DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.
Python ai-agentsai-tutorclawbotcli-tooldeepresearchinteractive-learninglearningmulti-agent-systemsrag
"""Opt-in provider proof: DEEPTUTOR_CACHE_E2E=1 pytest -s <this file>.
Uses the configured model with synthetic prompts and a local lookup tool.
No user conversations are changed; the session database lives in tmp_path.
"""
from __future__ import annotations
import asyncio
from copy import deepcopy
import json
import os
from types import SimpleNamespace
from unittest.mock import AsyncMock
import pytest
from deeptutor.agents.chat.agentic_pipeline import AgenticChatPipeline
from deeptutor.core.context import TurnRuntimeContext, UnifiedContext, WorkspaceRuntimeContext
from deeptutor.core.tool_protocol import BaseTool, ToolDefinition, ToolResult
from deeptutor.runtime.registry.tool_registry import ToolRegistry
from deeptutor.runtime.stream_bus import StreamBus
from deeptutor.services.llm.metrics import TurnUsage, current_usage
from deeptutor.services.session.context_builder import ContextBuilder
from deeptutor.services.session.sqlite_store import SQLiteSessionStore
pytestmark = [
pytest.mark.integration,
pytest.mark.real_llm_resolver,
pytest.mark.skipif(
os.getenv("DEEPTUTOR_CACHE_E2E") != "1", reason="explicit real API opt-in required"
),
]
class Lookup(BaseTool):
def get_definition(self):
return ToolDefinition(
name="cache_lookup",
description="Return the stored deployment color. Call once when asked to look it up.",
)
async def execute(self, **kwargs):
return ToolResult(content="The stored deployment color is azure-falcon-42.")
@pytest.mark.asyncio
async def test_real_provider_cache_survives_tool_turn_and_database_resume(monkeypatch, tmp_path):
from deeptutor.runtime.agentic.client import close_agentic_client_pool
from deeptutor.services.llm.config import get_llm_config
config = get_llm_config()
registry = ToolRegistry()
registry.register(Lookup())
requests = []
usage = TurnUsage()
token = current_usage.set(usage)
async def run(context, force_tool=False):
pipeline = AgenticChatPipeline(language="en", max_tokens=1024, max_rounds=3)
pipeline.initial_tool_choice = "cache_lookup" if force_tool else None
pipeline.registry = registry
monkeypatch.setattr(pipeline, "_prepare_deferred_tools", AsyncMock())
monkeypatch.setattr(pipeline, "_prepare_kb_manifests", AsyncMock())
monkeypatch.setattr(pipeline, "_build_notebook_manifest", lambda: "")
monkeypatch.setattr(pipeline, "_compose_enabled_tools", lambda _: ["cache_lookup"])
original = pipeline._build_openai_client()
async def create(**kwargs):
requests.append(
deepcopy({"messages": kwargs["messages"], "tools": kwargs.get("tools")})
)
return await original.chat.completions.create(**kwargs)
client = SimpleNamespace(chat=SimpleNamespace(completions=SimpleNamespace(create=create)))
monkeypatch.setattr(pipeline, "_build_openai_client", lambda: client)
bus = StreamBus()
async def consume():
async for _ in bus.subscribe():
pass
consumer = asyncio.create_task(consume())
await asyncio.sleep(0)
try:
await asyncio.wait_for(pipeline.run(context, bus), 120)
return pipeline.last_result
finally:
await bus.close()
await consumer
try:
store = SQLiteSessionStore(db_path=tmp_path / "session.db")
sid = (await store.create_session())["id"]
first = UnifiedContext(
session_id=sid,
user_message="Look up the deployment color with cache_lookup, then repeat the exact value in one short sentence.",
runtime=TurnRuntimeContext(
workspace=WorkspaceRuntimeContext(logical_output_dir="outputs/cache-test/turn-1")
),
)
result = await run(first, force_tool=True)
assert "azure-falcon-42" in result["response"]
await store.add_message(sid, "user", first.user_message)
await store.add_message(
sid, "assistant", result["response"], metadata={"model_turn": first.runtime.model_turn}
)
first_turn_calls = len(requests)
assert first_turn_calls >= 2
restored = await ContextBuilder(SQLiteSessionStore(db_path=tmp_path / "session.db")).build(
session_id=sid, llm_config=config
)
second = UnifiedContext(
session_id=sid,
user_message="Repeat that stored color once. Do not call any tools.",
runtime=TurnRuntimeContext(
model_history=restored.model_history,
previous_model_turn=restored.previous_model_turn,
workspace=WorkspaceRuntimeContext(logical_output_dir="outputs/cache-test/turn-2"),
),
)
result = await run(second)
assert "azure-falcon-42" in result["response"]
before, after = requests[first_turn_calls - 1], requests[first_turn_calls]
assert after["tools"] == before["tools"]
assert after["messages"][: len(before["messages"])] == before["messages"]
summary, _ = await ContextBuilder(store)._summarize(
session_id=sid,
language="en",
source_text="fallback",
summary_budget=512,
replay_request={"messages": before["messages"], "tools": before["tools"]},
)
assert "azure-falcon-42" in summary
counts = [
{k: call.get(k) for k in ("prompt_tokens", "cache_read_input_tokens", "cache_hit_rate")}
for call in usage.calls
]
print(json.dumps({"provider": config.binding, "model": config.model, "calls": counts}))
assert usage.calls[first_turn_calls].get("cache_read_input_tokens", 0) > 0
finally:
current_usage.reset(token)
await close_agentic_client_pool()