Skip to main content
The reference docs cover how to use the platform. This section covers the why behind what you’re calling. Each topic starts broad and then layers in details, so you can form an accurate high-level mental model of the complete system. You can read everything in order or cherry-pick whatever’s relevant to what you’re building:

Fundamentals

How LLMs work

Text in, probabilities out, repeat in a loop.

Tokens & tokenization

Subword chunks of input and output.

Context windows

The model’s working memory and hard limits.

Inference parameters & sampling

How to produce predictable or creative outputs.

Context engineering

Structuring prompts for optimal results.

Inference metrics

TTFT and TPS, how fast an LLM feels.

Core capabilities

Function calling & tool use

Model plans, your code runs the tool.

Structured outputs & JSON mode

Force schema-valid output via constrained decoding.

Customization & deployment

Fine-tune vs. prompt

Hone a model for a particular task.

Choosing a deployment option

Serverless, dedicated endpoints, or containers.

Quantization

The JPEG quality slider for a model.