Prompt Templates

Prompt templates help you build reusable, parameterised prompts and message lists.

---

PromptTemplate

f-string style single prompt.

from synapsekit import PromptTemplate

pt = PromptTemplate("Translate the following text to {language}:\n\n{text}")

prompt = pt.format(language="French", text="Hello, how are you?")
# → "Translate the following text to French:\n\nHello, how are you?"

---

ChatPromptTemplate

Build a List[dict] message structure for chat models.

from synapsekit import ChatPromptTemplate

cpt = ChatPromptTemplate([
    {"role": "system", "content": "You are a helpful {persona}."},
    {"role": "user",   "content": "Explain {topic} in simple terms."},
])

messages = cpt.format_messages(persona="science teacher", topic="black holes")
# → [
#     {"role": "system", "content": "You are a helpful science teacher."},
#     {"role": "user",   "content": "Explain black holes in simple terms."},
#   ]

Pass the result directly to any LLM's stream_with_messages():

from synapsekit.llm.openai import OpenAILLM
from synapsekit.llm.base import LLMConfig

llm = OpenAILLM(LLMConfig(model="gpt-4o-mini", api_key="sk-..."))

messages = cpt.format_messages(persona="chef", topic="pasta")
async for token in llm.stream_with_messages(messages):
    print(token, end="")

---

FewShotPromptTemplate

Render few-shot examples followed by the actual query.

from synapsekit import FewShotPromptTemplate

fsp = FewShotPromptTemplate(
    examples=[
        {"input": "2 + 2",   "output": "4"},
        {"input": "10 - 3",  "output": "7"},
        {"input": "3 × 3",   "output": "9"},
    ],
    example_template="Input: {input}\nOutput: {output}",
    suffix="Input: {question}\nOutput:",
)

prompt = fsp.format(question="5 + 8")
# → """
# Input: 2 + 2
# Output: 4
#
# Input: 10 - 3
# Output: 7
#
# Input: 3 × 3
# Output: 9
#
# Input: 5 + 8
# Output:
# """

---

PromptHub

Local filesystem prompt registry with versioning. Push, pull, and manage versioned prompt templates.

from synapsekit import PromptHub

hub = PromptHub()  # defaults to ~/.synapsekit/prompts/

# Push a prompt
hub.push("my-org/summarize", "Summarize: {text}", version="v1")
hub.push("my-org/summarize", "Please summarize concisely: {text}", version="v2")

# Pull by version
tpl = hub.pull("my-org/summarize:v2")
print(tpl.format(text="Hello world"))

# Pull latest version
tpl = hub.pull("my-org/summarize")  # resolves to v2

Directory layout

Prompts are stored as JSON files on the local filesystem:

~/.synapsekit/prompts/
  my-org/
    summarize/
      v1.json
      v2.json
    qa/
      v1.json

Methods

Method	Returns	Description
push(name, template, version, metadata)	Path	Save a prompt template
pull(ref)	PromptTemplate	Load a prompt; ref is "org/name:version" or "org/name" (latest)
list(org=)	list[str]	List all prompts, optionally filtered by org
versions(name)	list[str]	List all versions of a prompt

Custom hub directory

hub = PromptHub(hub_dir="/path/to/my/prompts")