2024-11-30 11:41:13 +00:00
|
|
|
"""
|
|
|
|
title: Gemini Protocol Tool
|
|
|
|
author: projectmoon
|
|
|
|
author_url: https://git.agnos.is/projectmoon/open-webui-filters
|
2024-12-13 22:08:01 +00:00
|
|
|
version: 0.2.1
|
2024-11-30 11:41:13 +00:00
|
|
|
license: AGPL-3.0+
|
|
|
|
required_open_webui_version: 0.4.3
|
|
|
|
requirements: ignition-gemini
|
|
|
|
"""
|
|
|
|
import re
|
|
|
|
import ignition
|
2024-11-30 19:54:08 +00:00
|
|
|
from ignition import RedirectResponse, SuccessResponse
|
2024-11-30 11:41:13 +00:00
|
|
|
from pydantic import BaseModel, Field
|
2024-11-30 19:54:08 +00:00
|
|
|
from typing import Optional
|
2024-11-30 11:41:13 +00:00
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
# From https://stackoverflow.com/questions/33404752/removing-emojis-from-a-string-in-python
|
|
|
|
def remove_emojis(text):
|
|
|
|
pattern = re.compile(pattern = "["
|
|
|
|
u"\U0001F600-\U0001F64F" # emoticons
|
|
|
|
u"\U0001F300-\U0001F5FF" # symbols & pictographs
|
|
|
|
u"\U0001F680-\U0001F6FF" # transport & map symbols
|
|
|
|
u"\U0001F1E0-\U0001F1FF" # flags (iOS)
|
|
|
|
"]+", flags = re.UNICODE)
|
|
|
|
return pattern.sub(r'',text)
|
|
|
|
|
|
|
|
def extract_title(content: str, fallback: str="") -> Optional[str]:
|
|
|
|
lines = [line.strip() for line in content.splitlines()]
|
|
|
|
title = next(filter(lambda line: line.startswith("#"), lines), None)
|
|
|
|
if title:
|
|
|
|
return remove_emojis(title.lstrip('#'))
|
|
|
|
else:
|
|
|
|
return fallback
|
|
|
|
|
2024-11-30 20:25:59 +00:00
|
|
|
def result_instructions(url: str, redirect: bool=False) -> str:
|
|
|
|
content_instructions = (
|
|
|
|
"Report the content to the user and answer their question."
|
2024-11-30 20:35:33 +00:00
|
|
|
"Do not make assumptions. This is the content you need. "
|
|
|
|
"Even if you think it's not the right content, report the "
|
|
|
|
"content anyway. Do not say you think it's the wrong content."
|
2024-12-01 11:12:59 +00:00
|
|
|
"You MUST use this content."
|
2024-11-30 20:25:59 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
return ("# Gemini Content Fetch Result\n"
|
|
|
|
f"Content was successfully fetched for the URL: {url}\n"
|
|
|
|
) + content_instructions + "\n\n"
|
|
|
|
|
|
|
|
def instructions(url: str, redirect: bool=False) -> str:
|
|
|
|
return result_instructions(url, redirect) + (
|
2024-12-01 11:12:59 +00:00
|
|
|
"Here are more instructions you must follow. "
|
2024-11-30 20:25:59 +00:00
|
|
|
"Render all Gemini links as Markdown links. Examples:\n\n"
|
|
|
|
" - [gemini://example.com](gemini://example.com)\n"
|
|
|
|
" - [a gemini capsule](gemini://example.com)\n"
|
|
|
|
" - [my personal capsule](gemini://example.com/personal)\n\n"
|
|
|
|
"A Gemini link starts with => on the line, and then the URL follows:\n\n"
|
|
|
|
" - `=> /page A Gemini Page`: A relative link to `/page` titled `A Gemini Page`\n"
|
|
|
|
" - `=> gemini://example.com/place` An absolute link with no URL title \n\n"
|
|
|
|
"When rendering relative links, always make them absolute links.\n"
|
|
|
|
"If the link has a title, render the link title verbatim:\n"
|
|
|
|
" - `=> gemini://example.com My Page` becomes `[My Page](gemini://example.com)`\n"
|
|
|
|
" - `=> gemini://example.com` becomes `[gemini://example.com](gemini://example.com)`.\n"
|
|
|
|
" - `=> gemini://example.com 🐂 My Page` becomes `[🐂 My Page](gemini://example.com)`\n\n"
|
|
|
|
"The Gemtext content is below in the code block."
|
|
|
|
)
|
2024-11-30 11:41:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
class GeminiFetcher:
|
2024-12-13 22:08:01 +00:00
|
|
|
def __init__(self, gemini_url: str, correct_urls: bool=False, web_proxy: str="", event_emitter=None):
|
2024-12-13 16:07:46 +00:00
|
|
|
self.original_url = gemini_url
|
|
|
|
self.current_url = None
|
|
|
|
self.correct_urls = correct_urls
|
2024-12-13 22:08:01 +00:00
|
|
|
self.web_proxy = web_proxy
|
2024-12-13 16:07:46 +00:00
|
|
|
self.event_emitter = event_emitter
|
|
|
|
|
|
|
|
async def fetch_event(self, done: bool):
|
|
|
|
if not self.event_emitter:
|
|
|
|
return
|
|
|
|
|
|
|
|
description = (f"Fetched: {self.original_url}" if done
|
|
|
|
else f"Fetching: {self.original_url}")
|
|
|
|
|
|
|
|
await self.event_emitter({
|
|
|
|
"type": "status",
|
|
|
|
"data": {
|
|
|
|
"status": "complete" if done else "in_progress",
|
|
|
|
"description": description,
|
|
|
|
"done": done,
|
|
|
|
},
|
|
|
|
})
|
|
|
|
|
|
|
|
async def error_event(self, message: str="There was an error"):
|
|
|
|
if not self.event_emitter:
|
|
|
|
return
|
|
|
|
|
|
|
|
await self.event_emitter({
|
|
|
|
"type": "status",
|
|
|
|
"data": {
|
|
|
|
"status": "error",
|
|
|
|
"description": message,
|
|
|
|
"done": True,
|
|
|
|
},
|
|
|
|
})
|
|
|
|
|
|
|
|
async def create_citation(self, title: str, content: str):
|
|
|
|
if not self.event_emitter:
|
|
|
|
return
|
|
|
|
|
|
|
|
document = content
|
2024-12-13 22:08:01 +00:00
|
|
|
link = (self.web_proxy.replace("<URL>", self.original_url)
|
|
|
|
if self.web_proxy != "" else self.original_url)
|
2024-12-13 16:07:46 +00:00
|
|
|
|
|
|
|
await self.event_emitter({
|
|
|
|
"type": "source",
|
|
|
|
"data": {
|
|
|
|
"document": [document],
|
|
|
|
"metadata": [{"source": title, "html": False }],
|
2024-12-13 22:08:01 +00:00
|
|
|
"source": {"name": title, "url": link},
|
2024-12-13 16:07:46 +00:00
|
|
|
}
|
|
|
|
})
|
|
|
|
|
|
|
|
async def complete_event(self, content: str):
|
|
|
|
title = extract_title(content, self.original_url)
|
|
|
|
await self.fetch_event(done=True)
|
|
|
|
await self.create_citation(title, content)
|
|
|
|
|
|
|
|
def correct_url(self, url: str) -> str:
|
|
|
|
if url.startswith("gemini://http://"):
|
|
|
|
match = re.match(r'gemini://http://(.+)', url)
|
|
|
|
if match:
|
|
|
|
return f"gemini://{match.group(1)}"
|
|
|
|
return url
|
|
|
|
|
|
|
|
if url.startswith("gemini://https://"):
|
|
|
|
match = re.match(r'gemini://https://(.+)', url)
|
|
|
|
if match:
|
|
|
|
return f"gemini://{match.group(1)}"
|
|
|
|
return url
|
|
|
|
|
|
|
|
if url.startswith("https://"):
|
|
|
|
match = re.match(r'https://(.+)', url)
|
|
|
|
if match:
|
|
|
|
return f"gemini://{match.group(1)}"
|
|
|
|
return url
|
|
|
|
|
|
|
|
if url.startswith("http://"):
|
|
|
|
match = re.match(r'http://(.+)', url)
|
|
|
|
if match:
|
|
|
|
return f"gemini://{match.group(1)}"
|
|
|
|
return url
|
|
|
|
|
|
|
|
if not url.startswith("gemini://"):
|
|
|
|
return f"gemini://{url}"
|
2024-11-30 11:41:13 +00:00
|
|
|
|
|
|
|
return url
|
|
|
|
|
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
async def fetch(self, prev_url: Optional[str]=None, redirects: int=0) -> dict:
|
|
|
|
await self.fetch_event(done=False)
|
|
|
|
gemini_url = (self.current_url
|
|
|
|
if self.current_url is not None
|
|
|
|
else self.original_url)
|
2024-11-30 19:54:08 +00:00
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
if redirects > 5:
|
|
|
|
return {
|
|
|
|
"success": False,
|
|
|
|
"content": f"Too many redirects (ended at {gemini_url})",
|
|
|
|
"redirected": prev_url is not None
|
|
|
|
}
|
2024-11-30 19:54:08 +00:00
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
if self.correct_urls and not prev_url:
|
|
|
|
corrected_url = self.correct_url(gemini_url)
|
|
|
|
if corrected_url != gemini_url:
|
|
|
|
print(f"[Gemini] URL '{gemini_url}' corrected to '{corrected_url}'")
|
|
|
|
gemini_url = corrected_url
|
2024-11-30 19:54:08 +00:00
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
if not prev_url:
|
|
|
|
print(f"[Gemini] Fetching: {gemini_url}")
|
|
|
|
else:
|
|
|
|
print(f"[Gemini] Fetching: {gemini_url} (redirected from {prev_url})")
|
|
|
|
|
|
|
|
try:
|
|
|
|
response = ignition.request(gemini_url, raise_errors=True, referer=prev_url)
|
|
|
|
|
|
|
|
if isinstance(response, SuccessResponse):
|
|
|
|
content = self.handle_content(response)
|
|
|
|
await self.complete_event(content)
|
|
|
|
return {
|
|
|
|
"success": True,
|
|
|
|
"content": content,
|
|
|
|
"redirected": prev_url is not None
|
|
|
|
}
|
|
|
|
elif isinstance(response, RedirectResponse):
|
|
|
|
prev_url = self.current_url
|
|
|
|
self.current_url = response.data()
|
|
|
|
return await self.fetch(prev_url, redirects + 1)
|
|
|
|
else:
|
|
|
|
await self.error_event(f"Unsupported status code {response.status}")
|
|
|
|
print(f"[Gemini] Unsupported {response.status} code for '{gemini_url}'")
|
|
|
|
message = (f"Tell the user there was a {response.status} status code. "
|
|
|
|
f"Support for handling {response.status} is not implemented yet.")
|
|
|
|
return { "success": False, "content": message, "redirected": prev_url is not None }
|
|
|
|
except Exception as e:
|
|
|
|
print(f"[Gemini] error: {e}")
|
|
|
|
message = f"Tell the user there was an error fetching the page: {e}"
|
|
|
|
await self.error_event(str(e))
|
2024-11-30 20:25:59 +00:00
|
|
|
return {
|
2024-12-13 16:07:46 +00:00
|
|
|
"success": False,
|
|
|
|
"content": message,
|
2024-11-30 20:25:59 +00:00
|
|
|
"redirected": prev_url is not None
|
|
|
|
}
|
2024-12-13 16:07:46 +00:00
|
|
|
|
|
|
|
def handle_content(self, resp: SuccessResponse) -> str:
|
|
|
|
try:
|
|
|
|
mime_type, encoding = resp.meta.split(";")
|
|
|
|
except ValueError:
|
|
|
|
mime_type = resp.meta
|
|
|
|
|
|
|
|
mime_type = mime_type.strip()
|
|
|
|
|
|
|
|
if mime_type == "text/gemini":
|
|
|
|
return resp.data().strip()
|
2024-11-30 19:54:08 +00:00
|
|
|
else:
|
2024-12-13 16:07:46 +00:00
|
|
|
raise ValueError(f"Not yet able to handle MIME type {mime_type}")
|
2024-12-01 11:12:59 +00:00
|
|
|
|
2024-11-30 11:41:13 +00:00
|
|
|
class Tools:
|
|
|
|
class Valves(BaseModel):
|
2024-12-13 22:08:01 +00:00
|
|
|
attempt_url_correction: bool = Field(
|
2024-11-30 11:41:13 +00:00
|
|
|
default=True, description="Attempt to correct malformed URLs (default enabled)."
|
|
|
|
)
|
2024-12-13 22:08:01 +00:00
|
|
|
gemini_web_proxy: str = Field(
|
|
|
|
default="", description="Web proxy to use for clickable citation links."
|
|
|
|
)
|
2024-11-30 11:41:13 +00:00
|
|
|
pass
|
|
|
|
|
|
|
|
def __init__(self):
|
|
|
|
self.valves = self.Valves()
|
|
|
|
self.user_valves = None
|
|
|
|
|
2024-12-13 16:07:46 +00:00
|
|
|
async def get_gemini_page(self, gemini_url: str, __event_emitter__) -> str:
|
2024-11-30 11:41:13 +00:00
|
|
|
"""
|
|
|
|
Fetches Gemini capsules, content, and web pages over Gemini Protocol.
|
|
|
|
Use this if the user requests a gemini:// URL.
|
|
|
|
:param gemini_url: The URL to fetch. The URL MUST begin with gemini://.
|
|
|
|
:return: The fetched data as Markdown.
|
|
|
|
"""
|
2024-12-13 16:07:46 +00:00
|
|
|
fetcher = GeminiFetcher(
|
|
|
|
gemini_url=gemini_url,
|
|
|
|
correct_urls=self.valves.attempt_url_correction,
|
2024-12-13 22:08:01 +00:00
|
|
|
web_proxy=self.valves.gemini_web_proxy,
|
2024-12-13 16:07:46 +00:00
|
|
|
event_emitter=__event_emitter__
|
|
|
|
)
|
|
|
|
|
|
|
|
resp = await fetcher.fetch()
|
2024-11-30 20:29:27 +00:00
|
|
|
if resp["success"]:
|
2024-11-30 20:25:59 +00:00
|
|
|
result_instructions = instructions(gemini_url, redirect=resp["redirected"])
|
2024-11-30 20:38:26 +00:00
|
|
|
return f"{result_instructions}\n\n```\n{resp['content']}\n```"
|
2024-11-30 20:25:59 +00:00
|
|
|
else:
|
|
|
|
return resp["content"]
|