"""
Raven AI — Main Application
Full-featured emotionally-aware AI assistant with streaming, multi-chat,
document processing, image understanding, voice input, data analysis, and web search.
"""
import json
import os
import uuid
import time
import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use("Agg")
from emotion_engine import detect_emotion, detect_intent, detect_crisis
from responder import (
get_response, get_response_stream, generate_title,
analyze_image, transcribe_audio, process_document, analyze_data,
search_and_respond,
)
# ── Page config ──────────────────────────────────────────────────────────────
st.set_page_config(
page_title="Raven AI",
page_icon="🐦⬛",
layout="wide",
initial_sidebar_state="expanded",
)
# ── Conversation persistence file ────────────────────────────────────────────
CONV_FILE = os.path.join(os.path.dirname(__file__), ".raven_conversations.json")
def save_conversations():
"""Save all conversations to disk for persistence."""
try:
data = {}
for cid, conv in st.session_state.conversations.items():
data[cid] = {
"title": conv.get("title", "New Chat"),
"messages": conv.get("messages", []),
"emotion_log": conv.get("emotion_log", []),
"current_emotion": conv.get("current_emotion", "neutral"),
"current_intent": conv.get("current_intent", "casual"),
"created_at": conv.get("created_at", time.time()),
}
with open(CONV_FILE, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=2)
except Exception:
pass
def load_conversations():
"""Load conversations from disk."""
try:
if os.path.exists(CONV_FILE):
with open(CONV_FILE, "r", encoding="utf-8") as f:
return json.load(f)
except Exception:
pass
return {}
# ── Custom CSS ───────────────────────────────────────────────────────────────
st.markdown("""
""", unsafe_allow_html=True)
# ── Style maps ───────────────────────────────────────────────────────────────
EMOTION_STYLE = {
"happy": {"emoji": "😊", "color": "#F59E0B"},
"sad": {"emoji": "😔", "color": "#60A5FA"},
"anxious": {"emoji": "😰", "color": "#A78BFA"},
"angry": {"emoji": "😤", "color": "#F87171"},
"confused": {"emoji": "😕", "color": "#34D399"},
"neutral": {"emoji": "😐", "color": "#9CA3AF"},
}
INTENT_STYLE = {
"venting": {"emoji": "💬", "label": "Venting"},
"question": {"emoji": "🔍", "label": "Question"},
"advice": {"emoji": "💡", "label": "Advice"},
"casual": {"emoji": "😌", "label": "Casual"},
"crisis": {"emoji": "🆘", "label": "Crisis"},
"coding": {"emoji": "💻", "label": "Coding"},
"learning": {"emoji": "📚", "label": "Learning"},
"creative": {"emoji": "🎨", "label": "Creative"},
"deep_dive": {"emoji": "🔬", "label": "Deep Dive"},
}
EMOTION_ORDER = ["happy", "neutral", "confused", "anxious", "sad", "angry"]
EMOTION_COLORS = {
"happy": "#F59E0B", "sad": "#60A5FA", "anxious": "#A78BFA",
"angry": "#F87171", "confused": "#34D399", "neutral": "#9CA3AF",
}
WELCOME_PROMPTS = [
{"emoji": "💻", "title": "Write Code", "prompt": "Write a Python script that scrapes the top 10 trending GitHub repos"},
{"emoji": "📚", "title": "Learn Something", "prompt": "Explain how neural networks work — from scratch, step by step"},
{"emoji": "🎨", "title": "Creative Writing", "prompt": "Write a short sci-fi story about an AI that develops emotions"},
{"emoji": "🧩", "title": "Solve a Problem", "prompt": "I need to build a portfolio website. Help me plan the architecture and tech stack"},
{"emoji": "📊", "title": "Analyze Data", "prompt": "Explain the key concepts of statistics I need for data science"},
{"emoji": "🔬", "title": "Deep Dive", "prompt": "Give me an exhaustive breakdown of how transformers work in AI"},
]
# ═══════════════════════════════════════════════════════════════════════════════
# SESSION STATE
# ═══════════════════════════════════════════════════════════════════════════════
def create_conversation():
"""Create a new conversation and return its ID."""
conv_id = str(uuid.uuid4())
st.session_state.conversations[conv_id] = {
"title": "New Chat",
"messages": [],
"emotion_log": [],
"current_emotion": "neutral",
"current_intent": "casual",
"created_at": time.time(),
}
return conv_id
def get_conv():
"""Get the active conversation dict."""
return st.session_state.conversations.get(st.session_state.active_conv_id, {})
def delete_conversation(conv_id):
"""Delete a conversation."""
if conv_id in st.session_state.conversations:
del st.session_state.conversations[conv_id]
# If we deleted the active one, switch to another or create new
if conv_id == st.session_state.active_conv_id:
remaining = list(st.session_state.conversations.keys())
if remaining:
st.session_state.active_conv_id = remaining[0]
else:
new_id = create_conversation()
st.session_state.active_conv_id = new_id
save_conversations()
def init_session_state():
"""Initialize all session state variables."""
if "conversations" not in st.session_state:
loaded = load_conversations()
if loaded:
st.session_state.conversations = loaded
else:
st.session_state.conversations = {}
if "active_conv_id" not in st.session_state:
# Always start with a fresh new chat — old chats stay in sidebar
conv_id = create_conversation()
st.session_state.active_conv_id = conv_id
# Ensure active conv exists
if st.session_state.active_conv_id not in st.session_state.conversations:
conv_id = create_conversation()
st.session_state.active_conv_id = conv_id
if "pending_prompt" not in st.session_state:
st.session_state.pending_prompt = None
if "regenerate_idx" not in st.session_state:
st.session_state.regenerate_idx = None
if "uploaded_context" not in st.session_state:
st.session_state.uploaded_context = None
if "preferences" not in st.session_state:
st.session_state.preferences = {"tone": "balanced", "name": ""}
if "voice_transcript" not in st.session_state:
st.session_state.voice_transcript = None
if "edit_msg_idx" not in st.session_state:
st.session_state.edit_msg_idx = None
if "upload_key" not in st.session_state:
st.session_state.upload_key = 0
init_session_state()
# ═══════════════════════════════════════════════════════════════════════════════
# HELPER FUNCTIONS
# ═══════════════════════════════════════════════════════════════════════════════
def build_emotion_chart(emotion_log):
"""Build the emotion trend chart."""
if len(emotion_log) < 2:
return None
emotions = [e for _, e in emotion_log]
y_vals = [EMOTION_ORDER.index(e) if e in EMOTION_ORDER else 2 for e in emotions]
x_vals = list(range(1, len(emotions) + 1))
colors = [EMOTION_COLORS.get(e, "#9CA3AF") for e in emotions]
fig, ax = plt.subplots(figsize=(3.5, 2.2))
fig.patch.set_facecolor("#0e1117")
ax.set_facecolor("#0e1117")
ax.plot(x_vals, y_vals, color="#6366F1", linewidth=1.5, alpha=0.6, zorder=1)
for x, y, c in zip(x_vals, y_vals, colors):
ax.scatter(x, y, color=c, s=50, zorder=2)
ax.set_yticks(range(len(EMOTION_ORDER)))
ax.set_yticklabels(EMOTION_ORDER, fontsize=7, color="#9CA3AF")
ax.set_xticks(x_vals[-10:])
ax.set_xticklabels([str(i) for i in x_vals[-10:]], fontsize=7, color="#9CA3AF")
ax.set_xlabel("Message", fontsize=7, color="#9CA3AF")
ax.tick_params(colors="#9CA3AF")
for spine in ax.spines.values():
spine.set_edgecolor("#2d2d3a")
ax.grid(axis="y", color="#2d2d3a", linewidth=0.5, linestyle="--")
plt.tight_layout(pad=0.5)
return fig
def get_recent_context():
"""Get recent user messages for context."""
conv = get_conv()
user_msgs = [m["content"] for m in conv.get("messages", []) if m.get("role") == "user"][-2:]
return " | ".join(user_msgs) if user_msgs else ""
def render_badges(emotion, intent):
"""Render emotion + intent badges."""
s = EMOTION_STYLE.get(emotion, EMOTION_STYLE["neutral"])
i = INTENT_STYLE.get(intent, INTENT_STYLE["casual"])
st.markdown(
f""
f"{s['emoji']} {emotion}"
f""
f"{i['emoji']} {i['label']}",
unsafe_allow_html=True,
)
def render_word_count(text):
"""Show word count below a response."""
words = len(text.split())
st.markdown(f"{words} words
", unsafe_allow_html=True)
def process_uploaded_file(uploaded_file):
"""Process an uploaded file and return (content_text, file_type)."""
name = uploaded_file.name.lower()
raw = uploaded_file.read()
if name.endswith((".png", ".jpg", ".jpeg", ".gif", ".webp")):
return raw, "image"
if name.endswith((".wav", ".mp3", ".m4a", ".ogg", ".flac", ".webm")):
return raw, "audio"
if name.endswith((".csv", ".tsv")):
try:
sep = "\t" if name.endswith(".tsv") else ","
uploaded_file.seek(0)
df = pd.read_csv(uploaded_file, sep=sep)
summary = f"Shape: {df.shape}\n\nColumns: {list(df.columns)}\n\nFirst 10 rows:\n{df.head(10).to_string()}\n\nDescribe:\n{df.describe().to_string()}"
return summary, "data"
except Exception:
return raw.decode("utf-8", errors="ignore"), "text"
if name.endswith(".pdf"):
try:
import PyPDF2
uploaded_file.seek(0)
reader = PyPDF2.PdfReader(uploaded_file)
text = ""
for page in reader.pages[:50]:
text += page.extract_text() or ""
return text, "document"
except Exception:
return "Could not read PDF.", "error"
try:
text = raw.decode("utf-8", errors="ignore")
return text, "document"
except Exception:
return "Could not read file.", "error"
# ═══════════════════════════════════════════════════════════════════════════════
# SIDEBAR — Clean layout: New Chat, Conversations, Mood, Preferences, Graph
# ═══════════════════════════════════════════════════════════════════════════════
with st.sidebar:
# ── i) Header + New Chat ─────────────────────────────────────────────────
st.markdown(
"🐦⬛ Raven AI
"
""
"Thought. Memory. Empathy.
",
unsafe_allow_html=True,
)
if st.button("➕ New Chat", use_container_width=True, type="primary"):
conv_id = create_conversation()
st.session_state.active_conv_id = conv_id
st.session_state.uploaded_context = None
st.session_state.upload_key += 1
save_conversations()
st.rerun()
st.divider()
# ── ii) Past Conversations (with 3-dot menu each) ────────────────────────
convs = sorted(
st.session_state.conversations.items(),
key=lambda x: x[1].get("created_at", 0),
reverse=True,
)
# Filter: show chats that have messages, plus the active one
visible_convs = [
(cid, c) for cid, c in convs
if len([m for m in c.get("messages", []) if m.get("role") == "user"]) > 0
or cid == st.session_state.active_conv_id
]
if visible_convs:
st.markdown(
"Conversations
",
unsafe_allow_html=True,
)
for conv_id, conv in visible_convs[:30]:
is_active = conv_id == st.session_state.active_conv_id
title = conv.get("title", "New Chat")
msg_count = len([m for m in conv.get("messages", []) if m.get("role") == "user"])
col_title, col_menu = st.columns([5, 1])
with col_title:
display = f"**{title}**" if is_active else title
if st.button(
display,
key=f"conv_{conv_id}",
use_container_width=True,
disabled=is_active,
):
st.session_state.active_conv_id = conv_id
st.session_state.uploaded_context = None
st.session_state.upload_key += 1
st.rerun()
with col_menu:
with st.popover("⋮", use_container_width=True):
if st.button("🗑️ Delete", key=f"del_{conv_id}", use_container_width=True):
delete_conversation(conv_id)
st.rerun()
if msg_count > 0:
if st.button("📥 Export", key=f"exp_{conv_id}", use_container_width=True):
export_data = [
{"role": m.get("role", ""), "content": m.get("content", ""),
"emotion": m.get("emotion", ""), "intent": m.get("intent", "")}
for m in conv.get("messages", [])
]
csv_data = pd.DataFrame(export_data).to_csv(index=False)
st.download_button(
"Download CSV", csv_data,
f"raven_{title[:20]}.csv", "text/csv",
key=f"dl_{conv_id}",
)
if st.button("📋 Summarize", key=f"sum_{conv_id}", use_container_width=True):
emotion_trend = [e for _, e in conv.get("emotion_log", [])]
summary_prompt = (
f"Summarise this conversation in 3-4 sentences. "
f"Detected emotions: {', '.join(emotion_trend) if emotion_trend else 'none'}. "
f"Base your summary strictly on these emotions."
)
clean_msgs = [
{"role": m["role"], "content": m["content"]}
for m in conv.get("messages", [])
if m.get("role") in ["user", "assistant"] and m.get("content")
]
summary = get_response(
summary_prompt,
conv.get("current_emotion", "neutral"),
clean_msgs,
)
st.info(summary)
st.divider()
# ── iii) Current Mood & Intent ───────────────────────────────────────────
conv = get_conv()
emotion = conv.get("current_emotion", "neutral")
style = EMOTION_STYLE.get(emotion, EMOTION_STYLE["neutral"])
color, emoji = style["color"], style["emoji"]
st.markdown(
"Mood & Intent
",
unsafe_allow_html=True,
)
st.markdown(
f""
f"{emoji} {emotion.capitalize()}
",
unsafe_allow_html=True,
)
intent = conv.get("current_intent", "casual")
intent_info = INTENT_STYLE.get(intent, INTENT_STYLE["casual"])
st.markdown(
f""
f"{intent_info['emoji']} {intent_info['label']}
",
unsafe_allow_html=True,
)
# ── iv) Preferences ──────────────────────────────────────────────────────
st.divider()
st.markdown(
"Preferences
",
unsafe_allow_html=True,
)
prefs = st.session_state.preferences
new_name = st.text_input(
"Your name",
value=prefs.get("name", ""),
placeholder="So Raven can address you",
key="pref_name",
label_visibility="collapsed",
)
prefs["name"] = new_name
new_tone = st.selectbox(
"Tone",
["Balanced", "Formal", "Casual", "Concise", "Detailed"],
index=["balanced", "formal", "casual", "concise", "detailed"].index(
prefs.get("tone", "balanced")
) if prefs.get("tone", "balanced") in ["balanced", "formal", "casual", "concise", "detailed"] else 0,
key="pref_tone",
label_visibility="collapsed",
)
prefs["tone"] = new_tone.lower()
# ── v) Emotion Trend Graph ───────────────────────────────────────────────
emotion_log = conv.get("emotion_log", [])
if len(emotion_log) >= 2:
st.divider()
st.markdown(
"Emotion Trend
",
unsafe_allow_html=True,
)
fig = build_emotion_chart(emotion_log)
if fig:
st.pyplot(fig, use_container_width=True)
plt.close(fig)
# ═══════════════════════════════════════════════════════════════════════════════
# MAIN CHAT AREA
# ═══════════════════════════════════════════════════════════════════════════════
conv = get_conv()
messages = conv.get("messages", [])
# ── Welcome screen (when no messages) ────────────────────────────────────────
if not messages:
st.markdown("")
st.markdown(
""
"
🐦⬛
"
"
Hello, I'm Raven
"
"
"
"Your emotionally aware AI assistant — I can code, explain, create, analyze, and more.
"
"
"
"
"
"📄 Upload docs • "
"📷 Analyze images • "
"🎙️ Voice input • "
"📊 Data analysis • "
"🌐 Web search
"
"
"
f"{type_emoji} {ctx.get('name', 'File')} attached
",
unsafe_allow_html=True,
)
with clear_col:
if st.button("✕", key="clear_upload", help="Remove attachment"):
st.session_state.uploaded_context = None
st.session_state.upload_key += 1
st.rerun()
# File upload popover (appears above chat, left side)
attach_col, spacer_col, voice_col = st.columns([1, 8, 1])
with attach_col:
with st.popover("📎", use_container_width=True, help="Attach file"):
uploaded_files = st.file_uploader(
"Upload files",
type=["pdf", "txt", "py", "js", "java", "c", "cpp", "csv", "tsv",
"json", "md", "html", "css", "png", "jpg", "jpeg", "gif",
"webp", "wav", "mp3", "m4a", "ogg", "flac", "webm",
"xlsx", "docx"],
label_visibility="collapsed",
accept_multiple_files=True,
key=f"file_upload_{st.session_state.upload_key}",
)
if uploaded_files:
# Check if all files are images
image_files = [f for f in uploaded_files if f.name.lower().endswith((".png", ".jpg", ".jpeg", ".gif", ".webp"))]
if len(image_files) == len(uploaded_files) and len(image_files) > 1:
# Multiple images — batch them together (max 5)
images_bytes = []
names = []
for img_file in image_files[:5]:
images_bytes.append(img_file.read())
names.append(img_file.name)
for img_file in image_files[:5]:
img_file.seek(0)
st.image(img_file, caption=img_file.name, width=100)
st.session_state.uploaded_context = {"type": "image", "content": images_bytes, "names": names}
st.success(f"📷 {len(images_bytes)} images ready — ask about them!")
if len(image_files) > 5:
st.warning("Max 5 images per request — first 5 selected.")
else:
# Single file or mixed — process the first file
uploaded_file = uploaded_files[0]
content, file_type = process_uploaded_file(uploaded_file)
if file_type == "image":
st.image(uploaded_file, caption=uploaded_file.name, use_container_width=True)
st.session_state.uploaded_context = {"type": "image", "content": content, "name": uploaded_file.name}
st.success("📷 Image ready — ask about it!")
elif file_type == "audio":
st.audio(uploaded_file)
with st.spinner("Transcribing..."):
transcript = transcribe_audio(content, uploaded_file.name)
st.session_state.uploaded_context = {"type": "audio", "content": transcript, "name": uploaded_file.name}
st.success("🎙️ Transcribed!")
elif file_type == "data":
st.session_state.uploaded_context = {"type": "data", "content": content, "name": uploaded_file.name}
st.success("📊 Data loaded!")
elif file_type == "document":
st.session_state.uploaded_context = {"type": "document", "content": content, "name": uploaded_file.name}
st.success("📄 Document ready!")
doc_action = st.selectbox(
"Quick action:",
["Chat about it", "Summarize", "Key Points", "Generate Q&A",
"Flashcards", "Simplify", "Study Notes", "Topics"],
key="doc_action_chat",
)
action_map = {
"Summarize": "summarize", "Key Points": "key_points",
"Generate Q&A": "questions", "Flashcards": "flashcards",
"Simplify": "simplify", "Study Notes": "notes", "Topics": "topics",
}
if doc_action != "Chat about it" and st.button("🚀 Process"):
with st.spinner(f"Processing: {doc_action}..."):
result = process_document(content, uploaded_file.name, action_map[doc_action])
st.markdown(result)
elif file_type == "error":
st.error(content)
with voice_col:
with st.popover("🎙️", use_container_width=True, help="Voice input"):
st.markdown("**🎙️ Record your message:**")
audio_input = st.audio_input("Record", key="voice_chat", label_visibility="collapsed")
if audio_input:
# Only transcribe if we haven't already processed this audio
audio_bytes = audio_input.read()
audio_hash = hash(audio_bytes[:100]) # quick fingerprint
if st.session_state.get("last_audio_hash") != audio_hash:
st.session_state.last_audio_hash = audio_hash
with st.spinner("Transcribing..."):
transcript = transcribe_audio(audio_bytes, "recording.wav")
if transcript and not transcript.startswith("Couldn't"):
st.session_state.voice_transcript = transcript
else:
st.error(transcript)
# Show editable transcript if available
if st.session_state.get("voice_transcript"):
st.markdown("---")
st.markdown("**✏️ Edit before sending:**")
edited = st.text_area(
"Edit transcript",
value=st.session_state.voice_transcript,
key="voice_edit",
height=150,
label_visibility="collapsed",
)
if st.button("📨 Send Message", key="voice_send", type="primary", use_container_width=True):
st.session_state.pending_prompt = edited
st.session_state.voice_transcript = None
st.rerun()
# Process pending voice/welcome prompt BEFORE chat_input renders
voice_pending = st.session_state.pending_prompt
# The actual chat input
user_input = st.chat_input("Talk to Raven — ask anything...")
# Voice/welcome prompt takes priority if no typed input
if not user_input and voice_pending:
user_input = voice_pending
st.session_state.pending_prompt = None
elif user_input:
# User typed something — clear any pending prompt
st.session_state.pending_prompt = None
if user_input:
user_input = user_input.strip()
if not user_input:
st.stop()
if len(user_input) > 5000:
st.warning("Message too long — keep under 5000 characters.")
st.stop()
conv = get_conv()
messages = conv.get("messages", [])
recent_context = get_recent_context()
# ── Build extra context ──────────────────────────────────────────────────
extra_context = ""
prefs = st.session_state.preferences
pref_parts = []
if prefs.get("name"):
pref_parts.append(f"The user's name is {prefs['name']}. Address them by name occasionally.")
if prefs.get("tone") and prefs["tone"] != "balanced":
pref_parts.append(f"The user prefers a {prefs['tone']} response style.")
if pref_parts:
extra_context = "User preferences: " + " ".join(pref_parts)
upload_ctx = st.session_state.uploaded_context
if upload_ctx:
if upload_ctx["type"] == "image":
# Image(s) — use vision model directly
with st.chat_message("user"):
st.markdown(user_input)
# Show thumbnail previews of attached images
if isinstance(upload_ctx["content"], list):
img_cols = st.columns(min(len(upload_ctx["content"]), 5))
for i, img_b in enumerate(upload_ctx["content"][:5]):
with img_cols[i]:
st.image(img_b, width=80)
conv["messages"].append({"role": "user", "content": user_input, "emotion": "neutral", "intent": "question"})
with st.chat_message("assistant"):
img_count = len(upload_ctx["content"]) if isinstance(upload_ctx["content"], list) else 1
with st.status("🐦⬛ Raven is thinking...", expanded=True) as status:
st.write(f"🎨 Analyzing {img_count} image{'s' if img_count > 1 else ''}...")
response = analyze_image(upload_ctx["content"], user_input)
status.update(label="Done!", state="complete")
st.markdown(response)
render_word_count(response)
conv["messages"].append({"role": "assistant", "content": response})
st.session_state.uploaded_context = None
st.session_state.upload_key += 1
if conv.get("title") == "New Chat":
conv["title"] = generate_title(user_input)
save_conversations()
st.rerun()
elif upload_ctx["type"] == "audio":
extra_context += f"\n\n[Transcribed audio from {upload_ctx['name']}]:\n{upload_ctx['content'][:8000]}"
elif upload_ctx["type"] == "data":
extra_context += f"\n\n[Data from {upload_ctx['name']}]:\n{upload_ctx['content'][:8000]}"
elif upload_ctx["type"] == "document":
extra_context += f"\n\n[Document: {upload_ctx['name']}]:\n{upload_ctx['content'][:8000]}"
# ── Web search ───────────────────────────────────────────────────────────
searched_web = False
try:
from web_search import should_search, search_web
preliminary_intent = detect_intent(user_input, "neutral", recent_context)
if should_search(user_input, preliminary_intent):
web_context, _ = search_web(user_input)
if web_context:
extra_context += f"\n\n{web_context}"
searched_web = True
except ImportError:
pass
# ── Emotion & intent ─────────────────────────────────────────────────────
primary_emotion, _ = detect_emotion(user_input, recent_context, conv.get("emotion_log", []))
intent = detect_intent(user_input, primary_emotion, recent_context)
# Crisis continuation: only if the previous user message had actual crisis keywords
direct_crisis = detect_crisis(user_input)
prev_user_msgs = [m for m in messages if m.get("role") == "user"]
recent_crisis = (
len(prev_user_msgs) > 0
and prev_user_msgs[-1].get("intent") == "crisis"
and detect_crisis(prev_user_msgs[-1].get("content", ""))
)
if direct_crisis or recent_crisis:
primary_emotion = "sad"
intent = "crisis"
conv["current_emotion"] = primary_emotion
conv["current_intent"] = intent
conv["messages"].append({
"role": "user", "content": user_input,
"emotion": primary_emotion, "intent": intent,
})
conv["emotion_log"].append((user_input, primary_emotion))
# ── Display user message ─────────────────────────────────────────────────
with st.chat_message("user"):
render_badges(primary_emotion, intent)
st.markdown(user_input)
# ── Stream response ──────────────────────────────────────────────────────
with st.chat_message("assistant"):
if intent == "crisis" and direct_crisis:
# First crisis detection (keywords found) — show helplines directly
response = ("I can hear that you're going through something really difficult right now.\n"
"You don't have to face this alone. Please reach out to someone who can help:\n\n"
"🆘 **iCall (India):** 9152987821\n"
"🆘 **Vandrevala Foundation:** 1860-2662-345 (24/7)\n"
"🆘 **AASRA:** 9820466726\n"
"🆘 **International:** https://www.iasp.info/resources/Crisis_Centres/\n\n"
"You matter. I'm here to talk if you need me. 💙")
st.markdown(response)
elif intent == "crisis" and recent_crisis:
# Crisis continuation — send to Groq with gentle crisis-aware prompt
extra_context += ("\n\n[IMPORTANT CONTEXT: The user recently expressed crisis-level distress. "
"Helpline numbers have ALREADY been provided. Do NOT repeat helpline numbers. "
"Instead, be deeply empathetic, warm, and supportive. Acknowledge their feelings, "
"gently encourage them, and remind them they are not alone. Do NOT say 'I cannot assist with self-harm' "
"or any cold safety refusal. Be a caring friend.]")
with st.status("🐦⬛ Raven is thinking...", expanded=True) as status:
st.write(f"🧠 Emotion: **{primary_emotion}**")
st.write(f"🎯 Intent: **{intent}**")
status.update(label="Streaming response...", state="running")
try:
clean_msgs = [
{"role": m["role"], "content": m["content"]}
for m in conv["messages"][:-1]
if m.get("role") in ["user", "assistant"] and m.get("content")
]
stream = get_response_stream(user_input, primary_emotion, clean_msgs, extra_context)
response = st.write_stream(stream)
except Exception:
response = ("I hear you, and I want you to know that your feelings are valid. "
"You don't have to go through this alone. I'm right here with you. 💙")
st.markdown(response)
else:
with st.status("🐦⬛ Raven is thinking...", expanded=True) as status:
st.write(f"🧠 Emotion: **{primary_emotion}**")
st.write(f"🎯 Intent: **{intent}**")
if searched_web:
st.write("🌐 Searched the web...")
if upload_ctx:
st.write(f"📎 Using: {upload_ctx.get('name', 'file')}")
status.update(label="Streaming response...", state="running")
try:
clean_msgs = [
{"role": m["role"], "content": m["content"]}
for m in conv["messages"][:-1]
if m.get("role") in ["user", "assistant"] and m.get("content")
]
stream = get_response_stream(user_input, primary_emotion, clean_msgs, extra_context)
response = st.write_stream(stream)
except Exception:
response = "I ran into a small issue. Could you try again?"
st.markdown(response)
render_word_count(response)
conv["messages"].append({"role": "assistant", "content": response})
# Clear attachment after it's been used in a message
if st.session_state.uploaded_context:
st.session_state.uploaded_context = None
st.session_state.upload_key += 1
# Auto-title
if conv.get("title") == "New Chat" and len([m for m in conv["messages"] if m["role"] == "user"]) == 1:
conv["title"] = generate_title(user_input)
save_conversations()
st.rerun()