refactor: 切换到 litellm，删除未使用的代码

- 添加 litellm 依赖，统一 LLM 调用层
- 新增 litellm_client.py 支持多厂商
- 更新 llm_catalog.yaml 添加 litellm_model 映射
- 删除旧的 cost_tracker.py (litellm 内置 cost 追踪)
- 删除未使用的 multimodal.py 和 storage_adapter.py
- 删除空文件 crewai/__init__.py, tools/__init__.py
- 更新测试以适配新代码

backend/tests/unit/core/agent/test_cost_tracker.py

@@ -2,81 +2,47 @@ from __future__ import annotations
 
 from decimal import Decimal
 
-import pytest
-
-from core.agent.cost_tracker import CostTracker
+from core.agent.litellm_client import get_model_cost
 
 
-def test_normalize_usage_and_cost_aggregation() -> None:
-    tracker = CostTracker()
-
-    tracker.add_usage(
-        {
-            "prompt_tokens": 7,
-            "completion_tokens": 5,
-            "cost": "0.002500",
-        }
-    )
-    tracker.add_usage(
-        {
-            "input_tokens": 5,
-            "output_tokens": 3,
-            "cost": "0.003000",
-            "currency": "USD",
-        }
-    )
-
-    snapshot = tracker.snapshot()
-
-    assert snapshot["input_tokens"] == 12
-    assert snapshot["output_tokens"] == 8
-    assert snapshot["total_tokens"] == 20
-    assert snapshot["cost"] == Decimal("0.005500")
-    assert snapshot["currency"] == "USD"
+def test_get_model_cost_returns_decimal() -> None:
+    usage = {
+        "prompt_tokens": 7,
+        "completion_tokens": 5,
+        "total_tokens": 12,
+        "cost": "0.002500",
+    }
+    cost = get_model_cost(usage)
+    assert cost == Decimal("0.002500")
 
 
-def test_add_usage_rejects_negative_values() -> None:
-    tracker = CostTracker()
-
-    with pytest.raises(ValueError):
-        tracker.add_usage({"input_tokens": -1})
-
-    with pytest.raises(ValueError):
-        tracker.add_usage({"cost": "-0.010000"})
+def test_get_model_cost_with_no_cost() -> None:
+    usage = {
+        "prompt_tokens": 7,
+        "completion_tokens": 5,
+        "total_tokens": 12,
+    }
+    cost = get_model_cost(usage)
+    assert cost == Decimal("0")
 
 
-def test_snapshot_is_zero_before_any_usage() -> None:
-    tracker = CostTracker()
-
-    snapshot = tracker.snapshot()
-
-    assert snapshot["input_tokens"] == 0
-    assert snapshot["output_tokens"] == 0
-    assert snapshot["total_tokens"] == 0
-    assert snapshot["cost"] == Decimal("0")
-    assert snapshot["currency"] == "USD"
+def test_get_model_cost_with_zero_cost() -> None:
+    usage = {
+        "prompt_tokens": 7,
+        "completion_tokens": 5,
+        "total_tokens": 12,
+        "cost": "0",
+    }
+    cost = get_model_cost(usage)
+    assert cost == Decimal("0")
 
 
-def test_add_usage_rejects_currency_mismatch() -> None:
-    tracker = CostTracker(currency="USD")
-    tracker.add_usage({"input_tokens": 1, "output_tokens": 1, "cost": "0.001000"})
-
-    with pytest.raises(ValueError):
-        tracker.add_usage(
-            {
-                "input_tokens": 1,
-                "output_tokens": 1,
-                "cost": "0.001000",
-                "currency": "CNY",
-            }
-        )
-
-
-def test_add_usage_rejects_non_integral_token_values() -> None:
-    tracker = CostTracker()
-
-    with pytest.raises(ValueError):
-        tracker.add_usage({"input_tokens": 1.5})
-
-    with pytest.raises(ValueError):
-        tracker.add_usage({"output_tokens": True})
+def test_get_model_cost_with_numeric_cost() -> None:
+    usage = {
+        "prompt_tokens": 7,
+        "completion_tokens": 5,
+        "total_tokens": 12,
+        "cost": 0.0025,
+    }
+    cost = get_model_cost(usage)
+    assert cost == Decimal("0.0025")

backend/tests/unit/core/agent/test_multimodal.py

@@ -1,89 +0,0 @@
-from __future__ import annotations
-
-import pytest
-
-from core.agent.multimodal import AttachmentInput, MultimodalProcessor
-from core.agent.storage_adapter import StorageAdapter
-from core.agent.tools.asr_fun_asr import FunASRTool
-
-
-def test_multimodal_processes_audio_and_builds_attachment_context() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-
-    def fake_transcribe(*, audio_bytes: bytes, filename: str) -> dict[str, str]:
-        assert audio_bytes == b"audio"
-        assert filename == "voice.wav"
-        return {"text": "hello world", "request_id": "req-1"}
-
-    processor = MultimodalProcessor(
-        storage=storage,
-        asr_tool=FunASRTool(transcribe_callable=fake_transcribe),
-        max_file_size_mb=1,
-    )
-
-    result = processor.process(
-        user_id="u1",
-        session_id="s1",
-        message_seq=4,
-        attachments=[
-            AttachmentInput(
-                filename="voice.wav",
-                mime_type="audio/wav",
-                content=b"audio",
-            )
-        ],
-    )
-
-    assert len(result.attachments) == 1
-    metadata = result.attachments[0]
-    assert (
-        metadata["object_path"]
-        == "agent-chat/u1/s1/4/6ed8919ce20490a5e3ad8630a4fab69475297abd07db73918dd5f36fcfaeb11b.wav"
-    )
-    assert metadata["mime_type"] == "audio/wav"
-    assert result.preview_texts == ["hello world"]
-
-
-def test_multimodal_rejects_unsupported_mime_type() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-    processor = MultimodalProcessor(
-        storage=storage, asr_tool=FunASRTool(lambda **_: {})
-    )
-
-    with pytest.raises(ValueError):
-        processor.process(
-            user_id="u1",
-            session_id="s1",
-            message_seq=1,
-            attachments=[
-                AttachmentInput(
-                    filename="malware.exe",
-                    mime_type="application/octet-stream",
-                    content=b"bad",
-                )
-            ],
-        )
-
-
-def test_multimodal_rejects_attachment_over_max_size() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-    processor = MultimodalProcessor(
-        storage=storage,
-        asr_tool=FunASRTool(lambda **_: {}),
-        max_file_size_mb=1,
-    )
-
-    oversized = b"x" * (1024 * 1024 + 1)
-    with pytest.raises(ValueError):
-        processor.process(
-            user_id="u1",
-            session_id="s1",
-            message_seq=1,
-            attachments=[
-                AttachmentInput(
-                    filename="big.wav",
-                    mime_type="audio/wav",
-                    content=oversized,
-                )
-            ],
-        )

backend/tests/unit/core/agent/test_storage_adapter.py

@@ -1,37 +0,0 @@
-from __future__ import annotations
-
-from core.agent.storage_adapter import StorageAdapter
-
-
-def test_build_object_path_uses_expected_pattern() -> None:
-    adapter = StorageAdapter(bucket="agent-chat-attachments")
-
-    path = adapter.build_object_path(
-        user_id="u1",
-        session_id="s1",
-        message_seq=3,
-        checksum_sha256="abc123",
-        extension="wav",
-    )
-
-    assert path == "agent-chat/u1/s1/3/abc123.wav"
-
-
-def test_build_attachment_metadata_contains_required_fields() -> None:
-    adapter = StorageAdapter(bucket="agent-chat-attachments")
-
-    metadata = adapter.build_attachment_metadata(
-        object_path="agent-chat/u1/s1/3/abc123.wav",
-        mime_type="audio/wav",
-        size=1024,
-        checksum_sha256="abc123",
-        origin="user_upload",
-        preview_text="hello",
-    )
-
-    assert metadata["object_path"] == "agent-chat/u1/s1/3/abc123.wav"
-    assert metadata["mime_type"] == "audio/wav"
-    assert metadata["size"] == 1024
-    assert metadata["checksum_sha256"] == "abc123"
-    assert metadata["origin"] == "user_upload"
-    assert metadata["preview_text"] == "hello"

backend/tests/unit/core/agent_chat/test_multimodal.py

@@ -1,89 +0,0 @@
-from __future__ import annotations
-
-import pytest
-
-from core.agent.multimodal import AttachmentInput, MultimodalProcessor
-from core.agent.storage_adapter import StorageAdapter
-from core.agent.tools.asr_fun_asr import FunASRTool
-
-
-def test_multimodal_processes_audio_and_builds_attachment_context() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-
-    def fake_transcribe(*, audio_bytes: bytes, filename: str) -> dict[str, str]:
-        assert audio_bytes == b"audio"
-        assert filename == "voice.wav"
-        return {"text": "hello world", "request_id": "req-1"}
-
-    processor = MultimodalProcessor(
-        storage=storage,
-        asr_tool=FunASRTool(transcribe_callable=fake_transcribe),
-        max_file_size_mb=1,
-    )
-
-    result = processor.process(
-        user_id="u1",
-        session_id="s1",
-        message_seq=4,
-        attachments=[
-            AttachmentInput(
-                filename="voice.wav",
-                mime_type="audio/wav",
-                content=b"audio",
-            )
-        ],
-    )
-
-    assert len(result.attachments) == 1
-    metadata = result.attachments[0]
-    assert (
-        metadata["object_path"]
-        == "agent-chat/u1/s1/4/6ed8919ce20490a5e3ad8630a4fab69475297abd07db73918dd5f36fcfaeb11b.wav"
-    )
-    assert metadata["mime_type"] == "audio/wav"
-    assert result.preview_texts == ["hello world"]
-
-
-def test_multimodal_rejects_unsupported_mime_type() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-    processor = MultimodalProcessor(
-        storage=storage, asr_tool=FunASRTool(lambda **_: {})
-    )
-
-    with pytest.raises(ValueError):
-        processor.process(
-            user_id="u1",
-            session_id="s1",
-            message_seq=1,
-            attachments=[
-                AttachmentInput(
-                    filename="malware.exe",
-                    mime_type="application/octet-stream",
-                    content=b"bad",
-                )
-            ],
-        )
-
-
-def test_multimodal_rejects_attachment_over_max_size() -> None:
-    storage = StorageAdapter(bucket="agent-chat-attachments")
-    processor = MultimodalProcessor(
-        storage=storage,
-        asr_tool=FunASRTool(lambda **_: {}),
-        max_file_size_mb=1,
-    )
-
-    oversized = b"x" * (1024 * 1024 + 1)
-    with pytest.raises(ValueError):
-        processor.process(
-            user_id="u1",
-            session_id="s1",
-            message_seq=1,
-            attachments=[
-                AttachmentInput(
-                    filename="big.wav",
-                    mime_type="audio/wav",
-                    content=oversized,
-                )
-            ],
-        )

backend/tests/unit/core/agent_chat/test_storage_adapter.py

@@ -1,37 +0,0 @@
-from __future__ import annotations
-
-from core.agent.storage_adapter import StorageAdapter
-
-
-def test_build_object_path_uses_expected_pattern() -> None:
-    adapter = StorageAdapter(bucket="agent-chat-attachments")
-
-    path = adapter.build_object_path(
-        user_id="u1",
-        session_id="s1",
-        message_seq=3,
-        checksum_sha256="abc123",
-        extension="wav",
-    )
-
-    assert path == "agent-chat/u1/s1/3/abc123.wav"
-
-
-def test_build_attachment_metadata_contains_required_fields() -> None:
-    adapter = StorageAdapter(bucket="agent-chat-attachments")
-
-    metadata = adapter.build_attachment_metadata(
-        object_path="agent-chat/u1/s1/3/abc123.wav",
-        mime_type="audio/wav",
-        size=1024,
-        checksum_sha256="abc123",
-        origin="user_upload",
-        preview_text="hello",
-    )
-
-    assert metadata["object_path"] == "agent-chat/u1/s1/3/abc123.wav"
-    assert metadata["mime_type"] == "audio/wav"
-    assert metadata["size"] == 1024
-    assert metadata["checksum_sha256"] == "abc123"
-    assert metadata["origin"] == "user_upload"
-    assert metadata["preview_text"] == "hello"