nucliadb-agentic-api 1.0.0.post216__tar.gz → 1.0.0.post226__tar.gz

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nucliadb_agentic_api
-Version: 1.0.0.post216
+Version: 1.0.0.post226
 Summary: NucliaDB Agentic API for Hyperforge
 Author: AI Data Team
 Author-email: AI Data Team <learning@nuclia.com>

nucliadb_agentic_api-1.0.0.post226/VERSION

@@ -0,0 +1 @@
+1.0.0.post226

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "uv_build"
 
 [project]
 name = "nucliadb_agentic_api"
-version = "1.0.0.post216"
+version = "1.0.0.post226"
 description = "NucliaDB Agentic API for Hyperforge"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -66,13 +66,13 @@ server = [
     "aiohttp<3.14.0",
     "backoff",
     "nats-py",
-    "nuclia>=4.11.5",
+    "nuclia>=4.14.0",
     "nuclia-models>=0.59.0",
     "nucliadb-protos",
     "nucliadb-sdk>=6.14.0.post6532",
     "nucliadb-telemetry[otel]>=6.14.0.post6534",
     "nucliadb-utils[storages,fastapi]>=6.15.0.post6671",
-    "hyperforge>=1.0.0.post88",
+    "hyperforge>=1.0.0.post237",
     "hyperforge_conditional>=1.0.0.post27",
     "hyperforge_mcp>=1.0.0.post41",
     "hyperforge_nucliadb>=1.0.0.post46",

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/pyproject.toml.orig

@@ -4,7 +4,7 @@ build-backend = "uv_build"
 
 [project]
 name = "nucliadb_agentic_api"
-version = "1.0.0.post216"
+version = "1.0.0.post226"
 authors = [{ name = "AI Data Team", email = "learning@nuclia.com" }]
 description = "NucliaDB Agentic API for Hyperforge"
 readme = "README.md"
@@ -61,13 +61,13 @@ server = [
     "aiohttp<3.14.0",
     "backoff",
     "nats-py",
-    "nuclia>=4.11.5",
+    "nuclia>=4.14.0",
     "nuclia-models>=0.59.0",
     "nucliadb-protos",
     "nucliadb-sdk>=6.14.0.post6532",
     "nucliadb-telemetry[otel]>=6.14.0.post6534",
     "nucliadb-utils[storages,fastapi]>=6.15.0.post6671",
-    "hyperforge>=1.0.0.post88",
+    "hyperforge>=1.0.0.post237",
     "hyperforge_conditional>=1.0.0.post27",
     "hyperforge_mcp>=1.0.0.post41",
     "hyperforge_nucliadb>=1.0.0.post46",

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/agentic/ask_handler.py

@@ -8,6 +8,7 @@ from hyperforge_nucliadb_agentic.ask.model import (
 from hyperforge_nucliadb_agentic.ask.search.ask import (
     handled_ask_exceptions,
 )
+from nuclia.lib.nua_responses import CitationsType
 from nucliadb_models.search import (
     NucliaDBClientType,
 )
@@ -34,6 +35,8 @@ async def create_agentic_response(
     extra_predict_headers: dict[str, str] | None = None,
 ) -> Response:
     ask_request.max_tokens = parse_max_tokens(ask_request.max_tokens)
+    if ask_request.citations is None:
+        ask_request.citations = CitationsType.LLM_FOOTNOTES
     ask_result = AgenticAskResult(
         app=app,
         kbid=kbid,

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/agentic/ask_result.py

@@ -27,8 +27,11 @@ from hyperforge_nucliadb_agentic.ask.model import (
     StatusAskResponseItem,
     TokensDetail,
 )
-from hyperforge_nucliadb_agentic.ask.predict import AnswerStatusCode
-from hyperforge_nucliadb_agentic.ask.search.ask import AskResult, RetrievalMatch
+from hyperforge_nucliadb_agentic.ask.search.ask import (
+    AnswerStatusCode,
+    AskResult,
+    RetrievalMatch,
+)
 from hyperforge_nucliadb_agentic.ask.search.metrics import (
     AskMetrics,
 )

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/app.py

@@ -8,23 +8,21 @@ from hyperforge.api.authentication import RaoAuthenticationBackend
 from hyperforge.broker import Broker
 from hyperforge.broker.redis import RedisBroker
 from hyperforge.driver import Driver
+from hyperforge.manager import Manager
 from hyperforge_nucliadb_agentic.ask.audit import (
     AuditMiddleware,
     start_audit_utility,
     stop_audit_utility,
 )
-from hyperforge_nucliadb_agentic.ask.predict import (
-    start_predict_engine,
-    stop_predict_engine,
-)
 from lru import LRU
 from mcp.server.lowlevel.server import Server as MCPServer
 from mcp.server.streamable_http import (
     StreamableHTTPServerTransport,
 )
+from nuclia.lib.nua import AsyncNuaClient
 from nucliadb_sdk import NucliaDBAsync
 from nucliadb_telemetry.utils import clean_telemetry, setup_telemetry
-from nucliadb_utils.settings import AuditSettings
+from nucliadb_utils.settings import AuditSettings, nuclia_settings
 from prometheus_client import CONTENT_TYPE_LATEST
 from starlette.middleware.authentication import AuthenticationMiddleware
 from starlette.responses import PlainTextResponse
@@ -61,6 +59,7 @@ class HTTPApplication(FastAPI):
     source_manager: Sources
     broker: Broker
     hyperforge_drivers: dict[str, "Driver"]
+    predict_manager: Manager
 
     def __init__(
         self,
@@ -105,9 +104,24 @@ class HTTPApplication(FastAPI):
     async def startup(self) -> None:
         await setup_telemetry(SERVICE_NAME)
 
-        await start_predict_engine()
         await start_audit_utility(SERVICE_NAME, self.audit_settings)
 
+        if nuclia_settings.onprem:
+            nua = AsyncNuaClient.onprem(
+                nuclia_settings.nuclia_public_url,
+                service_account=nuclia_settings.nuclia_service_account,
+                zone=nuclia_settings.nuclia_zone,
+                local_predict=nuclia_settings.local_predict,
+                local_predict_headers=nuclia_settings.local_predict_headers,
+            )
+        else:
+            nua = AsyncNuaClient.internal(nuclia_settings.nuclia_inner_predict_url)
+        self.predict_manager = await Manager.from_config(
+            drivers=[],
+            nua=nua,
+            send_rao_origin=False,
+        )
+
         self.broker = RedisBroker.from_url(
             url=self.settings.valkey_url,
             activate_subject=self.settings.activate_subject,
@@ -163,5 +177,5 @@ class HTTPApplication(FastAPI):
         await self.source_manager.finalize()
         await self.broker.finalize()
         await stop_audit_utility()
-        await stop_predict_engine()
+        await self.predict_manager.aclose()
         await clean_telemetry(SERVICE_NAME)

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/db/transform.py

@@ -55,7 +55,7 @@ async def transform_agentic_config(
         if agentic_config.rephrase.ask_to:
             # Same KB with different layers
             config["kbid"] = agentic_config.rephrase.ask_to
-        if agentic_config.rephrase.history:
+        if agentic_config.rephrase.history is not None:
             config["history"] = agentic_config.rephrase.history
 
         config["module"] = "rephrase"
@@ -78,10 +78,10 @@ async def transform_agentic_config(
                 config["planner_model"] = agentic_config.smart_agent.models.planner
             if agentic_config.smart_agent.models.executor:
                 config["executor_model"] = agentic_config.smart_agent.models.executor
-            if agentic_config.smart_agent.history:
-                config["history"] = agentic_config.smart_agent.history
             if agentic_config.smart_agent.extra_prompt:
                 config["extra_prompt"] = agentic_config.smart_agent.extra_prompt
+        if agentic_config.smart_agent.history is not None:
+            config["history"] = agentic_config.smart_agent.history
         registered_agents = []
         for source in agentic_config.smart_agent.sources:
             source_obj = await source_manager.get_source(account, kbid, source)
@@ -186,9 +186,10 @@ async def transform_agentic_config(
             config["user_prompt"] = agentic_config.summarize.user_prompt
         if agentic_config.summarize.system_prompt:
             config["system_prompt"] = agentic_config.summarize.system_prompt
-        if agentic_config.summarize.conversational:
-            config["conversational"] = agentic_config.summarize.conversational
-        if agentic_config.summarize.history:
+        config["conversational"] = agentic_config.summarize.conversational
+        config["citations"] = True
+        config["force_chunk_level_citations"] = False
+        if agentic_config.summarize.history is not None:
             config["history"] = agentic_config.summarize.history
 
         config["module"] = "summarize"

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/models.py

@@ -133,7 +133,7 @@ class AgenticRephraseConfiguration(BaseModel):
     ask_to: Optional[str] = None
     prompt: Optional[str] = None
     model: Optional[str] = None
-    history: Optional[bool] = None
+    history: Optional[bool] = True
 
 
 class AgenticSmartAgentMode(str, Enum):
@@ -218,15 +218,15 @@ class AgenticSmartAgentConfiguration(BaseModel):
     extra_prompt: Optional[str] = None
     models: Optional[AgenticSmartAgentModels] = None
     sources: List[str] = Field(default_factory=list)
-    history: Optional[bool] = None
+    history: Optional[bool] = True
 
 
 class AgenticSummarizeConfiguration(BaseModel):
     user_prompt: Optional[str] = None
     system_prompt: Optional[str] = None
-    conversational: bool = False
+    conversational: bool = True
     model: Optional[str] = None
-    history: Optional[bool] = None
+    history: Optional[bool] = True
 
 
 class AgenticConfigSchema(BaseModel):

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/server/server.py

@@ -7,7 +7,6 @@ from hyperforge.broker.redis import RedisBroker
 from hyperforge.feature_flag import get_flag_service
 from hyperforge.server.cache import ValkeyCache
 from hyperforge.server.run import run_metrics_server
-from hyperforge_nucliadb_agentic.ask.predict import start_predict_engine
 from nucliadb_telemetry.logs import setup_logging
 from nucliadb_telemetry.settings import LogFormatType, LogLevel, LogSettings
 from nucliadb_telemetry.utils import setup_telemetry
@@ -40,8 +39,6 @@ async def run_server(
     if tracer:
         await setup_telemetry(SERVICE_NAME)
 
-    await start_predict_engine()
-
     audit_settings: AuditSettings = AuditSettings()
     # Connect to Valkey
     broker = RedisBroker.from_url(

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/server/session.py

@@ -255,6 +255,12 @@ class NucliaDBAgenticSessionManager(SessionManager):
             errors.capture_exception(e)
             error = ARAGException(detail=str(e))
             observation.set_status("error")
+        finally:
+            try:
+                await state.manager.aclose()
+            except Exception as e:
+                logger.exception("Error closing Manager")
+                errors.capture_exception(e)
 
         observation.end()
         answer_running.dec()

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/v1/ask.py

@@ -2,6 +2,7 @@ import json
 from uuid import UUID
 
 from fastapi import Header, Request, Response
+from hyperforge.manager import Manager
 from hyperforge_nucliadb_agentic.ask.exceptions import (
     AnswerJsonSchemaTooLong,
 )
@@ -108,6 +109,7 @@ async def ask_knowledgebox_endpoint(
         )
 
     return await create_ask_response(
+        predict_manager=request.app.predict_manager,
         kbid=kbid,
         ask_request=item,
         user_id=x_nucliadb_user,
@@ -169,6 +171,7 @@ async def resource_ask_endpoint_by_uuid(
         )
 
     return await create_ask_response(
+        predict_manager=request.app.predict_manager,
         kbid=kbid,
         ask_request=item,
         user_id=x_nucliadb_user,
@@ -238,6 +241,7 @@ async def resource_ask_endpoint_by_slug(
         )
 
     return await create_ask_response(
+        predict_manager=request.app.predict_manager,
         kbid=kbid,
         ask_request=item,
         user_id=x_nucliadb_user,
@@ -251,6 +255,7 @@ async def resource_ask_endpoint_by_slug(
 
 @handled_ask_exceptions
 async def create_ask_response(
+    predict_manager: Manager,
     kbid: str,
     ask_request: AskRequest,
     user_id: str,
@@ -266,6 +271,7 @@ async def create_ask_response(
             search_sdk=rpc.get_sdk("search"),
             reader_sdk=rpc.get_sdk("reader"),
             kbid=kbid,
+            predict_manager=predict_manager,
             ask_request=ask_request,
             user_id=user_id,
             client_type=client_type,

{nucliadb_agentic_api-1.0.0.post216 → nucliadb_agentic_api-1.0.0.post226}/src/nucliadb_agentic_api/v1/mcp_nucliadb.py

@@ -10,6 +10,7 @@ import anyio
 import pydantic_core
 from fastapi import Header
 from hyperforge.api.authentication import requires_one
+from hyperforge.manager import Manager
 from mcp.server.fastmcp.exceptions import ResourceError
 from mcp.server.fastmcp.utilities.types import Image
 from mcp.server.lowlevel.helper_types import ReadResourceContents
@@ -34,7 +35,7 @@ from starlette.responses import Response
 from nucliadb_agentic_api import logger
 
 if TYPE_CHECKING:
-    from hyperforge.api.app import HTTPApplication
+    from nucliadb_agentic_api.app import HTTPApplication
 
 from anyio.abc import TaskStatus
 from hyperforge_nucliadb_agentic.ask.exceptions import (
@@ -80,6 +81,7 @@ BATCH_GET_DOCUMENTS_MAX = 20
 
 @dataclass
 class MCPContext:
+    predict_manager: Manager
     ndb_reader: NucliaDBAsync
     ndb_search: NucliaDBAsync
     kbid: str
@@ -249,6 +251,7 @@ async def _tool_search_documents(
             search_sdk=context.ndb_search,
             reader_sdk=context.ndb_reader,
             kbid=context.kbid,
+            predict_manager=context.predict_manager,
             ask_request=ask_request,
             user_id=context.x_nucliadb_user,
             client_type=context.x_ndb_client,
@@ -469,6 +472,7 @@ async def list_tools(description: str) -> list[Tool]:
 
 
 async def call_tool(
+    predict_manager: Manager,
     x_stf_account: str,
     x_nucliadb_user: str,
     x_ndb_client: NucliaDBClientType,
@@ -492,6 +496,7 @@ async def call_tool(
         )
         raise ResourceError(f"Invalid arguments for '{name}': {error_details}") from e
     context = MCPContext(
+        predict_manager=predict_manager,
         ndb_reader=ndb_reader,
         ndb_search=ndb_search,
         kbid=kbid,
@@ -678,6 +683,7 @@ async def mcp_handler(
         list_resources_partial = partial(list_resources, ndb_reader, kbid)
         call_tool_partial = partial(
             call_tool,
+            app.predict_manager,
             x_stf_account,
             x_nucliadb_user,
             x_ndb_client,

nucliadb_agentic_api-1.0.0.post216/VERSION

@@ -1 +0,0 @@
-1.0.0.post216