quickstart.py

"""
Quickstart: Get a dream engine running in 10 lines.

This example uses a simple callable as the LLM backend.
Replace it with OpenAIAdapter or AnthropicAdapter for real use.
"""

import json
from dream_memory import DreamEngine, MemoryStore, Memory
from dream_memory.llm import CallableLLM


# 1. Simple mock LLM (replace with real one)
def my_llm(prompt: str, max_tokens: int = 2000) -> str:
    """Fake LLM for demo purposes."""
    if "analyzing" in prompt:
        return json.dumps({
            "stale_topics": [], "redundant_groups": [],
            "new_signals": ["User prefers Python"], "contradictions": [],
            "health_score": 0.8,
        })
    elif "extracting" in prompt:
        return json.dumps({
            "memories": [
                {"content": "User prefers Python for data work", "topic": "preferences", "priority": 2},
            ]
        })
    elif "pruning" in prompt:
        return json.dumps({"decisions": []})
    return json.dumps({"merged": [], "updated": []})


# 2. Create store and engine
store = MemoryStore("./demo_memories")
engine = DreamEngine(store, llm=CallableLLM(my_llm))

# 3. Add some memories manually
engine.add_memory("User name is Alice", topic="identity", priority=4)
engine.add_memory("Timezone is PST", topic="preferences")

# 4. Log sessions (normally from your agent's conversations)
engine.log_session("User asked about pandas DataFrames and preferred Python solutions.")
engine.log_session("User set up a new ML pipeline. Discussed sklearn vs pytorch.")
engine.log_session("User mentioned they always use dark mode.")

# 5. Dream!
result = engine.dream()

# 6. Check results
print(f"Dream completed!")
print(f"  Sessions processed: {result.sessions_processed}")
print(f"  Memories: {result.memories_before} -> {result.memories_after}")
print(f"  Added: {result.total_added}, Pruned: {result.total_pruned}")
print()
print("Current memories:")
for mem in store.list_memories():
    print(f"  [{mem.topic}] {mem.content}")

# Cleanup demo files
import shutil
shutil.rmtree("./demo_memories", ignore_errors=True)