Promptly

The prompt engineering tools.

Features

• build a prompt.
• run a prompt (with available model service).
• using arguments for prompt.
• commit prompt into commit repository to save a snapshot.
• testing a prompt (with loop or cases).

Deps

• python: 3.8 and above
  • use fastapi for backend
• typescript:
  • use vuejs v3 for front
• build tools: taskfile, poetry, pnpm
  • brew install go-task node swagger-codegen
  • pip3 install poetry
    • if it is too slow, try pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
  • npm install -g pnpm
    • if it is too slow, try npm config set registry https://registry.npmmirror.com
• db: mongodb
• env: docker
  • install docker desktop: https://www.docker.com/products/docker-desktop/
  • use image container for db
• (DEPRECATED) gui: pysimplegui

Run

Install & Run [recommand]

• task install
• task dist
• task run

task is coming from Taskfile, you must install task. see Taskfile for more details. see Install to install task.

Build & Dev [for developer only]

• task install
• task dev
• task run

Run With Docker [WIP]

run bellow

# start db
task db-start

# start promptly
...

visit localhost:8000 to use.

Data

Run task data-dump to dump your data into data/mongodb.zip.

While put file to data/mongodb.zip, run task data-load will load your data with db restart.

Config

LLM api token is sensitive, so we only put a separate file for it.

• run task config to generate config file config/llm.yaml.
• edit config/llm.yaml with your own info.
• by default, task run will load config/llm.yaml.

Since some areas are block by OpenAI, if you can not access LLM service, config your proxy at first. e.g.

export ALL_PROXY="socks5h://0.0.0.0:10888"