Basic concepts: LLM, tokens, API, deployment
A short dictionary “to understand what is happening”, without unnecessary theory.
| Term | What is it in simple words |
|---|---|
| LLM | II, which generates text and code upon request. Can write confidently, but sometimes incorrectly (“hallucinations”). |
| Prompt | Your model instructions. Quality depends on clarity of context, purpose, limitations and format. |
| Tokens | “Pieces of text” with which the model calculates volume. Limits and prices depend on tokens. |
| API | The way programs “talk” to each other is a set of rules for integration. |
| REST API | The most common API format on the web: rules on top of HTTP. |
| API key | “Door Key” to the API. You cannot show the code on the front, in the public repo and screenshots. |
| Webhook | “Callback”: the service itself pulls your URL upon an event (payment, application, PR). |
| Deploy | Lay out the application so that it works via a link for users. |
| Preview / Production | Two environments: preview for testing, production for real users. |
| Environment variables | “Project settings outside the code”: addresses, modes, keys. |
| Git | “Save system” for code: fixes versions, allows you to roll back. |
| Database | Organized application data storage: users, orders, content. Without a database, the application “forgets” everything when rebooted. |
| No-code / Low-code | Visual platforms: no-code with almost no code, low-code - with a minimum of code. |
More about key concepts
LLM - what is it and why is it important to you
LLM (large language model) is a model that “essentially” learns to continue the text: the input is the text, the output is the next text. From this grow explanations, code, plans.
For vibe coding, it is important to understand one thing: LLM does not store the “truth” and does not store the “project” in the head. Every time it works with what is in the window context (input, files, rules, logs). Therefore, the quality is highly dependent on the context and its purity.
Tokens - “currency” and context limit
Tokens are the “text building blocks” with which the model operates. They are counted both in your input and in the response.
- Tokens = money (in API) and speed.
- Tokens = limit of “working memory” of the model.
Rule-on-your-fingers (hint) OpenAI): ~1 token ≈ 4 characters of English text, ≈ ¾ words.
API - what is it and why is it almost everywhere
API is a “contract” between programs: how one program requests data/action, and what it will receive in response.
In vibe coding, the API is important for two reasons:
- Most integrations (payments, mail, CRM, analytics, databases) are via API.
- Agents (Lovable, n8n, your backend) “talk” through HTTP requests and webhooks.
API keys and access tokens
The key/token is the “password” for the program. If it is stolen, the attacker can spend your limit/money, read/write data, and perform actions on behalf of your service.
Storage practice:
- Do not store keys in frontend (they are visible to the user).
- Keep in secret/secret store or at least in env vars on the deployment server/platform.
Webhooks - how services “call” each other
Webhook is when the service itself sends an HTTP request to your URL upon an event. B GitHub This is formulated directly: you select the URL and events, GitHub sends an HTTP request with the event data.
Important: webhooks must be verified (signature/secret), otherwise anyone can “fake the event”. GitHub recommends X-Hub-Signature-256 (HMAC-SHA256).
Deploy - what is it in simple terms
Deploy is when your code turns into a “live site/application via link”. In Vercel, a deployment is the result of a successful build, and each deployment receives a unique URL.
Ideal “beginner” environment scheme:
- Development: locally or in a dev environment
- Preview/Staging: test in a live environment, but not prod
- Production: what users see
Database - why is it needed
Database is the place where the application stores data: users, orders, posts, settings. Without a database, everything you entered disappears when you close the page.
Two main types for a beginner:
- Relational (SQL) —data in tables with a strict structure. Examples: PostgreSQL (Supabase), MySQL. Suitable for most MVPs.
- Document (NoSQL) —data in “documents” (as JSON). Example: Firebase Firestore. More flexible in structure, but more difficult to control integrity.
What is important to understand:
- Table —a set of records of the same type (for example, the “users” table).
- Row (row) —one entry (one user).
- Column —one field (name, email, registration date).
- Query — command to the database: “give me all the users registered today.”
In vibe coding, you usually don’t deploy the database yourself—you use BaaSservice (Supabase, Firebase), which provides a database + API + authorization out of the box.
(I deliberately do not delve into “architectures”, “microservices” and so on - it’s too early for a beginner. First you need results and the right habits.)