Better queries with indexed data

r2ai/index.py

@@ -10,6 +10,7 @@ try:
 except:
 	from utils import slurp
 
+MAXCHARS = 128
 MAXMATCHES = 5
 MASTODON_KEY = ""
 try:
@@ -96,12 +97,16 @@ def md2txt(text):
 
 def filter_line(line):
 	line = unidecode(line) # remove accents
+	line = re.sub(r'https?://\S+', '', line)
+	line = re.sub(r'http?://\S+', '', line)
 	line = line.replace(":", " ").replace("/", " ").replace("`", " ").replace("?", " ")
 	line = line.replace("\"", " ").replace("'", " ")
 	line = line.replace("<", " ").replace(">", " ").replace("@", " ").replace("#", " ")
-	nline = line.replace("-", " ").replace(".", " ").replace(",", " ").replace("(", " ").replace(")", " ").strip(" ")
+	line = line.replace("-", " ").replace(".", " ").replace(",", " ").replace("(", " ").replace(")", " ").strip(" ")
+	if len(line) > MAXCHARS:
+		line = line[:MAXCHARS]
 	words = []
-	for a in nline.split(" "):
+	for a in line.split(" "):
 		b = a.strip().lower()
 		try:
 			int(b)

r2ai/interpreter.py

@@ -91,7 +91,7 @@ def messages_to_prompt(self, messages):
   else:
     formatted_messages = template_llama(self, messages)
 
-  if "DEBUG" in self.env:
+  if self.env["debug"] == "true":
     builtins.print(formatted_messages)
   return formatted_messages
 
@@ -534,8 +534,7 @@ class Interpreter:
         max_tokens=(self.context_window-self.max_tokens-25),
         system_message=system_message)
 
-    # DEBUG DEBUG DEBUG DEBUG DEBUG DEBUG
-    if "DEBUG" in self.env:
+    if self.env["debug"] == "true":
       print(messages)
 
     # Code-Llama

r2ai/voice.py

@@ -1,9 +1,10 @@
 import subprocess
 import os
+import re
 
 have_whisper = False
 model = None
-voice_model = "base" # base
+voice_model = "large" # base
 LANGUAGE = "ca"
 DEVICE = None
 try:
@@ -42,7 +43,7 @@ def stt(seconds):
 	if device is None:
 		tts("(r2ai)", "cannot find a microphone")
 		return
-	tts("(r2ai) listening for 5s... ", "si?")
+	tts("(r2ai) listening for 5s... ", "digues?")
 	print(f"DEVICE IS {device}")
 	os.system("rm -f .audiomsg.wav")
 	os.system(f"ffmpeg -f avfoundation -t 5 -i '{device}' .audiomsg.wav > /dev/null 2>&1")
@@ -52,7 +53,7 @@ def stt(seconds):
 	else:
 		result = model.transcribe(".audiomsg.wav", language=LANGUAGE)
 	os.system("rm -f .audiomsg.wav")
-	tts("(r2ai)", "ah")
+	tts("(r2ai)", "ok")
 	text = result["text"].strip()
 	if text == "you":
 		return ""
@@ -60,5 +61,7 @@ def stt(seconds):
 	return text
 
 def tts(author, text):
+	clean_text = re.sub(r'https?://\S+', '', text)
+	clean_text = re.sub(r'http?://\S+', '', clean_text)
 	print(f"{author}: {text}")
-	subprocess.run(["say", text])
+	subprocess.run(["say", clean_text])