ai-talk-show/scripts/orchestrator.py

#!/usr/bin/env python3
"""
AI Talk Show — Episode Orchestrator
Manages turn-taking between two autonomous agents (Leonard & Charlie).
Each agent gets its own system prompt and sees the full conversation history.
"""

import json
import time
import sys
import requests
from pathlib import Path

# ── Config ──────────────────────────────────────────────────────────
BASE_URL = "http://100.64.0.2:39195/v1/chat/completions"
MODEL = "qwen36-27b-nvfp4-mtp-gguf"
TEMPERATURE = 0.8
MAX_TURNS = 8  # 4 exchanges each (Leonard opens)
MAX_TOKENS_OPENING = 250
MAX_TOKENS_TURN = 180
PROJECT_DIR = Path(__file__).parent.parent
OUTPUT_DIR = PROJECT_DIR / "outputs"
OUTPUT_DIR.mkdir(exist_ok=True)

# ── Load prompts ────────────────────────────────────────────────────
LEONARD_SYSTEM = (PROJECT_DIR / "prompts" / "leonard_system.md").read_text().strip()
CHARLIE_SYSTEM = (PROJECT_DIR / "prompts" / "charlie_system.md").read_text().strip()
PRODUCER_BRIEF = (PROJECT_DIR / "prompts" / "producer_brief_003.md").read_text().strip()

# ── Helpers ─────────────────────────────────────────────────────────
def call_agent(system: str, messages: list, max_tokens: int) -> str:
    """Call the model API with the agent's system prompt and conversation history."""
    payload = {
        "model": MODEL,
        "messages": [{"role": "system", "content": system}] + messages,
        "max_tokens": max_tokens,
        "temperature": TEMPERATURE,
    }
    resp = requests.post(BASE_URL, json=payload, timeout=120)
    resp.raise_for_status()
    data = resp.json()
    content = data["choices"][0]["message"]["content"].strip()
    usage = data.get("usage", {})
    return content, usage

def call_agent_turn(agent_name: str, conversation: list, is_opening: bool) -> dict:
    """Run one turn for an agent."""
    system = LEONARD_SYSTEM if agent_name == "Leonard" else CHARLIE_SYSTEM
    max_tokens = MAX_TOKENS_OPENING if is_opening else MAX_TOKENS_TURN

    # Build conversation history
    history = []
    for turn in conversation:
        speaker = turn["agent"]
        history.append({"role": "user", "content": f"{speaker}: {turn['content']}"})

    # Leonard's opening: give him the raw brief and let him start naturally
    if agent_name == "Leonard" and is_opening:
        history.insert(0, {"role": "user", "content": f"Here's what's new in Hermes v0.18:\n\n{PRODUCER_BRIEF}"})

    content, usage = call_agent(system, history, max_tokens)

    # Word count check
    word_count = len(content.split())
    print(f"  {agent_name} turn {len(conversation)+1}: {word_count} words "
          f"(prompt: {usage.get('prompt_tokens', '?')}, "
          f"completion: {usage.get('completion_tokens', '?')})", flush=True)

    return {
        "agent": agent_name,
        "content": content,
        "word_count": word_count,
        "turn": len(conversation) + 1,
        "tokens": usage,
    }

# ── Main ────────────────────────────────────────────────────────────
def main():
    episode_id = "003"
    topic = "hermes-v018"
    print(f"=== AI Talk Show — Episode {episode_id}: {topic} ===\n", flush=True)

    conversation = []
    turn = 0

    while turn < MAX_TURNS:
        # Determine speaker: Leonard opens, then alternate
        if turn == 0:
            speaker = "Leonard"
            is_opening = True
        elif turn % 2 == 1:
            speaker = "Charlie"
            is_opening = False
        else:
            speaker = "Leonard"
            is_opening = False

        print(f"\n--- Turn {turn+1}/{MAX_TURNS}: {speaker} ---", flush=True)
        result = call_agent_turn(speaker, conversation, is_opening)
        conversation.append(result)

        # Print a short excerpt
        excerpt = result["content"][:120].replace("\n", " ")
        print(f"  → {excerpt}...", flush=True)

        turn += 1
        if turn < MAX_TURNS:
            time.sleep(0.5)  # Brief pause between turns

    # ── Save transcript ────────────────────────────────────────────
    transcript_path = OUTPUT_DIR / f"ep{episode_id}_{topic}_transcript.json"
    transcript_data = {
        "episode": episode_id,
        "topic": topic,
        "turns": len(conversation),
        "conversation": conversation,
    }
    transcript_path.write_text(json.dumps(transcript_data, indent=2))
    print(f"\n✅ Transcript saved: {transcript_path}", flush=True)

    # ── Save readable transcript ───────────────────────────────────
    readable_path = OUTPUT_DIR / f"ep{episode_id}_{topic}_readable.txt"
    lines = []
    lines.append(f"AI TALK SHOW — Episode {episode_id}\n")
    lines.append(f"Topic: Hermes v0.18\n")
    lines.append("=" * 60 + "\n\n")
    for t in conversation:
        lines.append(f"**{t['agent']}**\n")
        lines.append(t["content"] + "\n\n")
    readable_path.write_text("\n".join(lines))
    print(f"✅ Readable transcript: {readable_path}", flush=True)

    # ── Word count summary ─────────────────────────────────────────
    total_words = sum(t["word_count"] for t in conversation)
    print(f"\n📊 Total: {total_words} words across {len(conversation)} turns", flush=True)
    estimated_minutes = total_words / 150
    print(f"   Estimated runtime: {estimated_minutes:.1f} minutes", flush=True)

if __name__ == "__main__":
    main()