langroid 0.8.0__py3-none-any.whl → 0.9.0__py3-none-any.whl

langroid/agent/typed_task.py

@@ -0,0 +1,19 @@
+# from langroid.agent.tool_message import ToolMessage
+# from langroid.agent.task import Task
+# from typing import Any
+#
+# class TypedTask:
+#     def __init__(self, Task: Task, input_type: Any, output_type: Any):
+#         self.Task = Task
+#         self.input_type = input_type
+#         self.output_type = output_type
+#
+#     def run(self, input: Any) -> Any:
+#         if not isinstance(input, self.input_type):
+#             raise ValueError(f"Input must be of type {self.input_type}")
+#         output = self.Task.run(input)
+#         if not isinstance(output, self.output_type):
+#             raise ValueError(f"Output must be of type {self.output_type}")
+#         return output
+#
+#

langroid/language_models/base.py

@@ -294,8 +294,9 @@ class LLMResponse(BaseModel):
         of the recipient name if specified.
 
         Two cases:
-        (a) `message` contains "TO: <name> <content>", or
-        (b) `message` is empty and `function_call` with `to: <name>`
+        (a) `message` contains addressing string "TO: <name> <content>", or
+        (b) `message` is empty and function_call/tool_call with explicit `recipient`
+
 
         Returns:
             (str): name of recipient, which may be empty string if no recipient

langroid/mytypes.py

@@ -76,7 +76,6 @@ class Document(BaseModel):
         return self.metadata.id
 
     def __str__(self) -> str:
-        # TODO: make metadata a pydantic model to enforce "source"
         return dedent(
             f"""
         CONTENT: {self.content}         

langroid/parsing/parse_json.py

@@ -89,11 +89,28 @@ def parse_imperfect_json(json_string: str) -> Union[Dict[str, Any], List[Any]]:
 
     # If ast.literal_eval fails or returns non-dict/list, try json.loads
     try:
-        str = add_quotes(json_string)
-        result = json.loads(str)
+        json_string = add_quotes(json_string)
+        result = json.loads(json_string)
         if isinstance(result, (dict, list)):
             return result
     except json.JSONDecodeError:
+        try:
+            # fallback on yaml
+            yaml_result = yaml.safe_load(json_string)
+            if isinstance(yaml_result, (dict, list)):
+                return yaml_result
+        except yaml.YAMLError:
+            pass
+
+    try:
+        # last resort: try to repair the json using a lib
+        from json_repair import repair_json
+
+        repaired_json = repair_json(json_string)
+        result = json.loads(repaired_json)
+        if isinstance(result, (dict, list)):
+            return result
+    except Exception:
         pass
 
     # If all methods fail, raise ValueError

langroid/utils/pydantic_utils.py

@@ -1,4 +1,5 @@
 import logging
+from collections.abc import MutableMapping
 from contextlib import contextmanager
 from typing import (
     Any,
@@ -21,6 +22,24 @@ from langroid.pydantic_v1 import BaseModel, ValidationError, create_model
 logger = logging.getLogger(__name__)
 
 
+def flatten_dict(
+    d: MutableMapping[str, Any], parent_key: str = "", sep: str = "."
+) -> Dict[str, Any]:
+    """Flatten a nested dictionary, using a separator in the keys.
+    Useful for pydantic_v1 models with nested fields -- first use
+        dct = mdl.model_dump()
+    to get a nested dictionary, then use this function to flatten it.
+    """
+    items: List[Tuple[str, Any]] = []
+    for k, v in d.items():
+        new_key = f"{parent_key}{sep}{k}" if parent_key else k
+        if isinstance(v, MutableMapping):
+            items.extend(flatten_dict(v, new_key, sep=sep).items())
+        else:
+            items.append((new_key, v))
+    return dict(items)
+
+
 def has_field(model_class: Type[BaseModel], field_name: str) -> bool:
     """Check if a Pydantic model class has a field with the given name."""
     return field_name in model_class.__fields__

langroid/vector_store/base.py

@@ -15,6 +15,7 @@ from langroid.utils.configuration import settings
 from langroid.utils.object_registry import ObjectRegistry
 from langroid.utils.output.printing import print_long_text
 from langroid.utils.pandas_utils import stringify
+from langroid.utils.pydantic_utils import flatten_dict
 
 logger = logging.getLogger(__name__)
 
@@ -136,7 +137,8 @@ class VectorStore(ABC):
         """Compute a result on a set of documents,
         using a dataframe calc string like `df.groupby('state')['income'].mean()`.
         """
-        dicts = [doc.dict() for doc in docs]
+        # convert each doc to a dict, using dotted paths for nested fields
+        dicts = [flatten_dict(doc.dict(by_alias=True)) for doc in docs]
         df = pd.DataFrame(dicts)
 
         try:

langroid/vector_store/lancedb.py

@@ -147,6 +147,8 @@ class LanceDB(VectorStore):
 
     def _create_lance_schema(self, doc_cls: Type[Document]) -> Type[BaseModel]:
         """
+        NOTE: NOT USED, but leaving it here as it may be useful.
+
         Create a subclass of LanceModel with fields:
          - id (str)
          - Vector field that has dims equal to

{langroid-0.8.0.dist-info → langroid-0.9.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.8.0
+Version: 0.9.0
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani
@@ -54,6 +54,7 @@ Requires-Dist: grpcio (>=1.62.1,<2.0.0)
 Requires-Dist: halo (>=0.0.31,<0.0.32)
 Requires-Dist: huggingface-hub (>=0.21.2,<0.22.0) ; extra == "hf-transformers" or extra == "all" or extra == "transformers"
 Requires-Dist: jinja2 (>=3.1.2,<4.0.0)
+Requires-Dist: json-repair (>=0.27.0,<0.28.0)
 Requires-Dist: lancedb (>=0.8.2,<0.9.0) ; extra == "vecdbs" or extra == "lancedb"
 Requires-Dist: litellm (>=1.30.1,<2.0.0) ; extra == "all" or extra == "litellm"
 Requires-Dist: lxml (>=4.9.3,<5.0.0)

{langroid-0.8.0.dist-info → langroid-0.9.0.dist-info}/RECORD

@@ -1,26 +1,22 @@
 langroid/__init__.py,sha256=z_fCOLQJPOw3LLRPBlFB5-2HyCjpPgQa4m4iY5Fvb8Y,1800
 langroid/agent/__init__.py,sha256=ll0Cubd2DZ-fsCMl7e10hf9ZjFGKzphfBco396IKITY,786
-langroid/agent/base.py,sha256=EOAdZ2K0Yjjdt3zkjNsUhHgRvYZT-eIgRbY4ldjt6hY,46995
+langroid/agent/base.py,sha256=dCGYbT2s35OEgbYtuDWOBhpeJzjo2TPnbbG4NrjVv70,58201
 langroid/agent/batch.py,sha256=feRA_yRG768ElOQjrKEefcRv6Aefd_yY7qktuYUQDwc,10040
 langroid/agent/callbacks/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/callbacks/chainlit.py,sha256=Qedk1-CBCgo9PdaIa7AboLBFCTgAMg9q5nGmoqpZ378,22050
-langroid/agent/chat_agent.py,sha256=4bgLHBFSxL1dw1UAEF5zfCqnp_ss0MUUTOnjKkfOJRQ,45401
-langroid/agent/chat_document.py,sha256=y3QBdpclTBuqs42c3lLO9HU4s70PyK133HXZxyaxNZ8,16322
+langroid/agent/chat_agent.py,sha256=VUuYweOLuH4lokvZLY74WXIG4AlxRU4L907ox_hoRA4,48107
+langroid/agent/chat_document.py,sha256=Cq-TbPefpxbx2QzGl-CUpIrCcEC34GHDGH6Fd0w7aYs,16693
 langroid/agent/helpers.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/junk,sha256=LxfuuW7Cijsg0szAzT81OjWWv1PMNI-6w_-DspVIO2s,339
-langroid/agent/openai_assistant.py,sha256=TlMi5QecrxxhZ7TW03wB161elSSQYwL7H0xk8g7HawU,33879
+langroid/agent/openai_assistant.py,sha256=2rjCZw45ysNBEGNzQM4uf0bTC4KkatGYAWcVcW4xcek,34337
 langroid/agent/special/__init__.py,sha256=gik_Xtm_zV7U9s30Mn8UX3Gyuy4jTjQe9zjiE3HWmEo,1273
 langroid/agent/special/doc_chat_agent.py,sha256=8NPAhMnHkFUolQ8EHos40tz5Vwuz_m33NjUfjheXWXY,54569
-langroid/agent/special/lance_doc_chat_agent.py,sha256=Hjpu6u9UPAFMg5J6K97PRFaLbNrGhInC0N9oGi09CeY,10006
+langroid/agent/special/lance_doc_chat_agent.py,sha256=s8xoRs0gGaFtDYFUSIRchsgDVbS5Q3C2b2mr3V1Fd-Q,10419
 langroid/agent/special/lance_rag/__init__.py,sha256=QTbs0IVE2ZgDg8JJy1zN97rUUg4uEPH7SLGctFNumk4,174
-langroid/agent/special/lance_rag/critic_agent.py,sha256=S3NA3OAO7XaXjCrmwhKB7qCPlgRZFvDxiB5Qra65Zhs,7959
+langroid/agent/special/lance_rag/critic_agent.py,sha256=lRHLBU9TWo4ixm_wEPUP9bCjW0iSIw7ZRe6u4D4v7jk,9519
 langroid/agent/special/lance_rag/lance_rag_task.py,sha256=l_HQgrYY-CX2FwIsS961aEF3bYog3GDYo98fj0C0mSk,2889
-langroid/agent/special/lance_rag/query_planner_agent.py,sha256=QB8UYITUCkgSPturEwu_3i4kU8jXxW_jXNGSLlH5tMc,10109
-langroid/agent/special/lance_rag_new/__init__.py,sha256=QTbs0IVE2ZgDg8JJy1zN97rUUg4uEPH7SLGctFNumk4,174
-langroid/agent/special/lance_rag_new/critic_agent.py,sha256=9TV0_ILQ1jNeO9Pjeirz3VsUsCKeImyVHPu-fJub6cY,8340
-langroid/agent/special/lance_rag_new/lance_rag_task.py,sha256=mKo4lCC1ff5Jt9BlxwclQ_y3omw2S_f0MVIJfNmXM6w,5267
-langroid/agent/special/lance_rag_new/query_planner_agent.py,sha256=JqO_5fKW8HPn-zqKsZzX1sl05RgtcT63qpDMR9-q33A,10251
-langroid/agent/special/lance_tools.py,sha256=BznV_r3LAFyybvBRa9KQ0oU7mPM3uQVfri7PFp7M_qc,1894
+langroid/agent/special/lance_rag/query_planner_agent.py,sha256=-ZKareApPQD0EbwB1ABeZdhYAA2lJbwSVkfSuhMcKMk,11480
+langroid/agent/special/lance_tools.py,sha256=qS8x4wi8mrqfbYV2ztFzrcxyhHQ0ZWOc-zkYiH7awj0,2105
 langroid/agent/special/neo4j/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/special/neo4j/csv_kg_chat.py,sha256=dRsAgMBa1H_EMI2YYgJR2Xyv1D7e4o3G9M64mTewq_c,6409
 langroid/agent/special/neo4j/neo4j_chat_agent.py,sha256=dvZ4rrMWwKuOIOjC3iWQIe-cSxNXkaU_6W0rQII_4Q8,13347
@@ -36,21 +32,22 @@ langroid/agent/special/sql/utils/populate_metadata.py,sha256=1J22UsyEPKzwK0XlJZt
 langroid/agent/special/sql/utils/system_message.py,sha256=qKLHkvQWRQodTtPLPxr1GSLUYUFASZU8x-ybV67cB68,1885
 langroid/agent/special/sql/utils/tools.py,sha256=vFYysk6Vi7HJjII8B4RitA3pt_z3gkSglDNdhNVMiFc,1332
 langroid/agent/special/table_chat_agent.py,sha256=d9v2wsblaRx7oMnKhLV7uO_ujvk9gh59pSGvBXyeyNc,9659
-langroid/agent/task.py,sha256=eCA91yr6DSaRRGteTbSki-liMl6nspzRd_5Gin_ZFYw,75176
-langroid/agent/team.py,sha256=88VNRSmK35WEl620GfBzuIrBASXYSeBZ8yDKX-nP_Bo,75778
-langroid/agent/tool_message.py,sha256=XPM6whazpUOQu_vM2D4Q_L9tV6Vs9OTXNrGchwcqD_o,9768
-langroid/agent/tools/__init__.py,sha256=e-63cfwQNk_ftRKQwgDAJQK16QLbRVWDBILeXIc7wLk,402
+langroid/agent/task.py,sha256=taSP80E_pcqa6OEGi9weOQXx7It8SZvgosfuVv4DQcw,80300
+langroid/agent/tool_message.py,sha256=pArap87HNODwBVH2Dyu4Oi_OOhK0NH65pDUOHvIKV0E,10101
+langroid/agent/tools/__init__.py,sha256=SL7oZ0wOO8PzViGhWzjqY2OipcxH3sGfVli2XJ5bpWg,539
 langroid/agent/tools/duckduckgo_search_tool.py,sha256=NhsCaGZkdv28nja7yveAhSK_w6l_Ftym8agbrdzqgfo,1935
 langroid/agent/tools/extract_tool.py,sha256=u5lL9rKBzaLBOrRyLnTAZ97pQ1uxyLP39XsWMnpaZpw,3789
 langroid/agent/tools/generator_tool.py,sha256=y0fB0ZObjA0b3L0uSTtrqRCKHDUR95arBftqiUeKD2o,663
 langroid/agent/tools/google_search_tool.py,sha256=y7b-3FtgXf0lfF4AYxrZ3K5pH2dhidvibUOAGBE--WI,1456
 langroid/agent/tools/metaphor_search_tool.py,sha256=qj4gt453cLEX3EGW7nVzVu6X7LCdrwjSlcNY0qJW104,2489
 langroid/agent/tools/note_tool.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-langroid/agent/tools/recipient_tool.py,sha256=NrLxIeQT-kbMv7AeYX0uqvGeMK4Q3fIDvG15OVzlgk8,9624
+langroid/agent/tools/orchestration.py,sha256=MN-y5erSiCnsPtI7aeD7rn5K3kXzeQnJHRlCkenkloY,7324
+langroid/agent/tools/recipient_tool.py,sha256=0m2kQhYKTeGujAxhSPqH5z6hSAhVB_Dqour6uul2U30,9427
 langroid/agent/tools/retrieval_tool.py,sha256=2q2pfoYbZNfbWQ0McxrtmfF0ekGglIgRl-6uF26pa-E,871
 langroid/agent/tools/rewind_tool.py,sha256=XAXL3BpNhCmBGYq_qi_sZfHJuIw7NY2jp4wnojJ7WRs,5606
 langroid/agent/tools/run_python_code.py,sha256=BvoxYzzHijU-p4703n2iVlt5BCieR1oMSy50w0tQZAg,1787
 langroid/agent/tools/segment_extract_tool.py,sha256=__srZ_VGYLVOdPrITUM8S0HpmX4q7r5FHWMDdHdEv8w,1440
+langroid/agent/typed_task.py,sha256=oxja0Z3uLTv0BcR1xIMqDpo85MIGOruz4XsZ4ghjsW4,689
 langroid/agent_config.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/cachedb/__init__.py,sha256=icAT2s7Vhf-ZGUeqpDQGNU6ob6o0aFEyjwcxxUGRFjg,225
 langroid/cachedb/base.py,sha256=ztVjB1DtN6pLCujCWnR6xruHxwVj3XkYniRTYAKKqk0,1354
@@ -71,7 +68,7 @@ langroid/language_models/.chainlit/config.toml,sha256=1t5lHORGzc2E6dkaO9P15jYHu2
 langroid/language_models/.chainlit/translations/en-US.json,sha256=DAFz2HjOFFfboCStrUfKFg2BpplJPK_OOtixwF_GivY,9931
 langroid/language_models/__init__.py,sha256=1sUGobooTqq77XC7LxKsvME0RgSd5GGmeyrPo9SMh4U,940
 langroid/language_models/azure_openai.py,sha256=G4le3j4YLHV7IwgB2C37hO3MKijZ1KjynbYlEvpIF7Y,6214
-langroid/language_models/base.py,sha256=L76syCytos5IaQ5tpgToYJHThch4gA9uyjeMmjWjZ8E,21757
+langroid/language_models/base.py,sha256=ytJ_0Jw5erbqrqLPp4JMCo_nIkwzUvBqoKUr8Sae9Qg,21792
 langroid/language_models/config.py,sha256=9Q8wk5a7RQr8LGMT_0WkpjY8S4ywK06SalVRjXlfCiI,378
 langroid/language_models/mock_lm.py,sha256=2Ka05SVGSUy096bsa2AyjaqC5jmcFoe7HycpdnICTIw,3031
 langroid/language_models/openai_gpt.py,sha256=Bv-FCva9q0oO85VSqNpaxEYI8zPwVXpHmmfV7O8QRhU,61325
@@ -80,7 +77,7 @@ langroid/language_models/prompt_formatter/base.py,sha256=eDS1sgRNZVnoajwV_ZIha6c
 langroid/language_models/prompt_formatter/hf_formatter.py,sha256=TFL6ppmeQWnzr6CKQzRZFYY810zE1mr8DZnhw6i85ok,5217
 langroid/language_models/prompt_formatter/llama2_formatter.py,sha256=YdcO88qyBeuMENVIVvVqSYuEpvYSTndUe_jd6hVTko4,2899
 langroid/language_models/utils.py,sha256=o6Zo2cnnvKrfSgF26knVQ1xkSxEoE7yN85296gNdVOw,4858
-langroid/mytypes.py,sha256=KRN_dBamplSl3SYekosT_Maj6ZA749LONypaqPVnmbI,2435
+langroid/mytypes.py,sha256=ptAFxEAtiwmIfUnGisNotTe8wT9LKBf22lOfPgZoQIY,2368
 langroid/parsing/__init__.py,sha256=ZgSAfgTC6VsTLFlRSWT-TwYco7SQeRMeZG-49MnKYGY,936
 langroid/parsing/agent_chats.py,sha256=sbZRV9ujdM5QXvvuHVjIi2ysYSYlap-uqfMMUKulrW0,1068
 langroid/parsing/code-parsing.md,sha256=--cyyNiSZSDlIwcjAV4-shKrSiRe2ytF3AdSoS_hD2g,3294
@@ -89,7 +86,7 @@ langroid/parsing/config.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/parsing/document_parser.py,sha256=WGnA5ADwMHliGJt6WW9rc4RiFXQcKU33b5zdPiGrtEY,24265
 langroid/parsing/image_text.py,sha256=sbLIQ5nHe2UnYUksBaQsmZGaX-X0qgEpPd7CEzi_z5M,910
 langroid/parsing/para_sentence_split.py,sha256=AJBzZojP3zpB-_IMiiHismhqcvkrVBQ3ZINoQyx_bE4,2000
-langroid/parsing/parse_json.py,sha256=Ivh6YcEOESAi9y8E55E7qhFqfncvnluyb6sZGWK0--U,5268
+langroid/parsing/parse_json.py,sha256=sKrYv9-IUqRFaTJA24_rmfjN1E7dQSrTBrtd1jYDE1s,5817
 langroid/parsing/parser.py,sha256=AgtmlVUvrkSG1l7-YZPX8rlldgXjh_HqXAMqpXkBxUo,11746
 langroid/parsing/repo_loader.py,sha256=3GjvPJS6Vf5L6gV2zOU8s-Tf1oq_fZm-IB_RL_7CTsY,29373
 langroid/parsing/routing.py,sha256=-FcnlqldzL4ZoxuDwXjQPNHgBe9F9-F4R6q7b_z9CvI,1232
@@ -124,20 +121,20 @@ langroid/utils/output/citations.py,sha256=PSY2cpti8W-ZGFMAgj1lYoEIZy0lsniLpCliMs
 langroid/utils/output/printing.py,sha256=yzPJZN-8_jyOJmI9N_oLwEDfjMwVgk3IDiwnZ4eK_AE,2962
 langroid/utils/output/status.py,sha256=rzbE7mDJcgNNvdtylCseQcPGCGghtJvVq3lB-OPJ49E,1049
 langroid/utils/pandas_utils.py,sha256=UctS986Jtl_MvU5rA7-GfrjEHXP7MNu8ePhepv0bTn0,755
-langroid/utils/pydantic_utils.py,sha256=X35qxjE4sSIi-oBMkI1s9fiUIJbpXHLmJqcJ7zsy0jg,19914
+langroid/utils/pydantic_utils.py,sha256=iRy7uQhHhQmIDZTTPNX5jXb6fqefMe9N67p3fPfOmTI,20624
 langroid/utils/system.py,sha256=nvKeeUAj4eviR4kYpcr9h-HYdhqUNMTRBTHBOhz0GdU,5182
 langroid/utils/web/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/utils/web/login.py,sha256=1iz9eUAHa87vpKIkzwkmFa00avwFWivDSAr7QUhK7U0,2528
 langroid/vector_store/__init__.py,sha256=6xBjb_z4QtUy4vz4RuFbcbSwmHrggHL8-q0DwCf3PMM,972
-langroid/vector_store/base.py,sha256=pkc4n0yWGVk7iRUOLFkU_ID5NiBFfAcA3lBlPNX79pU,13623
+langroid/vector_store/base.py,sha256=BDsGl9D8TeepZK8-8SSIxpx9LJRJF9ic_xw4WjBX3sI,13780
 langroid/vector_store/chromadb.py,sha256=KMfHrgovQEOeJR_LsMpGM8BteJ50wpisDu608RhU3SU,7940
-langroid/vector_store/lancedb.py,sha256=MLubJBhtNIFX6zY0qANqCoB6MlL-oZiJCg9gZp2H2rs,14620
+langroid/vector_store/lancedb.py,sha256=kWIQ9QxFKS8gJ7cSn11empvi7Q6_NvXNkc4f-CmVZMg,14686
 langroid/vector_store/meilisearch.py,sha256=6frB7GFWeWmeKzRfLZIvzRjllniZ1cYj3HmhHQICXLs,11663
 langroid/vector_store/momento.py,sha256=qR-zBF1RKVHQZPZQYW_7g-XpTwr46p8HJuYPCkfJbM4,10534
 langroid/vector_store/qdrant_cloud.py,sha256=3im4Mip0QXLkR6wiqVsjV1QvhSElfxdFSuDKddBDQ-4,188
 langroid/vector_store/qdrantdb.py,sha256=v88lqFkepADvlN6lByUj9I4NEKa9X9lWH16uTPPbYrE,17457
-pyproject.toml,sha256=pK1C4vWrb5-NkREBDkFao-3Pi-xXJXMjh_wjedpjFAk,7063
-langroid-0.8.0.dist-info/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.8.0.dist-info/METADATA,sha256=i2_aKdjkgYcGAYMXEG2F3NC59yVI1w3uidn3edm5wAY,54518
-langroid-0.8.0.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
-langroid-0.8.0.dist-info/RECORD,,
+pyproject.toml,sha256=d0X6zblgHm8pq9AAikR8iHVrp0vXpoxFc6kzGIU0DPg,7087
+langroid-0.9.0.dist-info/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.9.0.dist-info/METADATA,sha256=5s4JX52hFBNWyWetUOw7mPC5sjo8Zj7CNxwoXMHyHzs,54564
+langroid-0.9.0.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
+langroid-0.9.0.dist-info/RECORD,,

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.8.0"
+version = "0.9.0"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"
@@ -88,6 +88,7 @@ nest-asyncio = "^1.6.0"
 async-generator = "^1.10"
 
 python-magic = "^0.4.27"
+json-repair = "^0.27.0"
 
 [tool.poetry.extras]
 # install these using, e.g.,

langroid/agent/special/lance_rag_new/__init__.py

@@ -1,9 +0,0 @@
-from . import query_planner_agent
-from . import critic_agent
-from . import lance_rag_task
-
-__all__ = [
-    "query_planner_agent",
-    "critic_agent",
-    "lance_rag_task",
-]

langroid/agent/special/lance_rag_new/critic_agent.py

@@ -1,171 +0,0 @@
-"""
-QueryPlanCritic is a ChatAgent that is created with a specific document schema.
-
-Its role is to provide feedback on a Query Plan, which consists of:
-- filter condition if needed (or empty string if no filter is needed)
-- query - a possibly rephrased query that can be used to match the `content` field
-- dataframe_calc - a Pandas-dataframe calculation/aggregation string, possibly empty
-- original_query - the original query for reference
-- result - the answer received from an assistant that used this QUERY PLAN.
-
-This agent has access to two tools:
-- QueryPlanTool: The handler method for this tool re-writes the query plan
-  in plain text (non-JSON) so the LLM can provide its feedback using the
-  QueryPlanFeedbackTool.
-- QueryPlanFeedbackTool: LLM uses this tool to provide feedback on the Query Plan
-"""
-
-import logging
-
-from langroid.agent.chat_agent import ChatAgent
-from langroid.agent.chat_document import ChatDocument
-from langroid.agent.special.lance_rag.query_planner_agent import (
-    LanceQueryPlanAgentConfig,
-)
-from langroid.agent.special.lance_tools import (
-    QueryPlanAnswerTool,
-    QueryPlanFeedbackTool,
-)
-from langroid.mytypes import Entity
-from langroid.utils.constants import DONE, NO_ANSWER
-
-logger = logging.getLogger(__name__)
-
-
-class QueryPlanCriticConfig(LanceQueryPlanAgentConfig):
-    name = "QueryPlanCritic"
-    system_message = f"""
-    You are an expert at carefully planning a query that needs to be answered
-    based on a large collection of documents. These docs have a special `content` field
-    and additional FILTERABLE fields in the SCHEMA below, along with the 
-    SAMPLE VALUES for each field, and the DTYPE in PANDAS TERMINOLOGY.
-    
-    {{doc_schema}}
-    
-    The ORIGINAL QUERY is handled by a QUERY PLANNER who sends the PLAN to an ASSISTANT,
-    who returns an ANSWER.
-    
-    You will receive a QUERY PLAN consisting of:
-    - ORIGINAL QUERY from the user, which a QUERY PLANNER processes,
-      to create a QUERY PLAN, to be handled by an ASSISTANT.
-    - PANDAS-LIKE FILTER, WHICH CAN BE EMPTY (and it's fine if results sound reasonable)
-      FILTER SHOULD ONLY BE USED IF EXPLICITLY REQUIRED BY THE QUERY.
-    - REPHRASED QUERY (CANNOT BE EMPTY) that will be used to match against the 
-      CONTENT (not filterable) of the documents.
-      In general the REPHRASED QUERY should be relied upon to match the CONTENT 
-      of the docs. Thus the REPHRASED QUERY itself acts like a 
-      SEMANTIC/LEXICAL/FUZZY FILTER since the Assistant is able to use it to match 
-      the CONTENT of the docs in various ways (semantic, lexical, fuzzy, etc.). 
-        Keep in mind that the ASSISTANT does NOT know anything about the FILTER fields,
-        so the REPHRASED QUERY should NOT mention ANY FILTER fields.
-        The assistant will answer based on documents whose CONTENTS match the QUERY, 
-        possibly REPHRASED. 
-        !!!!****THE REPHRASED QUERY SHOULD NEVER BE EMPTY****!!!
-    - DATAFRAME CALCULATION, which must be a SINGLE LINE calculation (or empty),
-        [NOTE ==> This calculation is applied AFTER the FILTER and REPHRASED QUERY.],
-    - ANSWER received from an assistant that used this QUERY PLAN.
-      NOTE --the ANSWER will usually NOT contain any references to FILTERING conditions,
-      and this is ALLOWED, since the ANSWER is based on documents AFTER FILTERING.
-
-    In addition to the above SCHEMA fields there is a `content` field which:
-    - CANNOT appear in a FILTER, 
-    - CAN appear in the DATAFRAME CALCULATION.
-    THERE ARE NO OTHER FIELDS IN THE DOCUMENTS or in the RESULTING DATAFRAME.
-        
-    Your job is to act as a CRITIC and provide feedback, 
-    ONLY using the `query_plan_feedback` tool, and DO NOT SAY ANYTHING ELSE.
-    
-    Here is how you must examine the QUERY PLAN + ANSWER:
-    - ALL filtering conditions in the original query must be EXPLICITLY 
-      mentioned in the FILTER, and the QUERY field should not be used for filtering.
-    - If the ANSWER contains an ERROR message, then this means that the query
-      plan execution FAILED, and your feedback should say INVALID along 
-      with the ERROR message, `suggested_fix` that aims to help the assistant 
-      fix the problem (or simply equals "address the the error shown in feedback")
-    - Ask yourself, is the ANSWER in the expected form, e.g. 
-        if the question is asking for the name of an ENTITY with max SIZE,
-        then the answer should be the ENTITY name, NOT the SIZE!! 
-    - It is perfectly FINE if the ANSWER does NOT contain any references to FILTERING 
-      conditions, since the ANSWER is obtained from documents AFTER FILTERING!
-    - If the ANSWER is in the expected form, then the QUERY PLAN is likely VALID,
-      and your feedback should say VALID, with empty `suggested_fix`.
-      ===> HOWEVER!!! Watch out for a spurious correct-looking answer, for EXAMPLE:
-      the query was to find the ENTITY with a maximum SIZE, 
-      but the dataframe calculation is find the SIZE, NOT the ENTITY!!      
-    - If the ANSWER is {NO_ANSWER} or of the wrong form, 
-      then try to DIAGNOSE the problem IN THE FOLLOWING ORDER:
-      - DATAFRAME CALCULATION -- is it doing the right thing?
-        Is it finding the Index of a row instead of the value in a column?
-        Or another example: maybe it is finding the maximum population
-           rather than the CITY with the maximum population?
-        If you notice a problem with the DATAFRAME CALCULATION, then
-        ONLY SUBMIT FEEDBACK ON THE DATAFRAME CALCULATION, and DO NOT
-        SUGGEST ANYTHING ELSE.
-      - If the DATAFRAME CALCULATION looks correct, then check if 
-        the REPHRASED QUERY makes sense given the ORIGINAL QUERY and FILTER.
-        If this is the problem, then ONLY SUBMIT FEEDBACK ON THE REPHRASED QUERY,
-        and DO NOT SUGGEST ANYTHING ELSE.
-      - If the REPHRASED QUERY looks correct, then check if the FILTER makes sense.
-        REMEMBER: A filter should ONLY be used if EXPLICITLY REQUIRED BY THE QUERY.
-     
-     
-     IMPORTANT!! The DATAFRAME CALCULATION is done AFTER applying the 
-         FILTER and REPHRASED QUERY! Keep this in mind when evaluating 
-         the correctness of the DATAFRAME CALCULATION.
-    
-    ALWAYS use `query_plan_feedback` tool/fn to present your feedback
-    in the `feedback` field, and if any fix is suggested,
-    present it in the `suggested_fix` field.
-    DO NOT SAY ANYTHING ELSE OUTSIDE THE TOOL/FN.
-    IF NO REVISION NEEDED, simply leave the `suggested_fix` field EMPTY,
-    and SAY NOTHING ELSE
-    and DO NOT EXPLAIN YOURSELF.        
-    """
-
-
-def plain_text_query_plan(msg: QueryPlanAnswerTool) -> str:
-    plan = f"""
-    OriginalQuery: {msg.plan.original_query}
-    Filter: {msg.plan.filter}
-    Rephrased Query: {msg.plan.query}
-    DataframeCalc: {msg.plan.dataframe_calc}
-    Answer: {msg.answer}
-    """
-    return plan
-
-
-class QueryPlanCritic(ChatAgent):
-    """
-    Critic for LanceQueryPlanAgent, provides feedback on
-    query plan + answer.
-    """
-
-    def __init__(self, cfg: LanceQueryPlanAgentConfig):
-        super().__init__(cfg)
-        self.config = cfg
-        self.enable_message(QueryPlanAnswerTool, use=False, handle=True)
-        self.enable_message(QueryPlanFeedbackTool, use=True, handle=True)
-
-    def query_plan_answer(self, msg: QueryPlanAnswerTool) -> str:
-        """Present query plan + answer in plain text (not JSON)
-        so LLM can give feedback"""
-        return plain_text_query_plan(msg)
-
-    def query_plan_feedback(self, msg: QueryPlanFeedbackTool) -> ChatDocument:
-        """Format Valid so return to Query Planner"""
-        doc = self.create_agent_response(DONE)
-        doc.tool_messages = [msg]
-        return doc
-
-    def handle_message_fallback(
-        self, msg: str | ChatDocument
-    ) -> str | ChatDocument | None:
-        """Remind the LLM to use QueryPlanFeedbackTool since it forgot"""
-        if isinstance(msg, ChatDocument) and msg.metadata.sender == Entity.LLM:
-            return """
-            You forgot to use the `query_plan_feedback` tool/function.
-            Re-try your response using the `query_plan_feedback` tool/function,
-            remember to provide feedback in the `feedback` field,
-            and if any fix is suggested, provide it in the `suggested_fix` field.
-            """
-        return None

langroid/agent/special/lance_rag_new/lance_rag_task.py

@@ -1,144 +0,0 @@
-"""
-The LanceRAGTaskCreator.new() method creates a 3-Agent system that uses this agent.
-It takes a LanceDocChatAgent instance as argument, and adds two more agents:
-- LanceQueryPlanAgent, which is given the LanceDB schema in LanceDocChatAgent,
-and based on this schema, for a given user query, creates a Query Plan
-using the QueryPlanTool, which contains a filter, a rephrased query,
-and a dataframe_calc.
-- QueryPlanCritic, which is given the LanceDB schema in LanceDocChatAgent,
- and gives feedback on the Query Plan and Result using the QueryPlanFeedbackTool.
-
-The LanceRAGTaskCreator.new() method sets up the given LanceDocChatAgent and
-QueryPlanCritic as sub-tasks of the LanceQueryPlanAgent's task.
-
-Langroid's built-in task orchestration ensures that:
-- the LanceQueryPlanAgent reformulates the plan based
-    on the QueryPlanCritics's feedback,
-- LLM deviations are corrected via tools and overrides of ChatAgent methods.
-"""
-
-import logging
-
-from langroid.agent.special.lance_tools import (
-    QueryPlanAnswerTool,
-)
-from langroid.agent.task import Task
-from langroid.mytypes import Entity
-from langroid.utils.constants import NO_ANSWER
-
-from ..lance_doc_chat_agent import LanceDocChatAgent
-from .critic_agent import (
-    QueryPlanCritic,
-    QueryPlanCriticConfig,
-)
-from .query_planner_agent import (
-    LanceQueryPlanAgent,
-    LanceQueryPlanAgentConfig,
-)
-
-logger = logging.getLogger(__name__)
-
-
-def run_lance_rag_task(
-    query: str,
-    agent: LanceDocChatAgent,
-    interactive: bool = True,
-) -> str:
-    """
-    Add a LanceFilterAgent to the LanceDocChatAgent,
-    set up the corresponding Tasks, connect them,
-    and return the top-level query_plan_task.
-    """
-    doc_agent_name = "LanceRAG"
-    critic_name = "QueryPlanCritic"
-    query_plan_agent_config = LanceQueryPlanAgentConfig(
-        critic_name=critic_name,
-        doc_agent_name=doc_agent_name,
-        doc_schema=agent._get_clean_vecdb_schema(),
-    )
-    query_plan_agent_config.set_system_message()
-
-    query_planner = LanceQueryPlanAgent(query_plan_agent_config)
-    query_plan_task = Task(
-        query_planner,
-        interactive=interactive,
-        restart=False,
-        done_if_response=[Entity.AGENT],
-    )
-    # TODO - figure out how to define the fns so we avoid re-creating
-    # agents in each invocation. Right now we are defining the fn
-    # inside this context, which may not be great.
-
-    rag_task = Task(
-        agent,
-        name="LanceRAG",
-        restart=True,  # default; no need to accumulate dialog
-        interactive=False,
-        done_if_response=[Entity.LLM],  # done when non-null response from LLM
-        done_if_no_response=[Entity.LLM],  # done when null response from LLM
-    )
-
-    critic_config = QueryPlanCriticConfig(
-        doc_schema=agent._get_clean_vecdb_schema(),
-    )
-    critic_config.set_system_message()
-
-    critic_agent = QueryPlanCritic(critic_config)
-    critic_task = Task(
-        critic_agent,
-        interactive=False,
-        restart=True,  # default; no need to accumulate dialog
-    )
-
-    no_answer = False
-    feedback = None
-    i = 0
-    while i := i + 1 < 5:
-        # query, feedback (QueryPlanFeedbackTool) => ChatDocument[QueryPlanTool]
-        if feedback is not None and feedback.suggested_fix != "":
-            prompt = f"""
-                A Critic has seen your Query Plan and the Answer, and has given the 
-                following feedback. Take it into account and re-generate your Query Plan
-                for the QUERY:
-                
-                QUERY: {query}
-                FEEDBACK: {feedback.feedback}
-                SUGGESTED FIX: {feedback.suggested_fix}
-                """
-        elif no_answer:
-            prompt = f"There was a {NO_ANSWER} response; try a different query plan"
-        else:
-            prompt = query
-
-        while True:
-            plan_doc = query_plan_task.run(prompt)
-            if len(plan_doc.tool_messages) > 0:
-                break
-            # forgot to use QueryPlanTool
-            prompt = """You forgot to use the `query_plan` tool/function. Try again."""
-
-        # TODO if plan_doc does NOT have a QueryPlan, remind the agent
-
-        # ChatDocument with QueryPlanTool => ChatDocument with answer
-        rag_answer_doc = rag_task.run(plan_doc)
-
-        if rag_answer_doc is None:
-            rag_answer_doc = rag_task.agent.create_llm_response(NO_ANSWER)
-        # QueryPlan, answer => QueryPlanAnswerTool
-        plan_answer_tool = QueryPlanAnswerTool(
-            plan=plan_doc.tool_messages[0].plan,
-            answer=rag_answer_doc.content,
-        )
-        # QueryPlanAnswerTool => ChatDocument[QueryPlanAnswerTool]
-        plan_answer_doc = agent.create_agent_response(tool_messages=[plan_answer_tool])
-
-        # ChatDocument[QueryPlanAnswerTool] => ChatDocument[QueryPlanFeedbackTool]
-        feedback_doc = critic_task.run(plan_answer_doc)
-        # ChatDocument[QueryPlanFeedbackTool] => QueryPlanFeedbackTool
-        feedback = feedback_doc.tool_messages[0]  # QueryPlanFeedbackTool
-        no_answer = NO_ANSWER in rag_answer_doc.content
-        if feedback.suggested_fix == "" and not no_answer:
-            break
-
-    # query_plan_task.add_sub_task([critic_task, rag_task])
-    return rag_answer_doc.content