Glossary

Quick lookup for General Bots terms. If you’re lost, start here.

A

A2A Protocol - Agent-to-Agent Protocol. Enables bots to communicate and delegate tasks to each other in multi-agent systems. Messages include request, response, broadcast, and delegate types.

ADD BOT - BASIC keyword to add a bot to the current session with triggers, tools, or schedules.

Argon2 - Password hashing algorithm used for secure credential storage. Makes brute-force attacks computationally infeasible.

Auto-Bootstrap - The automatic first-run process that installs and configures all dependencies: PostgreSQL, cache, storage, and LLM servers.

B

BASIC - The scripting language for General Bots dialogs. Inspired by the 1964 language, simplified for conversational AI. Powers all .bas scripts with keywords like TALK, HEAR, and LLM.

BM25 - Best Match 25. Sparse retrieval algorithm for keyword-based search. Used in hybrid RAG alongside dense (semantic) search.

BOOK - BASIC keyword to schedule calendar appointments.

Bot Memory - Persistent storage scoped to a single bot, shared across all users. Access with SET BOT MEMORY and GET BOT MEMORY.

Bot Package - A folder ending in .gbai containing everything to run a bot: scripts, documents, and configuration.

BotSession - The active conversation between user and bot. Tracks state, history, and context. Persists to database, cached for speed.

Bootstrap - Initial setup process that installs all dependencies automatically on first launch.

BROADCAST TO BOTS - BASIC keyword to send a message to all bots in the current session.

C

Cache - In-memory storage component for sessions, temporary data, and semantic caching. Provides sub-millisecond access times.

Collection - A folder of documents in .gbkb/ that becomes searchable knowledge. Each subfolder is a separate collection.

Chunking - The process of splitting documents into smaller pieces for embedding and retrieval. Default chunk size is optimized for context windows.

config.csv - The configuration file for each bot. Simple key-value pairs in CSV format. Lives in the .gbot/ folder.

Context - Information available to the LLM during a conversation. Includes history, knowledge base results, and user-provided context via SET CONTEXT.

Context Compaction - Automatic summarization of older conversation history to fit within token limits while preserving important information.

CREATE DRAFT - BASIC keyword to compose and save an email draft to the user’s mailbox.

CREATE TASK - BASIC keyword to create a task with assignee and due date.

D

DELEGATE TO BOT - BASIC keyword to send a task to another bot and optionally wait for a response.

Dense Search - Semantic search using vector embeddings. Finds content by meaning rather than exact keywords.

Dialog - A .bas script defining conversation flow. Contains BASIC code with keywords like TALK and HEAR.

Drive - Built-in S3-compatible object storage. Stores documents, templates, and uploads. Auto-installed during bootstrap.

E

Embedding - Text converted to numerical vectors for similarity search. Similar meanings produce similar vectors.

Embedding Model - Neural network that generates embeddings. Default is BGE, replaceable with any GGUF-compatible model.

Episodic Memory - Summaries of past conversations stored for long-term context. Automatically generated when conversations end.

Event Handler - BASIC code triggered by events. Use ON keyword with triggers like "login", "email", or cron expressions.

F

FIND - BASIC keyword to search database tables with filter criteria. Returns matching records.

FOR EACH - BASIC keyword for iterating over collections and query results.

G

.gbai - “General Bot AI” package folder. Contains the entire bot. Example: support.gbai/ becomes the bot at /support.

.gbdialog - Subfolder containing BASIC scripts. Must include start.bas as the entry point. Tools go in tools/ subdirectory.

.gbdrive - File storage configuration subfolder. Maps to Drive buckets for document management.

.gbkb - “Knowledge Base” subfolder. Each subdirectory becomes a searchable collection with automatic indexing.

.gbot - Configuration subfolder containing config.csv with bot settings.

.gbtheme - Optional UI customization subfolder for CSS, images, and HTML templates.

General Bots - Open-source enterprise conversational AI platform. Combines LLMs with structured dialogs, knowledge bases, and multi-channel support.

GET - BASIC keyword to retrieve data from APIs, files, or session variables.

GET BOT MEMORY - BASIC keyword to retrieve persistent bot-level data.

GET USER MEMORY - BASIC keyword to retrieve cross-session user data accessible from any bot.

GraphQL - Query language for APIs. Supported via the GRAPHQL keyword for complex data retrieval.

H

HEAR - BASIC keyword to wait for and capture user input. name = HEAR stores the response in a variable.

Hot Reload - Automatic reloading of BASIC scripts when files change. No restart needed.

Hybrid Search - RAG approach combining dense (semantic) and sparse (keyword) retrieval using Reciprocal Rank Fusion.

HTMX - Frontend library used for dynamic UI updates without full page reloads.

I

INSERT - BASIC keyword to add records to database tables.

Intent - What the user wants to accomplish. Detected from natural language via LLM classification.

K

Keyword - A BASIC command like TALK, HEAR, or LLM. About 50+ available. Written in uppercase by convention.

Knowledge Base (KB) - Documents searchable by the bot. Organized in folders under .gbkb/. Activate with USE KB "foldername".

L

LiveKit - WebRTC platform used for video meetings in General Bots.

LLM - Large Language Model. The AI that powers natural conversation. Supports OpenAI, Anthropic, Groq, and local models via llama.cpp.

llama.cpp - C++ library for running LLM inference locally. Used for self-hosted model deployment.

Local-First - Architecture principle where everything runs locally by default. No cloud dependencies required.

M

MCP - Model Context Protocol. Standard format for defining tools that LLMs can call. Supported alongside OpenAI function format.

Memory - Data persistence system with four scopes: User Memory (cross-bot), Bot Memory (per-bot), Session Memory (temporary), and Episodic Memory (conversation summaries).

Model Routing - Dynamic selection of LLM models based on task requirements. Use USE MODEL "fast", "quality", "code", or "auto".

Multi-Agent - Architecture where multiple specialized bots collaborate on complex tasks.

Multi-Channel - Same bot works across WhatsApp, Telegram, Teams, Web, and other channels without modification.

N

No Forms - General Bots philosophy since 2017: people should converse, not fill forms. Conversations replace traditional UI forms.

O

ON - BASIC keyword to define event handlers for triggers, schedules, or webhooks.

OIDC - OpenID Connect. Authentication protocol handled by the Directory service (Zitadel).

P

Package Manager - Built-in system that installs bot packages. Drop a .gbai folder and it’s automatically loaded.

PARAM - Declares tool parameters. PARAM name, email means the tool needs these inputs. LLM collects them automatically.

PostgreSQL - The database for General Bots. Stores users, sessions, messages, and bot configuration. Auto-installed and auto-configured.

POST - BASIC keyword to make HTTP POST requests to external APIs.

Pragmatismo - Brazilian software company that created and maintains General Bots.

Q

Qdrant - Vector database for semantic search at scale. Optional component for large knowledge bases.

R

RAG - Retrieval-Augmented Generation. Pattern where relevant documents are retrieved and provided to the LLM as context.

Reranking - Optional LLM-based scoring of search results for improved relevance. Adds latency but improves quality.

Rhai - Rust scripting engine that powers the BASIC interpreter. Sandboxed and safe.

RRF - Reciprocal Rank Fusion. Algorithm for combining rankings from multiple search methods in hybrid RAG.

RUN PYTHON / JAVASCRIPT / BASH - BASIC keywords to execute code in sandboxed environments.

S

SAVE - BASIC keyword to write data to CSV files or database tables.

Script - A .bas file with BASIC code. start.bas is the entry point; other scripts are tools or utilities.

Semantic Cache - Caching system that matches similar (not just identical) queries to reuse LLM responses.

Semantic Search - Finding content by meaning rather than exact keywords. Powered by embeddings and vector similarity.

SEND MAIL - BASIC keyword to send emails with optional HTML and attachments.

Session - Active conversation state between user and bot. Expires after inactivity (default 30 minutes).

Session Memory - Temporary storage for the current conversation. Access with SET and GET.

SET - BASIC keyword to store values in session variables or update database records.

SET BOT MEMORY - BASIC keyword to store persistent bot-level data.

SET CONTEXT - BASIC keyword to add information to the LLM context. Influences all subsequent responses.

SET SCHEDULE - BASIC keyword for cron-based task scheduling. Accepts natural language like "every monday at 9am".

SET USER MEMORY - BASIC keyword to store cross-session user data accessible from any bot.

Sparse Search - Keyword-based search using algorithms like BM25. Excels at exact matches and rare terms.

SSE - Server-Sent Events. Used for real-time streaming of LLM responses.

Stalwart - Email server component providing IMAP/SMTP/JMAP support.

Suite - The complete General Bots workspace application with Chat, Drive, Tasks, Mail, Calendar, and other apps.

SWITCH - BASIC keyword for multi-way conditional branching.

T

TALK - BASIC keyword to send messages to the user. Supports text, markdown, and multimedia.

Template - Pre-built bot configuration in the templates/ folder. Copy and modify to create new bots.

Token - Unit of text for LLMs. Roughly 4 characters. Context windows are measured in tokens.

Tool - A .bas file the LLM can call automatically. Define with PARAM declarations and a DESCRIPTION. Place in the tools/ folder.

TRANSFER CONVERSATION - BASIC keyword to hand off the entire conversation to another bot.

U

UPDATE - BASIC keyword to modify existing database records.

USE KB - BASIC keyword to activate a knowledge base for semantic search. USE KB "policies" makes the policies collection searchable.

USE MODEL - BASIC keyword to switch LLM models. Options: "fast", "quality", "code", or "auto".

USE TOOL - BASIC keyword to enable a tool for LLM use. The AI determines when to call it.

User Memory - Persistent storage scoped to a user, accessible across all bots and sessions.

V

Vault - HashiCorp Vault. Secrets management service for storing credentials securely. Only VAULT_* environment variables are used.

Vector - Mathematical representation of meaning. Similar meanings produce similar vectors.

Vector Database - Database optimized for storing and searching embeddings. Qdrant is the default option.

W

WAIT - BASIC keyword to pause execution for a specified duration.

WEBHOOK - BASIC keyword to create HTTP endpoints that trigger bot actions.

WebSocket - Real-time connection for chat. Enables instant messaging without polling. Path: /ws.

Z

Zitadel - Identity and access management service. Handles authentication, users, and permissions.

Package Extensions

Common Confusions

“Do I need containers?” - No. BotServer installs everything directly or in optional LXC containers.

“What database?” - PostgreSQL, automatically installed and configured.

“What about scaling?” - Single server handles 1000+ concurrent users. Scale by running multiple instances.

“Is BASIC really BASIC?” - Inspired by BASIC, not strict implementation. Simplified and focused on conversations.

“Can I use TypeScript/Python/etc?” - BASIC handles conversation logic. Use RUN PYTHON/JAVASCRIPT for code execution, or integrate via REST API.

“Is it production-ready?” - Yes. Used in production since 2016, current Rust version since 2023.