the older mobiledoc format.

M

mbox — A traditional email storage format. One large file containing all messages in a folder, concatenated together.

MCP — Model Context Protocol. A standard for giving AI models access to external tools and data sources.

Memory file — A persistent file an AI agent reads at session start to recall context from previous sessions.

Mobiledoc — An older Ghost content format — a JSON structure of sections and cards, replaced by lexical in Ghost 5+.

MQTT — A lightweight messaging protocol commonly used in IoT. Devices publish and subscribe to topic-based message streams.

N

Node-RED — A visual programming tool for wiring together hardware devices, APIs, and services. Runs as a service on the Cerbo GX.

O

Orchestrator — A lightweight AI model that parses requests and routes them to the right specialist — a dispatcher, not a worker.

P

PicoLog — A data logging device from Pico Technology that records sensor measurements over time.

R

RAG — Retrieval-Augmented Generation. Searching a knowledge base to add relevant context before an AI generates a response.

S

Seccomp — A Linux kernel feature that filters which system calls a process is allowed to make — limits what a program can ask the OS to do.

Siemens LOGO PLC — A compact programmable logic controller made by Siemens for small automation tasks.

Sparse embeddings — A vector representation that captures specific keyword importance, complementing dense semantic vectors.

SQLite — A lightweight database stored as a single file — no server needed, just read and write directly.

SSH — Secure Shell. A protocol for encrypted remote access to another computer's command line.

Sunk cost — The tendency to continue investing in something because of what you've already spent, not what it's worth.

System prompt — Instructions loaded before any conversation. They define the AI's personality, rules, and boundaries.

T

Tokens — The unit of text that AI models process — roughly one word or part of a word. More tokens means more cost and computation.

Tool call — A structured request from an AI model to use an external function — run a command, read a file, search the web.