import asyncio
from memforge import ConversationMemory

async def main():
    async with ConversationMemory(agent_id="my-bot") as memory:
        # Store conversation turns
        await memory.add_turn("user", "I prefer dark mode and vim keybindings")
        await memory.add_turn("assistant", "Noted! I'll remember your preferences.")

        # Get relevant context for the next turn
        context = await memory.get_context("What are my preferences?", max_tokens=2000)
        print(context)

        # End session — consolidate memories
        await memory.end_session()

asyncio.run(main())

from memforge import MemForgeClient

async with MemForgeClient(base_url="http://localhost:3333", token="...") as client:
    # Store
    await client.add("agent-1", "Deployment uses GitHub Actions")

    # Search (keyword, semantic, hybrid, or code mode)
    results = await client.query("agent-1", q="deployment", mode="hybrid", limit=5)

    # Budget-controlled retrieval
    results = await client.query("agent-1", q="deployment", max_tokens=2000)

    # Knowledge graph
    entities = await client.search_entities("agent-1", q="Alice")
    graph = await client.graph_traverse("agent-1", entity="Alice", depth=2)

    # Sleep cycle (consolidate, revise, reflect)
    await client.sleep("agent-1")

    # Session resumption (warm-start context)
    context = await client.resume("agent-1")

from memforge import ResilientMemForgeClient

client = ResilientMemForgeClient(on_error=lambda e: print(f"memforge: {e}"))
results = await client.query("agent-1", q="test")  # returns [] on failure, never throws

from memforge.tools import openai_tools, anthropic_tools

# OpenAI function calling
response = openai.chat.completions.create(tools=openai_tools(), ...)

# Anthropic tool use
response = anthropic.messages.create(tools=anthropic_tools(), ...)