unique_web_search 2026.36.0.dev3__tar.gz → 2026.36.0.dev5__tar.gz

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-web-search
-Version: 2026.36.0.dev3
+Version: 2026.36.0.dev5
 Summary: 
 Author: Andreas Hauri, Gustav Hartz, Rami Azouz
 Author-email: Andreas Hauri <andreas@unique.ch>, Gustav Hartz <gustav.hartz.ext@unique.ch>, Rami Azouz <rami.ext@unique.ch>
@@ -16,7 +16,7 @@ Requires-Dist: tavily-python>=0.7.11,<1
 Requires-Dist: unidecode>=1.4.0,<2
 Requires-Dist: azure-ai-projects>=2.3,<3
 Requires-Dist: azure-identity>=1.25.0,<2
-Requires-Dist: unique-toolkit[monitoring]>=2026.36.0.dev6,<2026.36.0rc0
+Requires-Dist: unique-toolkit[monitoring]>=2026.36.0.dev12,<2026.36.0rc0
 Requires-Dist: azure-core>=1.36.0,<2
 Requires-Dist: google-cloud-aiplatform>=1.128.0,<2
 Requires-Dist: google-auth>=2.43.0,<3
@@ -25,8 +25,8 @@ Requires-Dist: rapidfuzz>=3.0.0,<4
 Requires-Dist: httpx>=0.28.1
 Requires-Dist: certifi>=2025.11.12
 Requires-Dist: jinja2>=3.1.6
-Requires-Dist: unique-search-proxy-core>=2026.36.0.dev2,<2026.36.0rc0
-Requires-Dist: unique-search-proxy-sdk>=2026.36.0.dev1,<2026.36.0rc0
+Requires-Dist: unique-search-proxy-core>=2026.36.0.dev4,<2026.36.0rc0
+Requires-Dist: unique-search-proxy-sdk>=2026.36.0.dev2,<2026.36.0rc0
 Requires-Dist: openai>=1.105.0,<3
 Requires-Python: >=3.12, <4.0
 Description-Content-Type: text/markdown
@@ -75,10 +75,11 @@ An upstream AI agent / grounded model searches and returns opaque text that this
 
 | Engine | Proxy | Notes |
 |--------|-------|-------|
-| Bing (Azure AI Foundry grounding) | legacy + proxy | Configurable `requires_scraping` |
+| Bing (Azure AI Foundry grounding) | legacy + proxy | Configurable `requires_scraping`; `market` / `setLang` / `freshness` are `ExposableParam` knobs |
 | VertexAI (Gemini grounding) | legacy + proxy | Configurable `requires_scraping` |
 
-Agent engines do **not** use `ExposableParam`.
+Bing's grounding knobs are baked into the hashed Foundry agent name, so changing
+one provisions a new agent version.
 
 ## Web crawlers (page readers)
 

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/README.md

@@ -42,10 +42,11 @@ An upstream AI agent / grounded model searches and returns opaque text that this
 
 | Engine | Proxy | Notes |
 |--------|-------|-------|
-| Bing (Azure AI Foundry grounding) | legacy + proxy | Configurable `requires_scraping` |
+| Bing (Azure AI Foundry grounding) | legacy + proxy | Configurable `requires_scraping`; `market` / `setLang` / `freshness` are `ExposableParam` knobs |
 | VertexAI (Gemini grounding) | legacy + proxy | Configurable `requires_scraping` |
 
-Agent engines do **not** use `ExposableParam`.
+Bing's grounding knobs are baked into the hashed Foundry agent name, so changing
+one provisions a new agent version.
 
 ## Web crawlers (page readers)
 

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_web_search"
-version = "2026.36.0.dev3"
+version = "2026.36.0.dev5"
 description = ""
 readme = "README.md"
 requires-python = ">=3.12,<4.0"
@@ -22,7 +22,7 @@ dependencies = [
     "unidecode>=1.4.0,<2",
     "azure-ai-projects>=2.3,<3",
     "azure-identity>=1.25.0,<2",
-    "unique-toolkit[monitoring]>=2026.36.0.dev6,<2026.36.0rc0",
+    "unique-toolkit[monitoring]>=2026.36.0.dev12,<2026.36.0rc0",
     "azure-core>=1.36.0,<2",
     "google-cloud-aiplatform>=1.128.0,<2",
     "google-auth>=2.43.0,<3",
@@ -31,8 +31,8 @@ dependencies = [
     "httpx>=0.28.1",
     "certifi>=2025.11.12",
     "jinja2>=3.1.6",
-    "unique-search-proxy-core>=2026.36.0.dev2,<2026.36.0rc0",
-    "unique-search-proxy-sdk>=2026.36.0.dev1,<2026.36.0rc0",
+    "unique-search-proxy-core>=2026.36.0.dev4,<2026.36.0rc0",
+    "unique-search-proxy-sdk>=2026.36.0.dev2,<2026.36.0rc0",
     "openai>=1.105.0,<3",
 ]
 

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/src/unique_web_search/service.py

@@ -4,18 +4,19 @@ from time import time
 
 from typing_extensions import override
 from unique_search_proxy_core.context import RequestContext
+from unique_search_proxy_core.param_policy.exposable_config import (
+    ExposableParamsConfig,
+)
 from unique_search_proxy_core.param_policy.exposed_params import ExposedParams
-from unique_search_proxy_core.search_engines.base import BaseSearchEngineConfig
 from unique_toolkit._common.chunk_relevancy_sorter.service import ChunkRelevancySorter
 from unique_toolkit.agentic.evaluation.schemas import EvaluationMetricName
-from unique_toolkit.agentic.feature_flags.feature_flags import (
-    feature_flags,
-)
+from unique_toolkit.agentic.feature_flags import FeatureFlagNames
 from unique_toolkit.agentic.tools.factory import ToolFactory
 from unique_toolkit.agentic.tools.schemas import ToolCallResponse
 from unique_toolkit.agentic.tools.tool import (
     Tool,
 )
+from unique_toolkit.experimental.resources.feature_flags import is_flag_enabled
 from unique_toolkit.language_model.infos import LanguageModelInfo
 from unique_toolkit.language_model.schemas import (
     LanguageModelFunction,
@@ -116,9 +117,13 @@ class WebSearchTool(Tool[WebSearchConfig]):
         self.content_reducer = content_reducer
 
     def _resolve_exposed_params_cls(self) -> type[ExposedParams] | None:
-        """Resolve the exposed-params model from the search-engine config, if any."""
+        """Resolve the exposed-params model from the search-engine config, if any.
+
+        Covers standard and agent engines alike; local engines (custom APIs) do
+        not carry exposable knobs.
+        """
         cfg = self.config.search_engine_config
-        if isinstance(cfg, BaseSearchEngineConfig):
+        if isinstance(cfg, ExposableParamsConfig):
             return cfg.exposed_params_model()
         return None
 
@@ -311,8 +316,9 @@ class WebSearchTool(Tool[WebSearchConfig]):
     async def _get_argument_screening_service_if_ff_enabled(
         self,
     ) -> ArgumentScreeningService | None:
-        if not feature_flags.enable_web_search_argument_screening_un_18741.is_enabled(
-            self.company_id
+        if not await is_flag_enabled(
+            FeatureFlagNames.enable_web_search_argument_screening_un_18741,
+            company_id=self.company_id,
         ):
             return None
 

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/src/unique_web_search/services/search_engine/README.md

@@ -28,12 +28,12 @@ Deployment configs for Google / Brave / Perplexity (and agent base fields) live
 |--|----------|-------------------|
 | Proxy route | `POST /v1/search` | `POST /v1/agent-search` |
 | Result shape | `SearchResponse` → `WebSearchResult` list | Opaque `answer` text → parsed locally |
-| LLM knobs | `ExposableParam` + `config.merge()` | No exposable params |
+| LLM knobs | `ExposableParam` + `config.merge()` | `ExposableParam` (Bing only) |
 | Typical scrape | Often yes (Google) | Often no (`requires_scraping=false`) |
 
 ---
 
-## Exposable parameters (standard search only)
+## Exposable parameters
 
 Optional provider knobs use `ExposableParam[T]` from `unique_search_proxy_core.param_policy.exposable_param`:
 
@@ -54,7 +54,7 @@ ExposableParam(expose=False, value=None)  # deactivated (not merged into the req
 - `config.exposed_params_model()` — LLM-facing subset
 - `config.merge(llm_overrides, query=...)` — flat HTTP body for the proxy (no nested config+invocation)
 
-Crawlers and agent engines do **not** use this pattern.
+All standard engines and the Bing agent engine use this pattern; crawlers do not.
 
 ```python
 from unique_search_proxy_core.param_policy.exposable_param import ExposableParam
@@ -149,6 +149,15 @@ CUSTOM_WEB_SEARCH_API_HEADERS='{"Authorization": "Bearer ..."}'
 
 **Provider:** Azure AI Foundry Agent with Bing grounding · Config extends `BingAgentConfig` with `requires_scraping`, `language_model`, etc.
 
+| Field | Notes |
+|-------|-------|
+| `fetch_size` | Bing result count |
+| `market`, `set_lang`, `freshness` | `ExposableParam`; `market` / `set_lang` restricted to Bing's documented codes; forwarded to `BingGroundingSearchConfiguration` |
+
+Bing bakes the tool configuration into the *agent version*, so these knobs take
+part in the hashed agent name (`unique-grounding-with-bing-<hash>`): changing one
+provisions a new agent version instead of reusing a mismatched one.
+
 ```bash
 AZURE_AI_PROJECT_ENDPOINT=...
 AZURE_AI_AGENT_ID=...          # optional; empty → auto-provision
@@ -161,6 +170,12 @@ AZURE_IDENTITY_CREDENTIAL_TYPE=workload  # or default
 
 Requires GCP ADC / workload identity. Fields include `vertexai_model_name`, `generation_instructions`, `enable_entreprise_search`, `enable_redirect_resolution`, `requires_scraping`.
 
+```bash
+VERTEXAI_AGENT_DEFAULT_VERTEX_AI_MODEL_NAME=gemini-3-flash-preview
+# ZDR tenants: pin "Enable Enterprise Search" on and disable the checkbox
+VERTEXAI_AGENT_FORCE_ACTIVATE_ENTERPRISE_SEARCH=false
+```
+
 ---
 
 ## Factory

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/src/unique_web_search/services/search_engine/base.py

@@ -8,6 +8,9 @@ from pydantic import BaseModel
 from unique_search_proxy_core.agent_engines.base import AgentEngineType
 from unique_search_proxy_core.agent_engines.config_types import ENGINE_NAME_TO_CONFIG
 from unique_search_proxy_core.context import LOCAL_REQUEST_CONTEXT, RequestContext
+from unique_search_proxy_core.param_policy.exposable_param import (
+    resolve_exposable_value,
+)
 from unique_search_proxy_core.param_policy.exposed_params import ExposedParams
 from unique_search_proxy_core.search_engines.base import (
     BaseSearchEngineConfig,
@@ -155,8 +158,7 @@ class SearchEngine(ABC, Generic[SearchEngineConfig]):
         engine: AgentEngineType,
     ) -> list[WebSearchResult]:
         """Dispatch a grounded agent search via the proxy SDK and parse the answer."""
-        del params  # Agent engines do not expose LLM knobs today.
-        invocation = self._agent_proxy_invocation(engine)
+        invocation = self._agent_proxy_invocation(engine, params)
         async with open_search_proxy_client(
             timeout=float(self.config.timeout),
             context=self._request_context,
@@ -172,8 +174,16 @@ class SearchEngine(ABC, Generic[SearchEngineConfig]):
         )
         return await self._postprocess_results(results)
 
-    def _agent_proxy_invocation(self, engine: AgentEngineType) -> dict[str, Any]:
-        """Collect the config fields accepted by the agent proxy call for ``engine``."""
+    def _agent_proxy_invocation(
+        self,
+        engine: AgentEngineType,
+        params: ExposedParams | None,
+    ) -> dict[str, Any]:
+        """Collect the config fields accepted by the agent proxy call for ``engine``.
+
+        Exposable knobs contribute their admin ``value``; per-call ``params``
+        chosen by the LLM win over the deployment defaults.
+        """
         core_config_cls = ENGINE_NAME_TO_CONFIG[engine.value]
         invocation: dict[str, Any] = {}
         for field_name in core_config_cls.model_fields:
@@ -181,11 +191,13 @@ class SearchEngine(ABC, Generic[SearchEngineConfig]):
                 continue
             if not hasattr(self.config, field_name):
                 continue
-            value = getattr(self.config, field_name)
+            value = resolve_exposable_value(getattr(self.config, field_name))
             # Empty values fall through to the SDK/server defaults.
             if value is None or value == "":
                 continue
             invocation[field_name] = value
+        if params is not None:
+            invocation.update(params.model_dump(exclude_none=True))
         return invocation
 
     async def _postprocess_results(

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/src/unique_web_search/services/search_engine/bing.py

@@ -3,8 +3,14 @@ from typing import override
 
 from pydantic import Field
 from unique_search_proxy_core.agent_engines.base import AgentEngineType
+from unique_search_proxy_core.agent_engines.bing.grounding import (
+    BingGroundingConfiguration,
+)
 from unique_search_proxy_core.agent_engines.bing.schema import BingAgentConfig
 from unique_search_proxy_core.context import LOCAL_REQUEST_CONTEXT, RequestContext
+from unique_search_proxy_core.param_policy.exposable_param import (
+    resolve_exposable_value,
+)
 from unique_search_proxy_core.param_policy.exposed_params import ExposedParams
 from unique_toolkit._common.default_language_model import DEFAULT_LANGUAGE_MODEL
 from unique_toolkit._common.validators import LMI, get_LMI_default_field
@@ -82,20 +88,38 @@ class BingSearch(SearchEngine[BingSearchConfig]):
     def requires_scraping(self) -> bool:
         return self.config.requires_scraping
 
+    def _grounding_configuration(
+        self,
+        params: ExposedParams | None,
+    ) -> BingGroundingConfiguration:
+        """Resolve the Bing tool knobs from deployment defaults + LLM overrides."""
+        overrides = params.model_dump(exclude_none=True) if params else {}
+
+        def knob(field_name: str) -> str | None:
+            if field_name in overrides:
+                return overrides[field_name]
+            return resolve_exposable_value(getattr(self.config, field_name))
+
+        return BingGroundingConfiguration(
+            fetch_size=self.config.fetch_size,
+            market=knob("market"),
+            set_lang=knob("set_lang"),
+            freshness=knob("freshness"),
+        )
+
     @override
     async def _legacy_search(
         self,
         query: str,
         params: ExposedParams | None,
     ) -> list[WebSearchResult]:
-        del params
         agent_client = get_project_client(self.credentials)
 
         async with agent_client:
             search_results = await create_and_process_run(
                 agent_client,
                 query=query,
-                fetch_size=self.config.fetch_size,
+                grounding=self._grounding_configuration(params),
                 response_parsers_strategies=self.response_parsers,
                 generation_instructions=self.config.generation_instructions,
             )

{unique_web_search-2026.36.0.dev3 → unique_web_search-2026.36.0.dev5}/src/unique_web_search/services/search_engine/utils/grounding/bing/runner.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import hashlib
 import logging
 import time
 
@@ -23,6 +22,10 @@ from openai.types.responses.response_output_text import (
     ResponseOutputText,
 )
 from openai.types.responses.response_text_delta_event import ResponseTextDeltaEvent
+from unique_search_proxy_core.agent_engines.bing.grounding import (
+    BingGroundingConfiguration,
+    bing_agent_name,
+)
 
 from unique_web_search.invocation_stats import collector
 from unique_web_search.services.search_engine.schema import (
@@ -41,8 +44,6 @@ from unique_web_search.services.search_engine.utils.grounding.bing.models import
 from unique_web_search.settings import env_settings
 
 _LOGGER = logging.getLogger(__name__)
-BING_AUTO_AGENT_NAME_PREFIX = "unique-grounding-with-bing"
-_CONFIG_HASH_LENGTH = 12
 
 
 # ---------------------------------------------------------------------------
@@ -50,32 +51,21 @@ _CONFIG_HASH_LENGTH = 12
 # ---------------------------------------------------------------------------
 
 
-def _config_hash(*, model: str, fetch_size: int, instructions: str) -> str:
-    """Return a short hex digest of model + fetch_size + instructions for agent naming."""
-    payload = f"{model}\0{fetch_size}\0{instructions}".encode()
-    return hashlib.sha256(payload).hexdigest()[:_CONFIG_HASH_LENGTH]
-
-
-def _agent_name_for_config(*, model: str, fetch_size: int, instructions: str) -> str:
-    """Build a Foundry-safe agent name unique to this config."""
-    return (
-        f"{BING_AUTO_AGENT_NAME_PREFIX}-"
-        f"{_config_hash(model=model, fetch_size=fetch_size, instructions=instructions)}"
-    )
-
-
 def resolve_bing_agent_name(
     *,
     model: str,
-    fetch_size: int,
     instructions: str,
+    grounding: BingGroundingConfiguration,
 ) -> str:
     """Return the hash-based agent name to use for Responses (no Foundry round-trip).
 
-    Derives a stable name from ``model`` + ``fetch_size`` + ``instructions``.
+    Derives a stable name from everything baked into the agent version: the
+    model, the instructions, and the Bing grounding tool configuration.
     """
-    return _agent_name_for_config(
-        model=model, fetch_size=fetch_size, instructions=instructions
+    return bing_agent_name(
+        model=model,
+        instructions=instructions,
+        grounding=grounding,
     )
 
 
@@ -84,8 +74,8 @@ async def create_bing_agent(
     *,
     agent_name: str,
     model: str,
-    fetch_size: int,
     instructions: str,
+    grounding: BingGroundingConfiguration,
 ) -> str:
     """Create a Foundry agent version and return its name."""
     started = time.perf_counter()
@@ -94,7 +84,7 @@ async def create_bing_agent(
         definition=PromptAgentDefinition(
             model=model,
             instructions=instructions,
-            tools=[get_bing_grounding_tool(fetch_size)],
+            tools=[get_bing_grounding_tool(grounding)],
             tool_choice="required",
         ),
         description="Unique Bing grounding agent",
@@ -117,8 +107,8 @@ async def get_or_create_agent_id(agent_client: AIProjectClient) -> str:
     del agent_client  # kept for call-site compatibility
     return resolve_bing_agent_name(
         model=env_settings.azure_ai_bing_agent_model,
-        fetch_size=5,
         instructions=RESPONSE_RULE,
+        grounding=BingGroundingConfiguration(fetch_size=5),
     )
 
 
@@ -127,11 +117,13 @@ async def get_or_create_agent_id(agent_client: AIProjectClient) -> str:
 # ---------------------------------------------------------------------------
 
 
-def get_bing_grounding_tool(fetch_size: int) -> BingGroundingTool:
+def get_bing_grounding_tool(
+    grounding: BingGroundingConfiguration,
+) -> BingGroundingTool:
     """Build a BingGroundingTool configured with the environment connection string.
 
     Args:
-        fetch_size: Maximum number of search results the tool should return.
+        grounding: Bing tool knobs; knobs left unset keep Bing's own defaults.
 
     Raises:
         ValueError: If the Bing resource connection string is not configured.
@@ -144,7 +136,10 @@ def get_bing_grounding_tool(fetch_size: int) -> BingGroundingTool:
             search_configurations=[
                 BingGroundingSearchConfiguration(
                     project_connection_id=connection_id,
-                    count=fetch_size,
+                    count=grounding.fetch_size,
+                    market=grounding.market,
+                    set_lang=grounding.set_lang,
+                    freshness=grounding.freshness,
                 )
             ]
         )
@@ -184,7 +179,7 @@ def _is_missing_agent_error(exc: BaseException, *, agent_name: str) -> bool:
 async def create_and_process_run(
     agent_client: AIProjectClient,
     query: str,
-    fetch_size: int,
+    grounding: BingGroundingConfiguration,
     response_parsers_strategies: list[ResponseParser],
     generation_instructions: str,
 ) -> list[WebSearchResult]:
@@ -196,7 +191,7 @@ async def create_and_process_run(
     Args:
         agent_client: Azure AI project client used to manage agents.
         query: The search query to send to the agent.
-        fetch_size: Maximum number of Bing results the agent should retrieve.
+        grounding: Bing tool knobs (result count, market, language, freshness).
         response_parsers_strategies: Ordered list of parsing strategies to try
             when converting the agent's free-text response into structured results.
         generation_instructions: Per-request system instructions for the agent.
@@ -211,8 +206,8 @@ async def create_and_process_run(
         agent_client,
         query=query,
         model=model,
-        fetch_size=fetch_size,
         instructions=instructions,
+        grounding=grounding,
     )
 
     return await convert_response_to_search_results(answer, response_parsers_strategies)
@@ -223,8 +218,8 @@ async def _run_responses_agent(
     *,
     query: str,
     model: str,
-    fetch_size: int,
     instructions: str,
+    grounding: BingGroundingConfiguration,
 ) -> str:
     """Invoke the agent via Responses API and return the full output text.
 
@@ -234,8 +229,8 @@ async def _run_responses_agent(
     """
     resolved_name = resolve_bing_agent_name(
         model=model,
-        fetch_size=fetch_size,
         instructions=instructions,
+        grounding=grounding,
     )
     openai_client = get_openai_client(agent_client)
 
@@ -257,8 +252,8 @@ async def _run_responses_agent(
             agent_client,
             agent_name=resolved_name,
             model=model,
-            fetch_size=fetch_size,
             instructions=instructions,
+            grounding=grounding,
         )
         stream = await _create_responses_stream(
             openai_client,