prompthub-my 0.3.0

Prompt version control system built on SQLite

PromptHub

Prompt version control system built on SQLite. Store, version, diff, and roll back LLM prompts — no external services required.

from prompthub import Storage

store = Storage("prompts.db")
prompt = store.create_prompt("my-prompt")
prompt.add_version([("system", "You are helpful."), ("user", "Hello")])

Installation

pip install prompthub                    # core only — zero dependencies
pip install prompthub[langchain]         # + LangChain adapter
pip install prompthub[openai]            # + OpenAI tokenizer
pip install prompthub[anthropic]         # + Anthropic tokenizer
pip install prompthub[huggingface]       # + HuggingFace tokenizer
pip install prompthub[all]              # everything

Features

• Version control — every add_version() stores a compact diff (Insert/Delete/Replace operations); full snapshots every N versions
• Rollback — soft rollback (new version with old content) or hard rollback (delete history)
• Diff — line diff, char diff, structured message diff between any two versions
• Tags — label prompts with custom PromptTag values; filter with a boolean DSL
• Token counting & cost estimation — attach model tags, count tokens, compute cost per 1M tokens from OpenRouter tariffs
• LangChain adapter — wrap any Prompt in LangChainPromptAdapter to work with ChatPromptTemplate
• Zero dependencies — SQLite only; LLM integrations are optional extras

Quick Start

from prompthub import Storage, PromptTag
from prompthub.core.tokenizers.openai_tag import OpenAIModelTag

store = Storage("prompts.db")

# Create prompt with initial version
prompt = store.create_prompt(
    "assistant",
    messages=[("system", "You are concise."), ("user", "Hi!")],
    tags=[PromptTag("production")],
    model_tags=[OpenAIModelTag("gpt-4o")],
)

# Add a new version
prompt.add_version([("system", "You are very concise."), ("user", "Hi!")])

# Diff two versions
diff = prompt.compare_versions_structured(1, 2)

# Soft rollback to version 1 (creates version 3)
prompt.rollback(target_seq=1)

# Fetch latest content as raw messages
messages = store.fetch_prompt("assistant")

# Fetch and work with LangChain
from prompthub.adapters import LangChainPromptAdapter
lc = LangChainPromptAdapter(store.get_prompt("assistant"))
template = lc.get_version_content(2)   # -> ChatPromptTemplate

Project Structure

prompthub/
├── __init__.py              # Public API: Storage, Prompt, PromptGroup, PromptTag, ModelTag
├── adapters/                # High-level adapters (user-facing)
│   └── langchain.py         # LangChainPromptAdapter
├── facade/
│   ├── storage.py           # Storage — main entry point
│   ├── prompt.py            # Prompt — versions, tags, diff, rollback
│   └── prompt_group.py      # PromptGroup — bulk queries
├── core/
│   ├── domain/              # Domain models: operations, diff, tags, tariffs
│   ├── adapters/            # LLM serializers: LangChainAdapter + registry
│   └── tokenizers/          # Token counting: OpenAI, Anthropic, HuggingFace
├── infrastructure/
│   ├── pricing_gateway.py   # Fetch tariffs from OpenRouter API
│   └── tariff_manager.py    # Write tariffs to DB
├── repository/              # SQL layer: PromptRepo, query DSL, schema
└── search/
    └── filters.py           # Boolean filter DSL: FieldEquals, TagFilter, ...

License

MIT