// Section 1 — The Model · 2 MIN READ

[✓] VERIFIED MANUAL ENTRY — This concept has been rewritten from primary sources and is legally cleared for production.

Model

The compiled, frozen parameters that calculate next-token probability distributions. A model is completely stateless and pure.

A model is a mathematical file containing billions of floating-point weights (parameters). It acts as a stateless, pure function: it accepts a sequence of input tokens and returns a probability distribution for the next token.

Day-to-day developers often conflate the model with the harness (the application code surrounding it). Mechanically, a model has no agency:

  • It cannot read your workspace files.
  • It cannot execute console commands.
  • It cannot remember prior requests or yesterday's conversation.
  • It has no internal timer or clock.

All apparent "agentic" capabilities are orchestrations run by the harness, which calls the model in a loop, parses its output strings, and feeds tool results back as fresh input tokens.

Configuration in code

When setting up APIs, always select the model tier based on the task requirement (heavy planning vs fast execution):

// Configured in a Node.js harness
const plannerModel = "claude-3-5-sonnet"; // Large parameter count, high reasoning
const parserModel = "claude-3-5-haiku";   // Smaller parameter count, low latency

# AVOID

Do not blame the model for failures in state preservation or loop execution. That is a harness responsibility.

  • Avoid: "The model forgot my previous file modifications."
  • Write: "The harness failed to append the file changes back into the context window for the next turn."

# USAGE

Developer A: "Why is this model not executing my git commit command?" Developer B: "Because the model is just a stateless parameters file. You need to verify if the client harness parsed the tool call correctly and ran the child shell process on your terminal."

// SEE_ALSO

ParametersThe internal floating-point weights and biases of a neural network, optimized during training, that define the model's behavior.
HarnessThe client-side application code that drives the model, parses tool calls, maintains conversation logs, and executes command operations on the local machine.
InferenceThe execution phase where a trained model processes input tokens to generate output tokens.
StatelessThe architectural characteristic of AI models where each API request has no memory of prior queries, requiring the client to send the entire conversation history in every call.

// SOCRATIC_VALIDATION

Interactive Concept Quiz

QUESTION 1 OF 3SCORE: 0/3

[EDIT_THIS_TERM_ON_GITHUB ↗]