Configuration

From Python

Because even genies need settings.

import wishful

wishful.configure(
    model="openai/gpt-5",          # Switch models - use litellm model IDs (default: "azure/gpt-4.1")
    cache_dir="/tmp/.wishful",     # Cache directory for generated modules (default: ".wishful")
    spinner=False,                 # Show/hide the "generating..." spinner (default: True)
    review=True,                   # Review code before execution (default: False)
    allow_unsafe=False,            # Disable safety checks - dangerous! (default: False)
    temperature=0.7,               # LLM sampling temperature (default: 1.0)
    max_tokens=8000,               # Maximum LLM response tokens (default: 4096)
    debug=True,                    # Enable debug logging (default: False)
    log_level="INFO",              # Logging level: DEBUG, INFO, WARNING, ERROR (default: WARNING)
    log_to_file=True,              # Write logs to cache_dir/_logs/ (default: True)
    system_prompt="Custom prompt", # Override the system prompt for LLM (advanced)
)

# Context radius is configured separately (it likes to be special)
wishful.set_context_radius(6)  # Lines of context around imports AND call sites (default: 3)

All Configuration Options:

Your wish, your rules.

Parameter	Type	Default	Description
`model`	`str`	`"azure/gpt-4.1"`	LLM model identifier (litellm format)
`cache_dir`	`str \| Path`	`".wishful"`	Directory for cached generated modules
`review`	`bool`	`False`	Prompt for approval before executing generated code
`spinner`	`bool`	`True`	Show spinner during LLM generation
`allow_unsafe`	`bool`	`False`	Disable safety validation (use with caution!)
`temperature`	`float`	`1.0`	LLM sampling temperature (0.0-2.0)
`max_tokens`	`int`	`4096`	Maximum tokens for LLM response
`debug`	`bool`	`False`	Enable debug mode (sets log_level to DEBUG)
`log_level`	`str`	`"WARNING"`	Logging level (DEBUG, INFO, WARNING, ERROR)
`log_to_file`	`bool`	`True`	Write logs to `{cache_dir}/_logs/`
`system_prompt`	`str`	(see source)	Custom system prompt for LLM (advanced)

Environment variables (loaded via python-dotenv)

The .env file: where wishes go before they’re spoken aloud.

All settings can also be configured via environment variables:

WISHFUL_MODEL or DEFAULT_MODEL - LLM model identifier
WISHFUL_CACHE_DIR - Cache directory path
WISHFUL_REVIEW - Set to "1" to enable review mode
WISHFUL_DEBUG - Set to "1" to enable debug mode
WISHFUL_UNSAFE - Set to "1" to disable safety checks
WISHFUL_SPINNER - Set to "0" to disable spinner
WISHFUL_MAX_TOKENS - Maximum tokens (integer)
WISHFUL_TEMPERATURE - Sampling temperature (float)
WISHFUL_CONTEXT_RADIUS - Context lines around imports and call sites (integer)
WISHFUL_LOG_LEVEL - Logging level (DEBUG, INFO, WARNING, ERROR)
WISHFUL_LOG_TO_FILE - Set to "0" to disable file logging
WISHFUL_SYSTEM_PROMPT - Custom system prompt
WISHFUL_FAKE_LLM - Set to "1" for deterministic stub generation (testing)

Safety

Safety checks block os, subprocess, sys imports; eval/exec; and open in write-ish modes unless allow_unsafe=True.
Tests flip allow_unsafe=True by default, so don’t blindly copy test fixtures into production code.

Logging

log_level & log_to_file piggyback on wishful.settings.
Logs live under <cache_dir>/_logs/YYYY-MM-DD.log when file logging is on (enabled by default; stays in the sandboxed cache dir).

Want to see the knobs in action? Pair this with the CLI.