Python SDK

The refrase Python package — install with pip install refrase.

Heads up: this is the v0.x deterministic SDK

The published refrasepackage applies prompt-pattern adaptations locally from a static rules table. It works offline and requires no API key, but it's not the same engine as the LLM-powered refrase.cc/enhance (which reads each model's full documentation and rewrites prompts via Claude Haiku 4.5 — that's the version our research measured).

v1.0 of the package will be rewritten as an API client that calls the Refrase API with an API key — same engine as the web app. Until then this page documents the v0.x behavior.

Installation

bash

pip install refrase

Quick Start

python

import refrase

result = refrase.adapt(
    prompt="Explain quantum computing in simple terms",
    model="claude-sonnet-4-20250514",
    task="chat",
    user_prompt="What is quantum computing?",
)

print(result.adapted)   # The optimized prompt
print(result.changes)   # List of changes made
print(result.model)     # Target model info

API Reference

`refrase.adapt(prompt, model, task, user_prompt)`

Adapt a system prompt for a specific model and task type.

Parameter	Type	Description
`prompt`	`str`	The system prompt to optimize
`model`	`str`	Target model identifier
`task`	`TaskType`	Optional. Task type for context-aware optimization
`user_prompt`	`str`	Optional. Example user message for calibration

Returns an AdaptResult object.

`refrase.list_models()`

Returns a list of all supported model configurations as dictionaries containing id, name, provider, and capabilities.

TaskType Enum

class TaskType(str, Enum):
    CHAT = "chat"
    CODE = "code"
    ANALYSIS = "analysis"
    CREATIVE = "creative"
    EXTRACTION = "extraction"
    SUMMARIZATION = "summarization"

Change Dataclass

@dataclass
class Change:
    type: str        # "add" | "remove" | "modify" | "reorder"
    description: str # Human-readable explanation
    before: str      # Original text (empty for additions)
    after: str       # Modified text (empty for removals)
    reason: str      # Why this change improves the prompt