Prompt Templates

PromptTemplate is a reusable prompt with {variable} placeholders and an optional dedicated system prompt — useful for keeping prompt text out of handler logic.

Defining and rendering

from shakti.ai.templates import PromptTemplate

tpl = PromptTemplate(
    "Summarize this in {language}: {text}",
    system="You are a helpful summarizer.",
)
prompt = tpl.render(language="French", text="Hello world")

render(**kwargs) does str.format(**kwargs) under the hood and raises ValueError naming the missing variable if you forget one. tpl.variables() lists the placeholder names found in the template. defaults lets you pre-fill variables that are usually the same:

tpl = PromptTemplate("Translate to {language}: {text}", defaults={"language": "Spanish"})
tpl.render(text="Good morning")   # language defaults to "Spanish"

Registering with AI

ai.register_template("summarize_fr", tpl)
answer = await ai.ask("summarize_fr", text="Long article text...")

ai.ask(template_name, **variables) renders the named template and sends it through ai.chat(), using the template's own system prompt if it has one. Registering by name is optional — you can also just call tpl.render(...) yourself and pass the result to ai.chat() directly.

Built-in templates

shakti.ai.templates ships a few ready to use or register:

from shakti.ai.templates import SUMMARIZE, TRANSLATE, CODE_REVIEW, EXPLAIN_CODE, EXTRACT_JSON, SQL_QUERY

ai.register_template("summarize", SUMMARIZE)
ai.register_template("translate", TRANSLATE)

Template	Variables	Purpose
SUMMARIZE	text	concise summary
TRANSLATE	language, text	translation
CODE_REVIEW	language, code	quality/bug/improvement review
EXPLAIN_CODE	code	plain-language explanation
EXTRACT_JSON	keys, text	structured extraction, JSON-only output
SQL_QUERY	request, schema	natural language → SQL

answer = await ai.ask("code_review", language="python", code=snippet)

(after ai.register_template("code_review", CODE_REVIEW).)