agstack 1.8.3__tar.gz → 1.9.0__tar.gz

{agstack-1.8.3 → agstack-1.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agstack
-Version: 1.8.3
+Version: 1.9.0
 Summary: Production-ready toolkit for building FastAPI and LLM applications
 Author-email: XtraVisions <gitadmin@xtravisions.com>, Chen Hao <chenhao@xtravisions.com>
 Maintainer-email: XtraVisions <gitadmin@xtravisions.com>, Chen Hao <chenhao@xtravisions.com>
@@ -34,8 +34,7 @@ Requires-Dist: pydantic>=2.12.4
 Requires-Dist: python-multipart>=0.0.20
 Requires-Dist: requests>=2.32.5
 Requires-Dist: RestrictedPython>=7.0
-Requires-Dist: sqlalchemy[asyncio]>=2.0.48
-Requires-Dist: sqlobjects>=1.6.0
+Requires-Dist: sqlobjects>=1.9.0
 Requires-Dist: tiktoken>=0.12.0
 Requires-Dist: uvicorn>=0.41.0
 Dynamic: license-file

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/nodes/detect_node.py

@@ -47,10 +47,12 @@ class DetectNodeHandler(NodeHandler):
         resolved_inputs = self.resolve_inputs(config, context)
 
         query = resolved_inputs.get("query", "")
-        instruction = config.get("instruction", "Classify the input")
-        options = config.get("options", [])
-        model = config.get("model", "gpt-4o-mini")
-        temperature = config.get("temperature", 0.0)
+        instruction = resolved_inputs.get("instruction") or config.get("instruction", "Classify the input")
+        _options = resolved_inputs.get("options")
+        options = _options if _options is not None else config.get("options", [])
+        model = resolved_inputs.get("model") or config.get("model", "gpt-4o-mini")
+        _raw_temp = resolved_inputs.get("temperature")
+        temperature: float = float(_raw_temp) if _raw_temp is not None else float(config.get("temperature", 0.0))
 
         messages = self._build_classification_prompt(instruction, options, query)
 

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/nodes/llm_chat_node.py

@@ -47,9 +47,11 @@ class LLMChatNodeHandler(NodeHandler):
         resolved_inputs = self.resolve_inputs(config, context)
         prompt_text = self._build_prompt(config.get("prompt", ""), resolved_inputs)
 
-        model = config.get("model", "gpt-4o")
-        temperature = config.get("temperature", 0.7)
-        max_tokens = config.get("max_tokens")
+        model = resolved_inputs.get("model") or config.get("model", "gpt-4o")
+        _temp = resolved_inputs.get("temperature")
+        temperature: float = float(_temp) if _temp is not None else float(config.get("temperature", 0.7))
+        _max = resolved_inputs.get("max_tokens")
+        max_tokens = _max if _max is not None else config.get("max_tokens")
 
         client = get_llm_client()
         messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": prompt_text}]
@@ -103,9 +105,11 @@ class LLMChatNodeHandler(NodeHandler):
         resolved_inputs = self.resolve_inputs(config, context)
         prompt_text = self._build_prompt(config.get("prompt", ""), resolved_inputs)
 
-        model = config.get("model", "gpt-4o")
-        temperature = config.get("temperature", 0.7)
-        max_tokens = config.get("max_tokens")
+        model = resolved_inputs.get("model") or config.get("model", "gpt-4o")
+        _temp = resolved_inputs.get("temperature")
+        temperature: float = float(_temp) if _temp is not None else float(config.get("temperature", 0.7))
+        _max = resolved_inputs.get("max_tokens")
+        max_tokens = _max if _max is not None else config.get("max_tokens")
 
         client = get_llm_client()
         messages: list[ChatCompletionMessageParam] = [{"role": "user", "content": prompt_text}]

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/nodes/llm_embed_node.py

@@ -29,7 +29,7 @@ class LLMEmbedNodeHandler(NodeHandler):
         if isinstance(texts, str):
             texts = [texts]
 
-        model = config.get("model", "bge-m3")
+        model = resolved_inputs.get("model") or config.get("model", "bge-m3")
 
         client = get_llm_client()
         embeddings = await client.embed(texts=texts, model=model)

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/nodes/llm_rerank_node.py

@@ -30,8 +30,9 @@ class LLMRerankNodeHandler(NodeHandler):
         if isinstance(documents, str):
             documents = [documents]
 
-        model = config.get("model", "bge-reranker-v2-m3")
-        top_n = config.get("top_n", 10)
+        model = resolved_inputs.get("model") or config.get("model", "bge-reranker-v2-m3")
+        _top_n = resolved_inputs.get("top_n")
+        top_n = _top_n if _top_n is not None else config.get("top_n", 10)
 
         client = get_llm_client()
         raw_results = await client.rerank(

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/nodes/tool_node.py

@@ -4,7 +4,7 @@
 
 from typing import TYPE_CHECKING, Any
 
-from ..exceptions import FlowError
+from ..exceptions import FlowError, ToolExecutionError
 from ..registry import registry
 from .base import NodeHandler
 
@@ -31,4 +31,7 @@ class ToolNodeHandler(NodeHandler):
         config = node.get("config", {})
         resolved = self.resolve_inputs(config, context)
         tool = self._create_tool(config)
-        return await tool.run(context, inputs=resolved)
+        result = await tool.execute_async(context, inputs=resolved)
+        if not result.success:
+            raise ToolExecutionError("TOOL_EXECUTION_FAILED", args={"tool_name": tool.name, "error": result.error})
+        return result.result

{agstack-1.8.3 → agstack-1.9.0}/agstack/llm/flow/tool.py

@@ -2,6 +2,7 @@
 
 """工具定义和执行"""
 
+import logging
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, Callable
 
@@ -9,6 +10,8 @@ from typing import TYPE_CHECKING, Any, Callable
 if TYPE_CHECKING:
     from .context import FlowContext
 
+logger = logging.getLogger(__name__)
+
 
 @dataclass
 class ToolResult:
@@ -52,6 +55,7 @@ class Tool:
 
             return ToolResult(name=self.name, arguments=inputs or {}, result=result, success=True)
         except Exception as e:
+            logger.warning("Tool %s failed: %s", self.name, e, exc_info=True)
             return ToolResult(
                 name=self.name,
                 arguments=inputs or {},

{agstack-1.8.3 → agstack-1.9.0}/agstack.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agstack
-Version: 1.8.3
+Version: 1.9.0
 Summary: Production-ready toolkit for building FastAPI and LLM applications
 Author-email: XtraVisions <gitadmin@xtravisions.com>, Chen Hao <chenhao@xtravisions.com>
 Maintainer-email: XtraVisions <gitadmin@xtravisions.com>, Chen Hao <chenhao@xtravisions.com>
@@ -34,8 +34,7 @@ Requires-Dist: pydantic>=2.12.4
 Requires-Dist: python-multipart>=0.0.20
 Requires-Dist: requests>=2.32.5
 Requires-Dist: RestrictedPython>=7.0
-Requires-Dist: sqlalchemy[asyncio]>=2.0.48
-Requires-Dist: sqlobjects>=1.6.0
+Requires-Dist: sqlobjects>=1.9.0
 Requires-Dist: tiktoken>=0.12.0
 Requires-Dist: uvicorn>=0.41.0
 Dynamic: license-file

{agstack-1.8.3 → agstack-1.9.0}/agstack.egg-info/requires.txt

@@ -12,7 +12,6 @@ pydantic>=2.12.4
 python-multipart>=0.0.20
 requests>=2.32.5
 RestrictedPython>=7.0
-sqlalchemy[asyncio]>=2.0.48
-sqlobjects>=1.6.0
+sqlobjects>=1.9.0
 tiktoken>=0.12.0
 uvicorn>=0.41.0

{agstack-1.8.3 → agstack-1.9.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agstack"
-version = "1.8.3"
+version = "1.9.0"
 description = "Production-ready toolkit for building FastAPI and LLM applications"
 readme = "README.md"
 license = "MIT"
@@ -53,8 +53,7 @@ dependencies = [
     "python-multipart>=0.0.20",
     "requests>=2.32.5",
     "RestrictedPython>=7.0",
-    "sqlalchemy[asyncio]>=2.0.48",
-    "sqlobjects>=1.6.0",
+    "sqlobjects>=1.9.0",
     "tiktoken>=0.12.0",
     "uvicorn>=0.41.0",
 ]

{agstack-1.8.3 → agstack-1.9.0}/tests/test_flow_io.py

@@ -257,6 +257,76 @@ class TestDetectNodeHandler:
         assert isinstance(result, dict)
         assert result == {"choice": "qa"}
 
+    @patch("agstack.llm.flow.nodes.detect_node.get_llm_client")
+    def test_dynamic_instruction_and_options(self, mock_get_client):
+        from agstack.llm.flow.nodes.detect_node import DetectNodeHandler
+
+        mock_response = MagicMock()
+        mock_choice = MagicMock()
+        mock_choice.message.content = '{"result": "billing"}'
+        mock_response.choices = [mock_choice]
+        mock_response.usage = MagicMock(prompt_tokens=10, completion_tokens=5, total_tokens=15)
+
+        mock_client = AsyncMock()
+        mock_client.chat = AsyncMock(return_value=mock_response)
+        mock_get_client.return_value = mock_client
+
+        handler = DetectNodeHandler()
+        ctx = FlowContext(
+            variables={
+                "my_instruction": "classify ticket type",
+                "my_options": ["billing", "technical", "general"],
+            }
+        )
+        node = {
+            "id": "detect1",
+            "type": "detect",
+            "config": {
+                "inputs": {
+                    "query": "$v.user_query",
+                    "instruction": "$v.my_instruction",
+                    "options": "$v.my_options",
+                },
+            },
+        }
+        ctx.variables["user_query"] = "I was charged twice"
+        result = asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        assert result == {"choice": "billing"}
+
+    @patch("agstack.llm.flow.nodes.detect_node.get_llm_client")
+    def test_dynamic_model_and_temperature(self, mock_get_client):
+        from agstack.llm.flow.nodes.detect_node import DetectNodeHandler
+
+        mock_response = MagicMock()
+        mock_choice = MagicMock()
+        mock_choice.message.content = '{"result": "qa"}'
+        mock_response.choices = [mock_choice]
+        mock_response.usage = MagicMock(prompt_tokens=10, completion_tokens=5, total_tokens=15)
+
+        mock_client = AsyncMock()
+        mock_client.chat = AsyncMock(return_value=mock_response)
+        mock_get_client.return_value = mock_client
+
+        handler = DetectNodeHandler()
+        ctx = FlowContext(variables={"chosen_model": "qwen2.5-72b", "temp": 0.1})
+        node = {
+            "id": "detect2",
+            "type": "detect",
+            "config": {
+                "options": ["qa", "chitchat"],
+                "inputs": {
+                    "query": "hello",
+                    "model": "$v.chosen_model",
+                    "temperature": "$v.temp",
+                },
+            },
+        }
+        result = asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.chat.call_args
+        assert call_args.kwargs["model"] == "qwen2.5-72b"
+        assert call_args.kwargs["temperature"] == 0.1
+        assert result == {"choice": "qa"}
+
 
 # ── LLMChatNodeHandler ──
 
@@ -344,6 +414,47 @@ class TestLLMChatNodeHandler:
         assert len(system_msg) == 1
         assert system_msg[0]["content"] == "You speak Chinese"
 
+    @patch("agstack.llm.flow.nodes.llm_chat_node.get_llm_client")
+    def test_dynamic_model_temperature_max_tokens(self, mock_get_client):
+        from agstack.llm.flow.nodes.llm_chat_node import LLMChatNodeHandler
+
+        mock_response = MagicMock()
+        mock_choice = MagicMock()
+        mock_choice.message.content = "response"
+        mock_response.choices = [mock_choice]
+        mock_response.usage = MagicMock(prompt_tokens=5, completion_tokens=3, total_tokens=8)
+
+        mock_client = AsyncMock()
+        mock_client.chat = AsyncMock(return_value=mock_response)
+        mock_get_client.return_value = mock_client
+
+        handler = LLMChatNodeHandler()
+        ctx = FlowContext(
+            variables={
+                "chosen_model": "qwen2.5-72b",
+                "temp": 0.2,
+                "max_tok": 512,
+            }
+        )
+        node = {
+            "id": "chat1",
+            "type": "llm_chat",
+            "config": {
+                "prompt": "Hello",
+                "model": "gpt-4o",
+                "inputs": {
+                    "model": "$v.chosen_model",
+                    "temperature": "$v.temp",
+                    "max_tokens": "$v.max_tok",
+                },
+            },
+        }
+        asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.chat.call_args
+        assert call_args.kwargs["model"] == "qwen2.5-72b"
+        assert call_args.kwargs["temperature"] == 0.2
+        assert call_args.kwargs["max_tokens"] == 512
+
 
 # ── Flow routing ──
 
@@ -433,3 +544,117 @@ class TestDataFlowIntegration:
         }
         result = asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
         assert result == {"result": 30}
+
+
+class TestLLMRerankNodeHandler:
+    """LLM Rerank 节点动态参数测试"""
+
+    @patch("agstack.llm.flow.nodes.llm_rerank_node.get_llm_client")
+    def test_dynamic_model_and_top_n(self, mock_get_client):
+        from agstack.llm.flow.nodes.llm_rerank_node import LLMRerankNodeHandler
+
+        mock_client = AsyncMock()
+        mock_client.rerank = AsyncMock(return_value=[(0, 0.95, "doc A"), (1, 0.80, "doc B")])
+        mock_get_client.return_value = mock_client
+
+        handler = LLMRerankNodeHandler()
+        ctx = FlowContext(variables={"rerank_model": "bge-reranker-large", "topk": 2})
+        node = {
+            "id": "rerank1",
+            "type": "llm_rerank",
+            "config": {
+                "model": "bge-reranker-v2-m3",
+                "inputs": {
+                    "query": "best python book",
+                    "documents": ["doc A", "doc B", "doc C"],
+                    "model": "$v.rerank_model",
+                    "top_n": "$v.topk",
+                },
+            },
+        }
+        result = asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.rerank.call_args
+        assert call_args.kwargs["model"] == "bge-reranker-large"
+        assert call_args.kwargs["top_n"] == 2
+        assert result == {
+            "results": [{"index": 0, "score": 0.95, "text": "doc A"}, {"index": 1, "score": 0.80, "text": "doc B"}]
+        }
+
+    @patch("agstack.llm.flow.nodes.llm_rerank_node.get_llm_client")
+    def test_static_fallback_still_works(self, mock_get_client):
+        from agstack.llm.flow.nodes.llm_rerank_node import LLMRerankNodeHandler
+
+        mock_client = AsyncMock()
+        mock_client.rerank = AsyncMock(return_value=[(0, 0.9, "doc A")])
+        mock_get_client.return_value = mock_client
+
+        handler = LLMRerankNodeHandler()
+        ctx = FlowContext()
+        node = {
+            "id": "rerank2",
+            "type": "llm_rerank",
+            "config": {
+                "model": "bge-reranker-v2-m3",
+                "top_n": 5,
+                "inputs": {
+                    "query": "test",
+                    "documents": ["doc A"],
+                },
+            },
+        }
+        asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.rerank.call_args
+        assert call_args.kwargs["model"] == "bge-reranker-v2-m3"
+        assert call_args.kwargs["top_n"] == 5
+
+
+class TestLLMEmbedNodeHandler:
+    """LLM Embed 节点动态参数测试"""
+
+    @patch("agstack.llm.flow.nodes.llm_embed_node.get_llm_client")
+    def test_dynamic_model(self, mock_get_client):
+        from agstack.llm.flow.nodes.llm_embed_node import LLMEmbedNodeHandler
+
+        mock_client = AsyncMock()
+        mock_client.embed = AsyncMock(return_value=[[0.1, 0.2, 0.3]])
+        mock_get_client.return_value = mock_client
+
+        handler = LLMEmbedNodeHandler()
+        ctx = FlowContext(variables={"embed_model": "text-embedding-3-large"})
+        node = {
+            "id": "embed1",
+            "type": "llm_embed",
+            "config": {
+                "model": "bge-m3",
+                "inputs": {
+                    "texts": ["hello world"],
+                    "model": "$v.embed_model",
+                },
+            },
+        }
+        result = asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.embed.call_args
+        assert call_args.kwargs["model"] == "text-embedding-3-large"
+        assert result == {"embeddings": [[0.1, 0.2, 0.3]]}
+
+    @patch("agstack.llm.flow.nodes.llm_embed_node.get_llm_client")
+    def test_static_model_fallback(self, mock_get_client):
+        from agstack.llm.flow.nodes.llm_embed_node import LLMEmbedNodeHandler
+
+        mock_client = AsyncMock()
+        mock_client.embed = AsyncMock(return_value=[[0.1, 0.2]])
+        mock_get_client.return_value = mock_client
+
+        handler = LLMEmbedNodeHandler()
+        ctx = FlowContext()
+        node = {
+            "id": "embed2",
+            "type": "llm_embed",
+            "config": {
+                "model": "bge-m3",
+                "inputs": {"texts": ["hello"]},
+            },
+        }
+        asyncio.get_event_loop().run_until_complete(handler.execute(node, ctx))
+        call_args = mock_client.embed.call_args
+        assert call_args.kwargs["model"] == "bge-m3"