Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 67 additions & 0 deletions openspace/llm_factory.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,67 @@
"""LLMFactory — centralized LLM client creation from config.

Extracted from OpenSpace.initialize() in Epic 4.5. Owns:
• Main LLM client creation with full config
• Optional tool retrieval LLM creation (shares credentials via llm_kwargs)
"""

from __future__ import annotations

from typing import TYPE_CHECKING, Optional

from openspace.llm import LLMClient
from openspace.utils.logging import Logger

if TYPE_CHECKING:
from openspace.tool_layer import OpenSpaceConfig

logger = Logger.get_logger(__name__)


class LLMFactory:
"""Creates LLM clients from OpenSpaceConfig."""

def __init__(self, *, config: OpenSpaceConfig) -> None:
self._config = config
self._llm_client: Optional[LLMClient] = None
self._tool_retrieval_llm: Optional[LLMClient] = None

@property
def llm_client(self) -> Optional[LLMClient]:
return self._llm_client

@property
def tool_retrieval_llm(self) -> Optional[LLMClient]:
return self._tool_retrieval_llm

def create_main(self) -> LLMClient:
"""Create the primary LLM client from config."""
self._llm_client = LLMClient(
model=self._config.llm_model,
enable_thinking=self._config.llm_enable_thinking,
rate_limit_delay=self._config.llm_rate_limit_delay,
max_retries=self._config.llm_max_retries,
timeout=self._config.llm_timeout,
**self._config.llm_kwargs,
)
return self._llm_client

def create_tool_retrieval(self) -> Optional[LLMClient]:
"""Create optional tool retrieval LLM. Returns None if not configured.

Inherits llm_kwargs (api_key, api_base, etc.) so credentials
from the host agent are shared across all internal LLM clients.

Note: intentionally omits ``enable_thinking`` and ``rate_limit_delay``
since tool retrieval is a simple selection task, not a reasoning task.
"""
if not self._config.tool_retrieval_model:
return None

self._tool_retrieval_llm = LLMClient(
model=self._config.tool_retrieval_model,
timeout=self._config.llm_timeout,
max_retries=self._config.llm_max_retries,
**self._config.llm_kwargs,
)
return self._tool_retrieval_llm
24 changes: 6 additions & 18 deletions openspace/tool_layer.py
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
from openspace.grounding.core.grounding_client import GroundingClient
from openspace.execution_engine import ExecutionEngine
from openspace.llm import LLMClient
from openspace.llm_factory import LLMFactory
from openspace.recording_service import RecordingService
from openspace.skill_engine import ExecutionAnalyzer, SkillRegistry, SkillStore
from openspace.skill_engine.evolver import SkillEvolver
Expand Down Expand Up @@ -125,6 +126,7 @@ def __init__(
self._container = container or AppContainer()

self._llm_client: Optional[LLMClient] = None
self._llm_factory: Optional[LLMFactory] = None
self._grounding_client: Optional[GroundingClient] = None
self._grounding_config = None # GroundingConfig reference for skill settings
self._grounding_agent: Optional[GroundingAgent] = None
Expand Down Expand Up @@ -205,14 +207,8 @@ async def initialize(self) -> None:
logger.info("Initializing OpenSpace...")

try:
self._llm_client = LLMClient(
model=self.config.llm_model,
enable_thinking=self.config.llm_enable_thinking,
rate_limit_delay=self.config.llm_rate_limit_delay,
max_retries=self.config.llm_max_retries,
timeout=self.config.llm_timeout,
**self.config.llm_kwargs,
)
self._llm_factory = LLMFactory(config=self.config)
self._llm_client = self._llm_factory.create_main()
logger.info(f"✓ LLM Client: {self.config.llm_model}")

# Load grounding config
Expand Down Expand Up @@ -300,16 +296,8 @@ async def initialize(self) -> None:
logger.info(f"✓ Recording enabled: {len(self._recording_manager.backends or [])} backends")

# Create separate LLM client for tool retrieval if configured
# Inherits llm_kwargs (api_key, api_base, etc.) so credentials
# from the host agent are shared across all internal LLM clients.
tool_retrieval_llm = None
if self.config.tool_retrieval_model:
tool_retrieval_llm = LLMClient(
model=self.config.tool_retrieval_model,
timeout=self.config.llm_timeout,
max_retries=self.config.llm_max_retries,
**self.config.llm_kwargs,
)
tool_retrieval_llm = self._llm_factory.create_tool_retrieval()
if tool_retrieval_llm:
logger.info(f"✓ Tool retrieval LLM: {self.config.tool_retrieval_model}")

self._grounding_agent = GroundingAgent(
Expand Down
217 changes: 217 additions & 0 deletions tests/test_llm_factory.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,217 @@
"""Tests for LLMFactory — extracted LLM client creation from OpenSpace."""

from __future__ import annotations

import pytest
from unittest.mock import MagicMock, patch, call

try:
from openspace.tool_layer import OpenSpace, OpenSpaceConfig
from openspace.llm_factory import LLMFactory

_HAS_TOOL_LAYER = True
except Exception:
_HAS_TOOL_LAYER = False

pytestmark = pytest.mark.skipif(not _HAS_TOOL_LAYER, reason="tool_layer deps unavailable")


# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------

@pytest.fixture
def config():
return OpenSpaceConfig(
llm_model="openrouter/anthropic/claude-sonnet-4.5",
llm_enable_thinking=True,
llm_timeout=60.0,
llm_max_retries=5,
llm_rate_limit_delay=0.5,
llm_kwargs={"api_key": "sk-test"},
tool_retrieval_model="openrouter/openai/gpt-4o",
)


@pytest.fixture
def minimal_config():
return OpenSpaceConfig()


# ---------------------------------------------------------------------------
# LLMFactory.__init__
# ---------------------------------------------------------------------------

class TestLLMFactoryInit:

def test_initial_state(self, config):
factory = LLMFactory(config=config)
assert factory.llm_client is None
assert factory.tool_retrieval_llm is None

def test_stores_config(self, config):
factory = LLMFactory(config=config)
assert factory._config is config


# ---------------------------------------------------------------------------
# LLMFactory.create_main()
# ---------------------------------------------------------------------------

class TestCreateMain:

def test_creates_llm_client(self, config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
mock_client = MagicMock()
MockLLM.return_value = mock_client

factory = LLMFactory(config=config)
result = factory.create_main()

assert result is mock_client
assert factory.llm_client is mock_client

def test_passes_all_config_fields(self, config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
MockLLM.return_value = MagicMock()

factory = LLMFactory(config=config)
factory.create_main()

MockLLM.assert_called_once_with(
model="openrouter/anthropic/claude-sonnet-4.5",
enable_thinking=True,
rate_limit_delay=0.5,
max_retries=5,
timeout=60.0,
api_key="sk-test",
)

def test_uses_defaults_when_minimal_config(self, minimal_config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
MockLLM.return_value = MagicMock()

factory = LLMFactory(config=minimal_config)
factory.create_main()

MockLLM.assert_called_once_with(
model="openrouter/anthropic/claude-sonnet-4.5",
enable_thinking=False,
rate_limit_delay=0.0,
max_retries=3,
timeout=120.0,
)

def test_create_main_twice_replaces_client(self, config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
first = MagicMock()
second = MagicMock()
MockLLM.side_effect = [first, second]

factory = LLMFactory(config=config)
factory.create_main()
assert factory.llm_client is first
factory.create_main()
assert factory.llm_client is second


# ---------------------------------------------------------------------------
# LLMFactory.create_tool_retrieval()
# ---------------------------------------------------------------------------

class TestCreateToolRetrieval:

def test_creates_when_model_configured(self, config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
mock_client = MagicMock()
MockLLM.return_value = mock_client

factory = LLMFactory(config=config)
result = factory.create_tool_retrieval()

assert result is mock_client
assert factory.tool_retrieval_llm is mock_client

def test_returns_none_when_no_model(self, minimal_config):
factory = LLMFactory(config=minimal_config)
result = factory.create_tool_retrieval()
assert result is None
assert factory.tool_retrieval_llm is None

def test_passes_correct_config(self, config):
with patch("openspace.llm_factory.LLMClient") as MockLLM:
MockLLM.return_value = MagicMock()

factory = LLMFactory(config=config)
factory.create_tool_retrieval()

MockLLM.assert_called_once_with(
model="openrouter/openai/gpt-4o",
timeout=60.0,
max_retries=5,
api_key="sk-test",
)

def test_inherits_llm_kwargs(self, config):
"""Tool retrieval LLM inherits credentials from llm_kwargs."""
config.llm_kwargs = {"api_key": "sk-shared", "api_base": "https://custom"}
with patch("openspace.llm_factory.LLMClient") as MockLLM:
MockLLM.return_value = MagicMock()

factory = LLMFactory(config=config)
factory.create_tool_retrieval()

kwargs = MockLLM.call_args.kwargs
assert kwargs["api_key"] == "sk-shared"
assert kwargs["api_base"] == "https://custom"

def test_returns_none_when_empty_string_model(self, config):
"""Empty string model treated as unconfigured."""
config.tool_retrieval_model = ""
factory = LLMFactory(config=config)
result = factory.create_tool_retrieval()
assert result is None
assert factory.tool_retrieval_llm is None

def test_create_tool_retrieval_twice_replaces(self, config):
"""Second call replaces the tool retrieval client."""
with patch("openspace.llm_factory.LLMClient") as MockLLM:
first, second = MagicMock(), MagicMock()
MockLLM.side_effect = [first, second]

factory = LLMFactory(config=config)
factory.create_tool_retrieval()
assert factory.tool_retrieval_llm is first
factory.create_tool_retrieval()
assert factory.tool_retrieval_llm is second


# ---------------------------------------------------------------------------
# Error handling
# ---------------------------------------------------------------------------

class TestErrorHandling:

def test_create_main_exception_propagates(self, config):
"""LLMClient constructor failure propagates, client stays None."""
with patch("openspace.llm_factory.LLMClient", side_effect=RuntimeError("boom")):
factory = LLMFactory(config=config)
with pytest.raises(RuntimeError, match="boom"):
factory.create_main()
assert factory.llm_client is None


# ---------------------------------------------------------------------------
# OpenSpace backward compatibility
# ---------------------------------------------------------------------------

class TestOpenSpaceDelegation:

def test_openspace_has_llm_factory_attr(self):
os_instance = OpenSpace()
assert hasattr(os_instance, "_llm_factory")

def test_llm_client_still_accessible(self):
os_instance = OpenSpace()
assert hasattr(os_instance, "_llm_client")
assert os_instance._llm_client is None
Loading