Skip to main content

​
Overview

Coevolved treats prompts as first-class inputs to LLM steps. You can build prompts as:
  • A simple str
  • A Prompt template (with metadata and versioning)
  • A full PromptPayload (text or chat messages + IDs/hashes)

​
Prompt templates

Use Prompt when you want:
  • Structured prompt versioning (id, version)
  • Templating via Python str.format
  • Stable hashes for caching or trace grouping
from coevolved.core.prompt import Prompt

greeting = Prompt(
    id="support.greeting",
    version="1.0",
    template="You are a helpful assistant. User: {question}",
)

​
Prompt payloads

When you need full control, build a PromptPayload with either:
  • text (single user prompt)
  • messages (chat-style prompts compatible with tool calling)
llm_step will coerce your prompt builder output into a PromptPayload and will compute a prompt hash if missing.

​
Prompt versioning and hashing

Hashes are useful for:
  • Caching and memoization
  • Trace grouping across runs
  • Regression testing (“did the prompt change?”)
Keep your prompt IDs stable and bump versions intentionally. This makes it easier to trace behavioral shifts over time.

​
Next steps

LLMs and providers

Wire prompts into llm_step and provider calls.

core.prompt reference

Prompt and rendering API details.