Module context_budget 

Context-window budgeting: estimating prompt token cost, deciding whether a request fits, and clamping the completion length to what remains.

Enums

Verdict

The outcome of assessing a prompt against a context window.

Constants

BUILTIN_CONTEXT_WINDOW

The context window Apple’s built-in model serves: fixed at 4096 tokens per session. The single source for every consumer — the runtime facade’s budget, the apple-foundation scanner’s record hint, and this module’s policy — so the value cannot drift apart.

COMPLETION_FLOOR

The minimum number of tokens reserved for the completion when deciding if a prompt fits the window.

Functions

assess

Decide whether a prompt of prompt_characters fits window, and clamp the completion length to the space that remains.

effective_window

The effective context window for record, honoring a per-runtime policy. A built-in model has a fixed window; other runtimes derive it from the record’s declared context length (and, for Ollama, a caller override).

estimated_tokens

A rough token estimate for characters of text (about four characters per token).

prompt_characters

Count the characters a chat/completion payload will send: the content of each message plus a top-level prompt string, if present.

stored_context_length

The user-set context length stored in the record’s parameter values, if any.