Raggedy

A refreshingly simple way to chat with LLMs/VLMs programmatically.

Installation

pip install raggedy

On Linux, you may also need to install libegl1.

Usage

Basic single message

from raggedy import chat

res = chat(to="ollama", model="llama3.2").message("Hello!")

print(res) # 'Hello! How can I assist you today?'

Message with files and streaming

c = chat(to="ollama", model="llama3.2-vision")

c.attach("test.png") # See below for supported file extensions

for chunk in c.message_stream("Describe this image."):
    print(chunk, end="", flush=True)

Multi-turn chatting (context memory)

c = chat(to="ollama", model="llama3.2")

print(c.message("My name is Evan. Please remember that."))
# 'I will make sure to remember your name, Evan! ...'

print(c.message("Why is the sky blue?"))
# 'The reason the sky appears blue is due to a phenomenon...'

print(c.message("What's my name again?"))
# 'Your name is Evan! I remember it from our conversation at the start.'

PDF attachments

Attach all pages

c = chat(to="ollama", model="llama3.2")

c.attach("test.pdf")

res = c.message("What are the contents of this PDF?")

Attach one page by direct text extraction

c = chat(to="ollama", model="llama3.2")

c.attach("test.pdf", page=0) # first page (0-indexed)

res = c.message("Describe this page from a PDF.")

Attach one page by rendering to an image

If the PDF page contains complex formatting, you can render to an image to preserve it:

c = chat(to="ollama", model="llama3.2-vision")

c.attach("test.pdf", page=0, as_image=True)

res = c.message("Extract the table in this image as Markdown.")

Parameters

chat(to, model, temperature?, num_ctx?) -> Chat

to: str

• Allowed values:
  • ollama: ensure Ollama is running and pull models in advance.
  • openai: not implemented yet; work in progress.
  • gemini: not implemented yet; work in progress.

model: str

• The model name to talk to. For example, "llama3.2" for ollama or "gpt-4o" for openai.

temperature?: float

• An optional parameter to specify temperature; 0 is objective, increase for creativity.

num_ctx?: int

• An optional parameter to specify the context size; for example, 32768 for a 32k window.

attach(str, page?, as_image?) -> Chat

The return value is the same Chat instance (so you can chain calls).

filepath: str

• The path to the file to attach. It is your responsibility to ensure it exists and is valid.
• Currently supported file extensions are:
  • Image formats: .jpg, .png
  • Textual formats: .txt, .csv, .json(l), .xml, .md
  • Other formats: .pdf

page?: int

• An optional parameter to specify page number (for PDFs). Default is all pages as text.

as_image?: bool

• An optional parameter for rendering a PDF page as an image to attach.

message(message: str) -> str

message: str

• The message to send to the LLM/VLM. Upon calling, any attachments will be sent with this call.

message_stream(message: str) -> Iterator[str]

• Same as message() but returns an iterator yielding streamed tokens from the LLM/VLM.