import anthropic import streamlit as st from streamlit.components.v1 import html from streamlit_extras.stylable_container import stylable_container import re import urllib.parse import traceback import os st.title("Claude Chat UI") if "api_key" not in st.session_state and os.path.exists("api_key.dat"): with open("api_key.dat", "r", encoding="utf-8") as f: st.session_state.api_key = f.readline().strip() if "api_key" not in st.session_state: api_key = st.text_input("Enter your API Key", type="password") if not api_key: st.warning("Please enter your API key to use the app.") st.stop() if not api_key.isascii(): st.warning("Please enter your API key correctly.") st.stop() st.session_state.api_key = api_key client = anthropic.Anthropic(api_key=api_key) st.rerun() else: client = anthropic.Anthropic(api_key=st.session_state.api_key) if "messages" not in st.session_state: st.session_state.messages = [] if "prefill" not in st.session_state: st.session_state.prefill = "" if "use_continue" not in st.session_state: st.session_state.use_continue = False if "message_continue" not in st.session_state: st.session_state.message_continue = "" if "exception" not in st.session_state: st.session_state.exception = None if "use_thinking" not in st.session_state: st.session_state.use_thinking = False if "thinking_budget_tokens" not in st.session_state: st.session_state.thinking_budget_tokens = 1024 if "last_thinking" not in st.session_state: st.session_state.last_thinking = "" def count_tokens(text): if len(text) > 0: response = client.messages.count_tokens( model=model, messages = [{ "role": "user", "content": text }] ) tokens = response.input_tokens return tokens else: return 0 def get_truncated_index(messages, limit_val, limit_unit, additional_tokens): """ Find index of first message within truncated context Args: messages (list): Input messages for anthropic API limit_val (int): Context length limitation limit_unit (str): "Turns" or "Tokens" additional_tokens (int): Token count of system prompt Returns: int: index of first message within truncated context """ if limit_val == 0: return 0 if limit_unit == "Turns": # Unit: Turns count_turn = 0 for i in reversed(range(len(messages))): if messages[i]["role"] == "user": count_turn += 1 if count_turn == limit_val: return i return 0 else: # Unit: Tokens last_user_index = len(messages) total_tokens = additional_tokens for i in reversed(range(len(messages))): total_tokens += count_tokens(messages[i]["content"]) if total_tokens > limit_val: return last_user_index if messages[i]["role"] == "user": last_user_index = i return last_user_index def get_truncated_context(messages, limit_val, limit_unit, additional_tokens): """ Return truncated context Args: messages (list): Input messages for anthropic API limit_val (int): Context length limitation limit_unit (str): "Turns" or "Tokens" additional_tokens (int): Token count of system prompt Returns: list: Truncated input messages for anthropic API """ first_message_index = get_truncated_index(messages, limit_val, limit_unit, additional_tokens) if first_message_index >= len(messages): return [] return messages[first_message_index:] def get_ai_response(messages): st.session_state.is_streaming = True st.session_state.response = "" shown_message = "" st.session_state.last_thinking = "" st.session_state.message_continue = st.session_state.message_continue.strip() st.session_state.prefill = st.session_state.prefill.strip() if st.session_state.message_continue != "": messages.append({"role": "assistant", "content": st.session_state.message_continue}) st.session_state.response += st.session_state.message_continue shown_message = st.session_state.message_continue.replace("\n", " \n") elif st.session_state.prefill: messages.append({"role": "assistant", "content": st.session_state.prefill}) st.session_state.response += st.session_state.prefill shown_message = st.session_state.prefill.replace("\n", " \n") api_messages = [] for msg in messages: api_msg = {"role": msg["role"], "content": msg["content"]} api_messages.append(api_msg) api_messages = get_truncated_context(api_messages, limit_val, limit_unit, count_tokens(system_prompt)) st.session_state.exception = None try: with st.chat_message("assistant", avatar=st.session_state.assistant_avatar): placeholder = st.empty() with stylable_container( key="stop_generating", css_styles=""" button { position: fixed; bottom: 100px; left: 50%; transform: translateX(-50%); z-index: 1; } """, ): st.button("Stop generating") # Configure thinking parameter thinking_param = None if st.session_state.use_thinking: thinking_param = { "type": "enabled", "budget_tokens": st.session_state.thinking_budget_tokens } # Prepare kwargs for the API call stream_kwargs = { "messages": api_messages, "model": model, "max_tokens": max_tokens, "system": system_prompt, "temperature": temperature, } # Add thinking parameter if enabled if thinking_param: stream_kwargs["thinking"] = thinking_param elif "-4-5-" not in model: # Only add top_p and top_k when thinking is disabled and not a 4-5 model # Claude 4-5 models don't support temperature + top_p/top_k simultaneously stream_kwargs["top_p"] = top_p stream_kwargs["top_k"] = top_k with client.messages.stream(**stream_kwargs) as stream: # Stream event handling current_block_type = None for event in stream: if event.type == "content_block_start": current_block_type = event.content_block.type elif event.type == "content_block_delta": if event.delta.type == "thinking_delta" and current_block_type == "thinking": content = str(event.delta.thinking) if event.delta.thinking is not None else "" st.session_state.last_thinking += content shown_message += content.replace("\n", " \n")\ .replace("<", "\\<")\ .replace(">", "\\>") placeholder.markdown(shown_message) elif event.delta.type == "text_delta" and current_block_type == "text": content = str(event.delta.text) if event.delta.text is not None else "" st.session_state.response += content shown_message += content.replace("\n", " \n")\ .replace("<", "\\<")\ .replace(">", "\\>") placeholder.markdown(shown_message) except Exception as e: st.session_state.exception = e st.session_state.is_streaming = False return st.session_state.response def normalize_code_block(match): return match.group(0).replace(" \n", "\n")\ .replace("\\<", "<")\ .replace("\\>", ">") def normalize_inline(match): return match.group(0).replace("\\<", "<")\ .replace("\\>", ">") code_block_pattern = r"(```.*?```)" inline_pattern = r"`([^`\n]+?)`" def display_messages(): for i, message in enumerate(st.session_state.messages): if message["role"] == "user": avatar = st.session_state.user_avatar else: avatar = st.session_state.assistant_avatar with st.chat_message(message["role"], avatar=avatar): # Add expander for thinking if it exists if message["role"] == "assistant" and "thinking" in message and message["thinking"]: with st.expander("Show Claude's thinking process"): thinking_text = message["thinking"].replace("\n", " \n")\ .replace("<", "\\<")\ .replace(">", "\\>") if "```" in thinking_text: thinking_text = re.sub(code_block_pattern, normalize_code_block, thinking_text, flags=re.DOTALL) if "`" in thinking_text: thinking_text = re.sub(inline_pattern, normalize_inline, thinking_text) st.markdown(thinking_text) shown_message = message["content"].replace("\n", " \n")\ .replace("<", "\\<")\ .replace(">", "\\>") if "```" in shown_message: # Replace " \n" with "\n" within code blocks shown_message = re.sub(code_block_pattern, normalize_code_block, shown_message, flags=re.DOTALL) if "`" in shown_message: shown_message = re.sub(inline_pattern, normalize_inline, shown_message) st.markdown(shown_message) if st.session_state.get("show_message_tokens"): message_tokens = count_tokens(message["content"]) st.info(f'Tokens: {message_tokens}') col1, col2, col3, col4 = st.columns([1, 1, 1, 1]) with col1: if st.button("Edit", key=f"edit_{i}_{len(st.session_state.messages)}"): st.session_state.edit_index = i st.rerun() with col2: if st.session_state.is_delete_mode and st.button("Delete", key=f"delete_{i}_{len(st.session_state.messages)}"): del st.session_state.messages[i] st.rerun() with col3: text_to_copy = message["content"] # Encode the string to escape text_to_copy_escaped = urllib.parse.quote(text_to_copy) copy_button_html = f""" """ html(copy_button_html, height=50) if i == len(st.session_state.messages) - 1 and message["role"] == "assistant": with col4: if st.button("Retry", key=f"retry_{i}_{len(st.session_state.messages)}"): if len(st.session_state.messages) >= 2: if st.session_state.get("use_continue"): st.session_state.message_continue = st.session_state.messages[-1]["content"] del st.session_state.messages[-1] st.session_state.retry_flag = True st.rerun() if "edit_index" in st.session_state and st.session_state.edit_index == i: with st.form(key=f"edit_form_{i}_{len(st.session_state.messages)}"): new_content = st.text_area("Edit message", height=200, value=st.session_state.messages[i]["content"]) col1, col2 = st.columns([1, 1]) with col1: if st.form_submit_button("Save"): st.session_state.messages[i]["content"] = new_content del st.session_state.edit_index st.rerun() with col2: if st.form_submit_button("Cancel"): del st.session_state.edit_index st.rerun() if st.session_state.exception: st.exception(st.session_state.exception) # Add sidebar for advanced settings with st.sidebar: settings_tab, appearance_tab = st.tabs(["Settings", "Appearance"]) with settings_tab: st.markdown("Help (Japanese): https://rentry.org/9hgneofz") # Copy Conversation History button log_text = "" for message in st.session_state.messages: if message["role"] == "user": log_text += "\n" log_text += message["content"] + "\n\n" else: log_text += "\n" if "thinking" in message and message["thinking"]: log_text += "\n" log_text += message["thinking"] + "\n\n" log_text += message["content"] + "\n\n" log_text = log_text.rstrip("\n") # Encode the string to escape log_text_escaped = urllib.parse.quote(log_text) copy_log_button_html = f""" """ html(copy_log_button_html, height=50) if st.session_state.get("is_history_shown") != True: if st.button("Display History as Code Block"): st.session_state.is_history_shown = True st.rerun() else: if st.button("Hide History"): st.session_state.is_history_shown = False st.rerun() st.code(log_text) st.session_state.is_delete_mode = st.toggle("Enable Delete button") st.session_state.use_continue = st.toggle("Use Continue instead of Retry", value=st.session_state.use_continue) st.header("Advanced Settings") model_list = ["claude-sonnet-4-5-20250929", "claude-opus-4-5-20251101", "claude-haiku-4-5-20251001", "claude-opus-4-1-20250805", "claude-opus-4-20250514", "claude-sonnet-4-20250514", "claude-3-7-sonnet-20250219", "claude-3-5-haiku-20241022", "claude-3-haiku-20240307", "claude-3-opus-20240229", ] model = st.selectbox("Model", options=model_list, index=0) system_prompt = st.text_area("System prompt", height=200) st.session_state.prefill = st.text_area("Prefill", height=68, value=st.session_state.prefill, help="You can prefill the assistant's responses. You can also directly type the @prefill command into the chat field (e.g., \"Write a novel. @prefill Sure! I'd be happy to write a novel for you.\")") save_prefill = st.toggle("Save the @prefill command input in the sidebar", value=True) temperature = st.slider("Temperature", min_value=0.0, max_value=1.0, value=1.0, step=0.1) top_k = st.slider("Top-K", min_value=0, max_value=500, value=0, step=1) top_p = st.slider("Top-P", min_value=0.01, max_value=1.00, value=1.00, step=0.01) max_tokens = st.slider("Max Output Tokens", min_value=1, max_value=4096, value=4096, step=1) st.header("Extended Thinking") st.session_state.use_thinking = st.toggle( "Enable extended thinking", value=st.session_state.use_thinking, help="Enable Claude's enhanced reasoning capabilities" ) if st.session_state.use_thinking: st.session_state.thinking_budget_tokens = st.slider( "Thinking budget tokens", min_value=1024, max_value=4000, value=st.session_state.thinking_budget_tokens, step=100, help="Maximum tokens Claude can use for internal reasoning" ) # Check relationship between max_tokens and budget_tokens if st.session_state.thinking_budget_tokens >= max_tokens: st.warning("Thinking budget tokens must be less than Max Output Tokens") st.header("Context limitation") col_limit_val, col_limit_unit = st.columns([1, 1]) with col_limit_val: limit_val = st.number_input("Limit", min_value=0, max_value=200000, value="min", help="0 means no limit") with col_limit_unit: limit_unit = st.selectbox("Unit", options=["Turns","Tokens"], index=0, help="Tokens Unit is a rough estimate") st.header("Tokens") st.session_state.show_message_tokens = st.toggle("Show message tokens") system_prompt_tokens = count_tokens(system_prompt) prefill_tokens = count_tokens(st.session_state.prefill) messages_tokens = 0 for message in st.session_state.messages: messages_tokens += count_tokens(message["content"]) total_tokens = system_prompt_tokens + prefill_tokens + messages_tokens tokens_table_md = f"| Part | Tokens |\n" tokens_table_md += f"| ---- | ---- |\n" tokens_table_md += f"| System | {system_prompt_tokens} |\n" tokens_table_md += f"| Prefill | {prefill_tokens} |\n" tokens_table_md += f"| Message | {messages_tokens}({len(st.session_state.messages)}) |\n" tokens_table_md += f"| Total | {total_tokens} |\n\n" tokens_table_md += f"These token counts are very rough estimates.\n" st.markdown(tokens_table_md) st.header("Restore History") history_input = st.text_area("Paste conversation history:", height=200) if st.button("Restore History"): st.session_state.messages = [] st.session_state.exception = None messages = re.split(r"^(|)\n", history_input, flags=re.MULTILINE) role = None text = "" for message in messages: if message.strip() in ["", ""]: if role and text: if role == "assistant": thinking_match = re.search(r"\n(.*?)\n", text, re.DOTALL) if thinking_match: thinking_content = thinking_match.group(1).strip() content = re.sub(r"\n.*?\n", "", text, flags=re.DOTALL).strip() st.session_state.messages.append({ "role": role, "content": content, "thinking": thinking_content }) else: st.session_state.messages.append({"role": role, "content": text.strip()}) else: st.session_state.messages.append({"role": role, "content": text.strip()}) text = "" role = "user" if message.strip() == "" else "assistant" else: text += message if role and text: if role == "assistant": thinking_match = re.search(r"\n(.*?)\n", text, re.DOTALL) if thinking_match: thinking_content = thinking_match.group(1).strip() content = re.sub(r"\n.*?\n", "", text, flags=re.DOTALL).strip() st.session_state.messages.append({ "role": role, "content": content, "thinking": thinking_content }) else: st.session_state.messages.append({"role": role, "content": text.strip()}) else: st.session_state.messages.append({"role": role, "content": text.strip()}) st.rerun() st.header("Clear History") if st.button("Clear Chat History"): st.session_state.messages = [] st.session_state.exception = None st.rerun() st.header("Change API Key") new_api_key = st.text_input("Enter new API Key", type="password") if st.button("Update API Key"): if new_api_key: st.session_state.api_key = new_api_key client = anthropic.Anthropic(api_key=new_api_key) st.success("API Key updated successfully!") else: st.warning("Please enter a valid API Key.") with appearance_tab: st.header("Font Selection") font_options = { "Zen Maru Gothic": "Zen Maru Gothic", "Noto Sans JP": "Noto Sans JP", "Sawarabi Mincho": "Sawarabi Mincho" } selected_font = st.selectbox("Choose a font", ["Default"] + list(font_options.keys())) st.header("Change the font size") st.session_state.font_size = st.slider("Font size", min_value=16.0, max_value=50.0, value=16.0, step=1.0) st.header("Change the user's icon") st.session_state.user_avatar = st.file_uploader("Choose an image", type=["png", "jpg", "jpeg", "webp", "gif", "bmp", "svg",], key="user_avatar_uploader") st.header("Change the assistant's icon") st.session_state.assistant_avatar = st.file_uploader("Choose an image", type=["png", "jpg", "jpeg", "webp", "gif", "bmp", "svg",], key="assistant_avatar_uploader") st.header("Change the icon size") st.session_state.avatar_size = st.slider("Icon size", min_value=2.0, max_value=20.0, value=2.0, step=0.2) # After Stop generating if st.session_state.get("is_streaming"): message_content = {"role": "assistant", "content": st.session_state.response} if st.session_state.last_thinking: message_content["thinking"] = st.session_state.last_thinking st.session_state.messages.append(message_content) st.session_state.is_streaming = False if "retry_flag" in st.session_state and st.session_state.retry_flag: st.session_state.retry_flag = False st.session_state.message_continue = "" st.rerun() # Change the font if selected_font != "Default": with open("style.css") as css: st.markdown(f'', unsafe_allow_html=True) st.markdown(f'', unsafe_allow_html=True) # Change font size st.markdown(f'', unsafe_allow_html=True) # Change icon size # (CSS element names may be subject to change.) # (Contributor: ★31 >>538) AVATAR_SIZE_STYLE = f""" """ st.markdown(AVATAR_SIZE_STYLE, unsafe_allow_html=True) display_messages() # After Retry if st.session_state.get("retry_flag"): if len(st.session_state.messages) > 0: messages = st.session_state.messages.copy() response = get_ai_response(messages) message_content = {"role": "assistant", "content": response} if st.session_state.last_thinking: message_content["thinking"] = st.session_state.last_thinking st.session_state.messages.append(message_content) st.session_state.retry_flag = False st.session_state.message_continue = "" st.rerun() else: st.session_state.retry_flag = False st.session_state.message_continue = "" if prompt := st.chat_input("Enter your message here..."): used_prefill = False prefill_pattern = r"([@@](prefill|ぷれふぃる|プレフィル)\s?(.*))" prefill_match = re.search(prefill_pattern, prompt) if prefill_match: used_prefill = True if not save_prefill: original_prefill = st.session_state.prefill st.session_state.prefill = prefill_match.group(3) prompt = prompt.replace(prefill_match.group(1), '') st.session_state.messages.append({"role": "user", "content": prompt}) messages = st.session_state.messages.copy() shown_message = prompt.replace("\n", " \n")\ .replace("<", "\\<")\ .replace(">", "\\>") with st.chat_message("user", avatar=st.session_state.user_avatar): st.write(shown_message) response = get_ai_response(messages) message_content = {"role": "assistant", "content": response} if st.session_state.last_thinking: message_content["thinking"] = st.session_state.last_thinking st.session_state.messages.append(message_content) if used_prefill and not save_prefill: st.session_state.prefill = original_prefill st.rerun()