tools/orchestrator.py

import os
import json
import re
import logging
import time

from model_logic import call_model_stream, MODELS_BY_PROVIDER, get_default_model_display_name_for_provider
from memory_logic import retrieve_memories_semantic, retrieve_rules_semantic
from tools.websearch import search_and_scrape_duckduckgo, scrape_url
from tools.space_builder import create_huggingface_space, update_huggingface_space_file
import prompts
from utils import format_insights_for_prompt

logger = logging.getLogger(__name__)

WEB_SEARCH_ENABLED = os.getenv("WEB_SEARCH_ENABLED", "true").lower() == "true"
TOOL_DECISION_PROVIDER = os.getenv("TOOL_DECISION_PROVIDER", "groq")
TOOL_DECISION_MODEL_ID = os.getenv("TOOL_DECISION_MODEL", "llama3-8b-8192")
MAX_HISTORY_TURNS = int(os.getenv("MAX_HISTORY_TURNS", 7))

def decide_on_tool(user_input: str, chat_history_for_prompt: list, initial_insights_ctx_str: str):
    user_input_lower = user_input.lower()

    if "http://" in user_input or "https://" in user_input:
        url_match = re.search(r'(https?://[^\s]+)', user_input)
        if url_match:
            return "scrape_url_and_report", {"url": url_match.group(1)}

    tool_trigger_keywords = ["what is", "how to", "explain", "search for", "build", "create", "make", "update", "modify", "change", "fix"]
    if len(user_input.split()) > 3 or "?" in user_input or any(w in user_input_lower for w in tool_trigger_keywords):
        history_snippet = "\n".join([f"{msg['role']}: {msg['content'][:100]}" for msg in chat_history_for_prompt[-2:]])
        guideline_snippet = initial_insights_ctx_str[:200].replace('\n', ' ')
        tool_user_prompt = prompts.get_tool_user_prompt(user_input, history_snippet, guideline_snippet)
        tool_decision_messages = [{"role": "system", "content": prompts.TOOL_SYSTEM_PROMPT}, {"role": "user", "content": tool_user_prompt}]
        tool_model_display = next((dn for dn, mid in MODELS_BY_PROVIDER.get(TOOL_DECISION_PROVIDER.lower(), {}).get("models", {}).items() if mid == TOOL_DECISION_MODEL_ID), None) or get_default_model_display_name_for_provider(TOOL_DECISION_PROVIDER)
        
        if tool_model_display:
            try:
                tool_resp_raw = "".join(list(call_model_stream(provider=TOOL_DECISION_PROVIDER, model_display_name=tool_model_display, messages=tool_decision_messages, temperature=0.0, max_tokens=2048)))
                json_match_tool = re.search(r"\{.*\}", tool_resp_raw, re.DOTALL)
                if json_match_tool:
                    action_data = json.loads(json_match_tool.group(0))
                    action_type = action_data.get("action", "quick_respond")
                    action_input = action_data.get("action_input", {})
                    if not isinstance(action_input, dict):
                        action_input = {}
                    return action_type, action_input
            except Exception as e:
                logger.error(f"Tool decision LLM error: {e}")

    return "quick_respond", {}

def orchestrate_and_respond(user_input: str, provider_name: str, model_display_name: str, chat_history_for_prompt: list[dict], custom_system_prompt: str = None, ui_api_key_override: str = None):
    request_id = os.urandom(4).hex()
    logger.info(f"ORCHESTRATOR [{request_id}] Start. User: '{user_input[:50]}...'")
    history_str_for_prompt = "\n".join([f"{t['role']}: {t['content']}" for t in chat_history_for_prompt])
    
    yield "status", "[Checking guidelines...]"
    initial_insights = retrieve_rules_semantic(f"{user_input}\n{history_str_for_prompt}", k=5)
    initial_insights_ctx_str, parsed_initial_insights_list = format_insights_for_prompt(initial_insights)
    
    yield "status", "[Choosing best approach...]"
    action_type, action_input = decide_on_tool(user_input, chat_history_for_prompt, initial_insights_ctx_str)
    logger.info(f"ORCHESTRATOR [{request_id}]: Tool Decision: Action='{action_type}', Input='{action_input}'")
    
    yield "status", f"[Path: {action_type}]"
    final_system_prompt = custom_system_prompt or prompts.DEFAULT_SYSTEM_PROMPT
    context_str = None

    if action_type == "create_huggingface_space":
        yield "status", "[Tool: Creating Space...]"
        params = ["owner", "space_name", "sdk", "markdown_content"]
        if all(p in action_input for p in params):
            result = create_huggingface_space(**action_input)
            context_str = f"Tool Result (Create Space): {result.get('result') or result.get('error', 'Unknown outcome from tool.')}"
        else:
            context_str = "Tool Failed: Missing required parameters for create_huggingface_space. Required: " + ", ".join(params)
    elif action_type == "update_huggingface_space_file":
        yield "status", "[Tool: Updating file...]"
        params = ["owner", "space_name", "file_path", "new_content", "commit_message"]
        if all(p in action_input for p in params):
            result = update_huggingface_space_file(**action_input)
            context_str = f"Tool Result (Update File): {result.get('result') or result.get('error', 'Unknown outcome from tool.')}"
        else:
            context_str = "Tool Failed: Missing required parameters for update_huggingface_space_file. Required: " + ", ".join(params)
    elif action_type == "search_duckduckgo_and_report" and WEB_SEARCH_ENABLED:
        query = action_input.get("search_engine_query")
        if query:
            yield "status", f"[Web: '{query[:60]}'...]"
            results = search_and_scrape_duckduckgo(query, num_results=2)
            context_str = "Web Content:\n" + "\n".join([f"Source {i+1} ({r.get('url','N/A')}):\n{r.get('content', r.get('error', 'N/A'))[:3000]}\n---" for i, r in enumerate(results)])
        else:
            context_str = "Tool Failed: Missing 'search_engine_query' for web search."
    elif action_type == "scrape_url_and_report" and WEB_SEARCH_ENABLED:
        url = action_input.get("url")
        if url:
            yield "status", f"[Web: '{url[:60]}'...]"
            result = scrape_url(url)
            context_str = f"Web Content for {url}:\n{result.get('content', result.get('error', 'No content scraped.'))}"
        else:
            context_str = "Tool Failed: Missing 'url' for scraping."
    elif action_type == "answer_using_conversation_memory":
        yield "status", "[Searching conversation memory...]"
        mems = retrieve_memories_semantic(f"User query: {user_input}\nContext:\n{history_str_for_prompt[-1000:]}", k=2)
        context_str = "Relevant Past Interactions:\n" + "\n".join([f"- User:{m.get('user_input','')}->AI:{m.get('bot_response','')} (Takeaway:{m.get('metrics',{}).get('takeaway','N/A')})" for m in mems]) if mems else "No relevant past interactions found."
    
    final_user_prompt = prompts.get_final_response_prompt(history_str_for_prompt, initial_insights_ctx_str, user_input, context_str)
    final_llm_messages = [{"role": "system", "content": final_system_prompt}, {"role": "user", "content": final_user_prompt}]
    
    streamed_response = ""
    try:
        for chunk in call_model_stream(provider_name, model_display_name, final_llm_messages, ui_api_key_override, max_tokens=4096):
            if isinstance(chunk, str) and chunk.startswith("Error:"):
                streamed_response += f"\n{chunk}\n"; yield "response_chunk", f"\n{chunk}\n"; break
            streamed_response += chunk
            yield "response_chunk", chunk
    except Exception as e:
        streamed_response += f"\n\n(Error: {e})"; yield "response_chunk", f"\n\n(Error: {e})"
    
    final_bot_text = streamed_response.strip() or "(No response)"
    logger.info(f"ORCHESTRATOR [{request_id}]: Finished. Response length: {len(final_bot_text)}")
    yield "final_response_and_insights", {"response": final_bot_text, "insights_used": parsed_initial_insights_list}