https://github.com/tolitius/jemma
jemma & her ai agents that build software
https://github.com/tolitius/jemma
Last synced: 3 months ago
JSON representation
jemma & her ai agents that build software
- Host: GitHub
- URL: https://github.com/tolitius/jemma
- Owner: tolitius
- Created: 2024-04-10T04:57:23.000Z (over 1 year ago)
- Default Branch: master
- Last Pushed: 2024-05-25T23:39:16.000Z (over 1 year ago)
- Last Synced: 2025-05-29T20:53:10.525Z (5 months ago)
- Language: Python
- Size: 4.81 MB
- Stars: 95
- Watchers: 5
- Forks: 10
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 
jemma
> hey, I am Jemma. I convert your thoughts to code
- [đ§Ŧ what I do](#-what-i-do)
- [am skechin'](#-am-sketchin)
- [đšī¸ can I play?](#%EF%B8%8F-can-i-play)
- [install me](#install-me)
- [convert ideas to code](#convert-ideas-to-code)
- [đ ī¸ how I do it](#%EF%B8%8F-how-i-do-it)
- [models](#models)
- [problems](#problems)
- [development](#development)
- [license](#license)
# đ§Ŧ what I do
I take an idea in a form of:
* a few words, such as "`Bill Pay Service`", "`2048`" or "`Kanban Board`"
* OR a text file with requirements
and I create a web based prototype đ
> _in fact I just created all three đ (so you can quickly see what I mean):_
after the prototype is built, I take feedback and refactor it.
## đ¨ am sketchin'
I also dabble in converting sketches to web app mockups:
```bash
$ jemma --prompt "Learning Portal" --sketch ~/tmp/sketch.png --build-prototype --claude
```
_this does require one or two hints of feedback, but I'm getting better_
# đšī¸ can I play?
of course!
## install me
```
$ pip install jemma
```
add a "`.env`" file from where I am going to be called from with API keys of your choice:
```bash
export ANTHROPIC_API_KEY=sk-ant-api03...
export OPENAI_API_KEY=sk-puk...
export REPLICATE_API_TOKEN=r8_ai...
```
ready to rock! :metal:
## convert ideas to code
```
$ jemma --prompt "Bill Pay Service" --build-prototype --claude
```
I will assemble a team who will build a prototype, open a browser with it, and wait for your feedback:
```bash
Claude đ§ claude-3-haiku-20240307 â
> Project Manager:
Dear Business Owner, in this meeting we'll work on creating requirements based on the đĄ idea
> Business Owner: đ creating detailed requirements ...đī¸
> Project Manager:
Dear Engineer, in this meeting let's put our heads together to build a prototype based on the requirements.
> Engineer: đĢ creating a prototype based on the requirements...
> Engineer: crafting css đ¨ (a.k.a. "visual beauty")
> Engineer: cooking javascript đŽ (a.k.a. "master of interactions")
> Engineer: creating html đ¸ī¸ (a.k.a. "the skeleton of the web")
prototype files created successfully:
- prototype/index.html
- prototype/app.js
- prototype/app.css
opened prototype in the web browser
tell me how to make it better >
```
# đ ī¸ how I do it
I rely on my team of project managers, business owners and engineers
yes... "AI Agents"
When I get an idea a Project Manager meets with a Business Owner to take it in and create a comprehensive set of requirements
then the Project Manager meets with an Engineer to build the idea based on these new requirements.
## models
I best work with Claude models, that is why my examples all end in "`--claude`":
```bash
$ jemma --prompt "Trivia Game" --build-prototype --claude
```
by default though I will call Ollama (llama3 model):
```bash
$ jemma --prompt "Trivia Game" --build-prototype
Ollama đ§ llama3:8b-instruct-fp16 â
```
here are the default models I would use:
| model param | default model|
| ---- | ---- |
| `--claude` | `claude-3-haiku-20240307` |
| `--openai` | `gpt-3.5-turbo`|
| `--ollama` | `llama3:8b-instruct-fp16`|
| `--replicate` | `meta/meta-llama-3-70b-instruct`|
| `--copilot` | `gpt-3.5-turbo`|
but you can override all of these with your (local, or not) models:
```bash
$ jemma --prompt "Trivia Game" --build-prototype --claude claude-3-opus-20240229
$ jemma --prompt "Trivia Game" --build-prototype --ollama dolphin-mistral:7b-v2.6-dpo-laser-fp16
$ jemma --prompt "Trivia Game" --build-prototype --openai gpt-4-turbo-preview
$ jemma --prompt "Trivia Game" --build-prototype --replicate meta/llama-2-70b-chat
$ jemma --prompt "Trivia Game" --build-prototype --copilot gpt-4
$ ...
```
> _but, at least for now, the best results are produced with **Claude** based models_
## problems
I am still learning, so some prototypes that I build might result in errors
this would especially be likely with non Claude based models
but, we are all learning, _and_ I love feedback:
```bash
tell me how to make it better > I see an error "app.js:138: uncaught TypeError: chordProgressionData.find(...) is undefined"
> Project Manager:
Dear Engineer, we have met with the user and received a valuable feedback. sudo make it better! đ ī¸
> Engineer: đĢ refactoring prototype based on the feedback...
> Engineer: âģī¸ crafting css đ¨ (a.k.a. "visual beauty")
> Engineer: âģī¸ cooking javascript đŽ (a.k.a. "master of interactions")
> Engineer: âģī¸ creating html đ¸ī¸ (a.k.a. "the skeleton of the web")
prototype files created successfully:
- prototype/index.html
- prototype/app.js
- prototype/app.css
opened prototype in the web browser
tell me how to make it better >
```
_you can check for / find errors in your browser console_
>_iff you know "how to HTML", you can help fix the code as well
_
>_it is often something simple: adding a CSS class, updating the "width", etc._
## development
in order to run from source
clone jemma:
```bash
$ git clone git@github.com:tolitius/jemma.git
```
and
```bash
$ cd jemma
$ python huddle.py --prompt "Code Editor" --build-prototype --claude
Claude đ§ claude-3-haiku-20240307 â
...
```
# license
Copyright Š 2024 tolitius
Distributed under the Eclipse Public License either version 1.0 or (at
your option) any later version.