From 81b5bfff35d38ce13c59f1f76b1a1ffc0ac62d0d Mon Sep 17 00:00:00 2001
From: Mikkel Georgsen <msgeorgsen@gmail.com>
Date: Fri, 16 Jan 2026 19:36:13 +0000
Subject: [PATCH] feat(05-02): create orchestrator with parallel model queries

Add orchestrator module with SYSTEM_PROMPT constant and query_models_parallel
function that uses asyncio.gather() to query all models simultaneously.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---
 src/moai/core/orchestrator.py | 71 +++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)
 create mode 100644 src/moai/core/orchestrator.py

diff --git a/src/moai/core/orchestrator.py b/src/moai/core/orchestrator.py
new file mode 100644
index 0000000..5116a8e
--- /dev/null
+++ b/src/moai/core/orchestrator.py
@@ -0,0 +1,71 @@
+"""AI orchestrator for managing multi-model discussions.
+
+Provides functions for orchestrating parallel and sequential AI queries
+across multiple models, building context, and managing discussion flow.
+"""
+
+import asyncio
+import logging
+
+from moai.core.ai_client import get_ai_client
+
+logger = logging.getLogger(__name__)
+
+# System prompt for roundtable discussions
+SYSTEM_PROMPT = """You are participating in a roundtable discussion with other AI models.
+Other participants: {models}
+Current topic: {topic}
+
+Guidelines:
+- Be concise but substantive
+- You can agree, disagree, or build upon others' points
+- Reference other models by name when responding to their points
+- Focus on practical, actionable insights
+- If you reach agreement with others, state it clearly"""
+
+
+async def query_models_parallel(
+    models: list[str],
+    question: str,
+    project_name: str,
+) -> dict[str, str]:
+    """Query multiple AI models in parallel.
+
+    Sends the same question to all models simultaneously and collects
+    their responses. Each model receives a system prompt identifying
+    the other participants and the discussion topic.
+
+    Args:
+        models: List of model short names (e.g., ["claude", "gpt", "gemini"]).
+        question: The question to ask all models.
+        project_name: The project name for context.
+
+    Returns:
+        Dict mapping model name to response text. If a model fails,
+        its value will be an error message string.
+    """
+    client = get_ai_client()
+
+    # Build system prompt with participant info
+    models_str = ", ".join(models)
+    system_prompt = SYSTEM_PROMPT.format(models=models_str, topic=project_name)
+
+    async def query_single_model(model: str) -> tuple[str, str]:
+        """Query a single model and return (model_name, response)."""
+        try:
+            response = await client.complete(
+                model=model,
+                messages=[{"role": "user", "content": question}],
+                system_prompt=system_prompt,
+            )
+            logger.info("Model %s responded successfully", model)
+            return (model, response)
+        except Exception as e:
+            logger.error("Model %s failed: %s", model, e)
+            return (model, f"[Error: {e}]")
+
+    # Run all queries in parallel
+    tasks = [query_single_model(model) for model in models]
+    results = await asyncio.gather(*tasks)
+
+    return dict(results)