"j "%SSKJr SSKrSSKrSSKrSSKrSSKrSSKrSSK r SSK J r J r SSK JrJrJrJr SSKJr SSKJrJrJrJr SSKJr SSKJr SS KJr SS KJr SS KJrJrJ r J!r!J"r"J#r#J$r$J%r%J&r&J'r'J(r( SS K)J*r* SSK+r+SSK,r,SS K-J.r/ SSK0J1r1 SSK2Jr3 SSK4J5r5J6r6J7r7 SSK8J9r9 SSK:J;r;Jr?J@rAJBrCJDrEJFrGJHrIJJrKJLrMJNrOJPrQ SSKRJSrSJTrT SSK5 \bRS?S@5 \#SArjSB\kSC'\#SDrlSB\kSE'\#SFrmSB\kSG'\#SHrnSB\kSI'\#SJroSB\kSK'\#SLrpSB\kSM'\#SNrqSB\kSO'\#SPrrSB\kSQ'\#SRrsSB\kSS'\#STrtSB\kSU'\#SVruSB\kSW'\v\#SX-rwSB\kSY'GSSZjrx\\y\z\ 4\\y\z\ 4S-/\y\z\ 44-S-\+R"\x\y\z\ 4S-S[S\94r|SB\kS]'\"S^S_9"S`Sa55r}GSSbjr~\3"S^\+R"ScSd9Se9"SfSg\ 55r\3"S^\+R"ScSd9Se9"ShSi\55r\3"S^\+R"ScSd9Se9"SjSk\55r\3"S^\+R"ScSd9Se9"SlSm\55r\3"S^\+R"ScSd9Se9"SnSo\55r\"S^S_9"SpSq55r\3"S^\+R"SrSrSs9Se9"StSu55r\3"S^\+R"SrSrSs9Se9"SvSw\55r\3"S^\+R"SrSrSs9Se9"SxSy\55r\"SzS{55r\#S|rSB\kS}'\3"S^\+R"ScSd9Se9"S~S55r\\\-\-\-\\\+GR"\GR54-\-\+GR"S54r\\\\\\4rS\kS'\GR"\GR"\GR"\GR"4rS\kS'GSSjr\z\-\-\-rSB\kS'\z\/\Q\P7rS\kS'Sr\6"S\ S9r\"S^S_9"SS\"\55rSSSSSSSSSSF. rS\kS'SSSSSSS.rS\kS'SSSSSD.rS\kS'SSSSSSSSSH.rS\kS'SSSS.rS\kS'GSSjrGSSjr\"S^S_9"SS55rSr\+GRD"\ \+R"ScSrSrS9S9rS\kS'"SS5r"SS5r\(a\\\ -\\z\ 4-\ -rSB\kS'O;\5"S\\\z\4\\\4-\\zS4-\S-\ -\+GRN"SS945r\+GRD"\\+R"ScS9S9rS\kS'\#SrSB\kS'\'"\5rS\kS'GSSjrSrGSSjr\"S^S_9"SS55r\"S^S_9"SS\55r\"S^S_9"SS\55r\+GRD"\\,GRd\+R"ScS9S9r\"S^S_9"SS55r\"ScS^S9"SS55r\"ScS^S9"SS55r\"ScS^S9"SS55r\\-\-rSB\kS'\"ScS^S9"SS55rGSSjrGSSjr\\S-\+GRx"\5\+R"\\zS-S[S\94rSB\kS'\"S^S_9"SS55r\"S^ScS9"SS55r\"S^S_9"SS55r\"S^S_9"SS55r\"S^S_9"SS55rSr\"S^S_9"SS55r\"S^S_9"SS55r\"S^ScS9"SS55r\"S^S_9"SS55r\"S^S_9"SS\55r\"S^S_9"SS\55r0rS\kS'0rGS\kGS'0rGS\kGS'0rGS\kGS'\6"GSSGS9r\6"GSSGS9rGSGS jrGSGS jr0rGS \kGS '0rGS \kGS'SGSKJrJr SGSKJrJrJrJrJrJrJr GSGSjr\\\}\+GR"GS54\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\+GR"\54rGSGSjrGSGSjr\\\\+GR"GS54\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS54-\\\+GR"GS 54-\\\+GR"GS!54-\\\+GR"GS"54-\\\+GR"GS#54-\\\+GR"GS$54-\\\+GR"GS%54-\\\+GR"GS&54-\\\+GR"GS54-\+GR"\54r\"S^S_9"GS'GS(55rGSGS)jr\\\-\+GR"S54r\+GRD"\\\+R"ScSrSrS9S9rGSGS*jrS^GS+.GSGS,jjrGSGS-jrGSGS.jrGSGS/jrGSGS0jr\6"GS1\GS9rGSGS2jrScS^GS3GS4S^GS4GS5.GSGS6jjrGSGS7jrGSGS8jrGSGS9jrGSGS:jrGSGS;jrGSGS<jr\"S^S_9"GS=GS>55r\"S^ScS9"GS?GS@55r\"S^ScS9"GSAGSB55r\"S^ScS9"GSCGSD55r\\\-\-\-\+GR"GSE54r\"S^ScS9"GSFGSG55r\"S^ScS9"GSHGSI55r\"S^ScS9"GSJGSK55r\"S^ScS9"GSLGSM55Gr\\\-\-G\-\+GR"GSN54Gr\"S^ScS9"GSOGSP55Gr\"S^S_9"GSQGSR55Gr\"S^S_9"GSSGSTG\55Gr\"S^S_9"GSUGSVG\55Gr\"S^S_9"GSWGSX55Gr\"S^S_9"GSYGSZG\55Gr\"S^ScS9"GS[GS\55Gr\"S^S_9"GS]GS^G\55Gr SGS_GK GJ Gr GJ Gr \"S^S_9"GS`GSa55Gr \"S^S_9"GSbGSc55Gr\"S^S_9"GSdGSe55Gr\"S^S_9"GSfGSg55Gr\"S^S_9"GShGSi55Gr\"S^S_9"GSjGSk55Gr\"S^S_9"GSlGSm55Gr\"S^S_9"GSnGSo55Gr\"S^S_9"GSpGSq55Gr\"S^S_9"GSrGSs55Gr\"S^S_9"GStGSu55Gr\G\G\-G\-G\-G\ -G\ -G\-\+GR"GSN54Gr\AG\GSv"5GrGSw\kGSx'\"S^ScS9"GSyGSv55Gr\"S^ScS9"GSzGS{G\S^GS|95GrG\"GS}\GR<"G\555Gr\AG\GS~"5Gr GS\kGS'GSGSjGr!\"S^ScS9"GSGS~55Gr"\"S^ScS9"GSGSG\"S^GS|95Gr#G\"GS\GR<"G\#555Gr$\G\G\-G\-G\-G\-G\-G\-G\-G\-\+GR"GSN54Gr%\G\G\-G\-G\%-G\-G\"-\+GR"GSN54Gr&g() annotationsN)ABCabstractmethod)Callable CollectionMappingSequence)deepcopy)KW_ONLY dataclassfieldreplace)datetime) MimeTypes)PathLike)Path) TYPE_CHECKING AnnotatedAnyClassVarGenericLiteral TypeAlias TypeGuardcastget_argsoverload)urlparse)types)to_snake)r ) TypeAliasTypeTypeVar assert_never)calculate_price_for_usage)_otel_messages_utils) RESERVED_EVENT_TAGS EventRegistryevent_family_schemaguard_post_initinherited_namespaceinject_tag_fieldis_redefinitionkeeps_canonical_registrationshadowed_envelope_fieldsundecorated_field_base)redact_binary_content serialize_any)generate_tool_call_idnow_utc) ModelRetryUnexpectedModelBehavior UserError) RequestUsage)InstrumentationSettings INVALID_JSONzapplication/rtfz.rtfAapplication/vnd.openxmlformats-officedocument.spreadsheetml.sheetz.xlsxGapplication/vnd.openxmlformats-officedocument.wordprocessingml.documentz.docx text/markdownz.mdxz.mdztext/x-asciidocz .asciidoc image/webpz.webp video/3gppz .three_gpvideo/x-matroskaz.mkvvideo/x-ms-wmvz.wmv video/x-flvz.flv audio/aacz.aac audio/aiffz.aiff audio/flacz.flac audio/oggz.oga audio/wavz.wavzapplication/yamlz.yamlz.ymlzapplication/tomlz.tomlzapplication/xmlz.xml)rH audio/mpegrGrFrErDrAudioMediaType)z image/jpegz image/pngz image/gifr?ImageMediaType) zapplication/pdfz text/plainztext/csvr=r<z text/htmlr>zapplication/mswordzapplication/vnd.ms-excelDocumentMediaType)rAzvideo/quicktime video/mp4z video/webmrCz video/mpegrBr@VideoMediaType)wavmp3ogaflacaiffaac AudioFormat)jpegpnggifwebp ImageFormat) csvdocdocxhtmlmdpdftxtxlsxlsxDocumentFormat) mkvmovmp4webmflvmpegmpgwmvthree_gp VideoFormat)stoplengthcontent_filter tool_callerror FinishReason)complete incomplete suspended interruptedModelResponseState)rurxModelRequestStatez allow-localForceDownloadModec([U5(agU$N)callablevalues Q/home/mande/repo/quber/.venv/lib/python3.13/site-packages/pydantic_ai/messages.py!_serialize_provider_details_deltars Ljson) return_type when_usedProviderDetailsDeltaF)reprc\rSrSr%SrS\S'S\S'\"\S9rS\S 'S r S \S 'S r S\S'SSjr \ RrSrg )SystemPromptPartzA system prompt, generally written by the application developer. This gives the model context and guidance on how to respond. strcontentr _default_factoryr timestampN str | None dynamic_ref system-promptzLiteral['system-prompt'] part_kindcx[R"SSS0UR(aSUR0D6/$0D6/$)Ntypetextr)r&TextPartinclude_contentr)selfsettingss rotel_message_parts#SystemPromptPart.otel_message_partss<''uVuT\TlTl 4<<7Puvvrtuvvrrrr:returnz list[_otel_messages.MessagePart])__name__ __module__ __qualname____firstlineno____doc____annotations__r _now_utcrrrrr'dataclasses_no_defaults_repr__repr____static_attributes__rrrrrsZ L$J9Ix9&"K" +:I'9Rw22Hrrc[U[5(aURS5n[R"USS9R 5SS$)zqGenerate stable identifier for multi-modal content to help LLM in finding a specific file in tool call responses.zutf-8F)usedforsecurityN) isinstancerencodehashlibsha1 hexdigest) identifiers r_multi_modal_content_identifierrs@*c""&&w/ << E : D D Fr JJrT)validate_by_name)rconfigch\rSrSr%SrS\S'S\S'SrS\S 'S rS \S '\"SS S 9r S\S'\"SS S 9r S\S'S S SS S S S.SSjjr \ R\SSj55r\ R\SSj55r\SSj5r\\SSj55r\R*rSrg )FileUrlz+Abstract base class for any URL-based file.rurlr rFr{force_downloadNdict[str, Any] | Nonevendor_metadatacomparedefaultUAnnotated[str | None, pydantic.Field(alias='media_type', default=None, exclude=True)] _media_typeUAnnotated[str | None, pydantic.Field(alias='identifier', default=None, exclude=True)] _identifier) media_typerrrrrcgr}r)rrrrrrrrs r__init__FileUrl.__init__srcHUR=(d UR5$)zQReturn the media type of the file, based on the URL or the provided `media_type`.)r_infer_media_typers rrFileUrl.media_type)s;4#9#9#;;rcRUR=(d [UR5$)aThe identifier of the file, such as a unique ID. This identifier can be provided to the model in a message to allow it to refer to this file in a tool call argument, and the tool can look up the file in question by iterating over the message history and finding the matching `FileUrl`. This identifier is only automatically passed to the model when the `FileUrl` is returned by a tool. If you're passing the `FileUrl` as a user message, it's up to you to include a separate text part with the identifier, e.g. "This is file :" preceding the `FileUrl`. It's also included in inline-text delimiters for providers that require inlining text documents, so the model can distinguish multiple files. )rrrrs rrFileUrl.identifier/sL#B488#LLrc[e)z2Infer the media type of the file based on the URL.NotImplementedErrorrs rrFileUrl._infer_media_type@s "!rc[e)zThe file format.rrs rformatFileUrl.formatEs "!rr)rrrrrrrr{rrrrrrrNonerr)rrrrrrrrr rrrpydanticcomputed_fieldpropertyrrrrrr'rrrrrrrrsV5 HJ(-N%-.2O*1jotjKfjotjKf"&!%,115"&"&       * /        <<  M M""""22Hrrc\rSrSr%SrS\S'S\S'SrS\S 'S S S S SS S S .SS jjrSSjr\ SSj5r \ SSj5r Sr g )VideoUrliNzA URL to a video.rrr r video-urlLiteral['video-url']kindNFrrrrrrrcgr}r rrrrrrrrrs rrVideoUrl.__init__\rcUR(ag[RUR5upUc[ SURS35eU$)z5Return the media type of the video, based on the url.rMz+Could not infer media type from video URL: ,. Explicitly provide a `media_type` instead.) is_youtube _mime_types guess_typer ValueErrorr mime_typers rrVideoUrl._infer_media_typejsP ??"--dhh7  =dhhZGst rcF[UR5RS;$)zTrue if the URL is on a YouTube host that models can resolve directly. This is a specific set of hosts rather than every YouTube-owned domain, so `music.youtube.com` is deliberately not one of them. )zyoutu.bez youtube.comzwww.youtube.comz m.youtube.com)rrhostnamers rrVideoUrl.is_youtubeys!**.mmmrc([UR$)zThe file format of the video. The choice of supported formats were based on the Bedrock Converse API. Other APIs don't require to use a format. )_video_format_lookuprrs rrVideoUrl.format $DOO44rr)rrrrrrrr{rrrrrrrrrrrrbool)rrn) rrrrrrrrrrrrrrrrrrNs HJ!,D ,M"&!%,115%0"&"&       * / #         n n55rrc\rSrSr%SrS\S'S\S'SrS\S 'S S S S SS S S .SS jjrSSjr\ SSj5r Sr g )AudioUrlizA URL to an audio file.rrr r audio-urlLiteral['audio-url']rNFrcgr}rrs rrAudioUrl.__init__rrc[RUR5upUc[SURS35eU$)zReturn the media type of the audio file, based on the url. References: - Gemini: https://ai.google.dev/gemini-api/docs/audio#supported-formats z+Could not infer media type from audio URL: rrrrrrs rrAudioUrl._infer_media_typesF #--dhh7  =dhhZGst rc([UR$)z"The file format of the audio file.)_audio_format_lookuprrs rrAudioUrl.formats$DOO44rr)rrrrrrrr{rrrrrrrrrrr)rrU rrrrrrrrrrrrrrrrrs! H$J!,D ,M"&!%,115%0"&"&       * / #        55rrc\rSrSr%SrS\S'S\S'SrS\S 'S S S S SS S S .SS jjrSSjr\ SSj5r Sr g )ImageUrlizA URL to an image.rrr r image-urlLiteral['image-url']rNFrcgr}rrs rrImageUrl.__init__rrc[RUR5upUc[SURS35eU$)z5Return the media type of the image, based on the url.z+Could not infer media type from image URL: rrrs rrImageUrl._infer_media_typesD"--dhh7  =dhhZGst rc([UR$)zThe file format of the image. The choice of supported formats were based on the Bedrock Converse API. Other APIs don't require to use a format. )_image_format_lookuprrs rrImageUrl.formatrrr)rrrrrrrr{rrrrrrrrrrr)rrZrrrrrrs HJ!,D ,M"&!%,115%0"&"&       * / #       55rrc\rSrSr%SrS\S'S\S'SrS\S 'S S S S SS S S .SS jjrSSjr\ SSj5r Sr g ) DocumentUrlizThe URL of the document.rrr rz document-urlLiteral['document-url']rNFrcgr}rrs rrDocumentUrl.__init__rrc[RUR5upUc[SURS35eU$)z8Return the media type of the document, based on the url.z.Could not infer media type from document URL: rrrs rrDocumentUrl._infer_media_type sD"--dhh7  @ Jvw rcpURn[U$![an[SU35UeSnAff=f)zThe file format of the document. The choice of supported formats were based on the Bedrock Converse API. Other APIs don't require to use a format. zUnknown document media type: N)r_document_format_lookupKeyErrorrrres rrDocumentUrl.formatsF __  R*:6 6 R:VWWrc[U5nUR5(d[SU35e[R U5up#UcSnUR U"UR 5US95$)zCreate a `BinaryContent` from a path. Defaults to 'application/octet-stream' if the media type cannot be inferred. Raises: FileNotFoundError: if the file does not exist. PermissionError: if the file cannot be read. zFile not found: application/octet-streamr8)rexistsFileNotFoundErrorrrr3 read_bytes)r=pathrrs r from_pathBinaryContent.from_pathseDz{{}}#&6tf$=> >#..t4  3Js(9jQRRrcRUR=(d [UR5$)aIdentifier for the binary content, such as a unique ID. This identifier can be provided to the model in a message to allow it to refer to this file in a tool call argument, and the tool can look up the file in question by iterating over the message history and finding the matching `BinaryContent`. This identifier is only automatically passed to the model when the `BinaryContent` is returned by a tool. If you're passing the `BinaryContent` as a user message, it's up to you to include a separate text part with the identifier, e.g. "This is file :" preceding the `BinaryContent`. It's also included in inline-text delimiters for providers that require inlining text documents, so the model can distinguish multiple files. )rrr&rs rrBinaryContent.identifiersM#B499#MMrc:SURSUR3$)z*Convert the `BinaryContent` to a data URI.r6r7)rr rs rr>BinaryContent.data_uris t'x }==rc^[R"UR5R5$)zMReturn the binary data as a base64-encoded string. Default encoding is UTF-8.)r  b64encoder&decoders rr BinaryContent.base64s" *1133rc8URRS5$)z1Return `True` if the media type is an audio type.audio/rr9rs ris_audioBinaryContent.is_audio))(33rc8URRS5$)z1Return `True` if the media type is an image type.image/rUrs rr0BinaryContent.is_imagerXrc8URRS5$)z0Return `True` if the media type is a video type.video/rUrs ris_videoBinaryContent.is_videorXrc(UR[;$)z3Return `True` if the media type is a document type.)rrrs r is_documentBinaryContent.is_documents"999rcXUR(a[UR$UR(a[UR$UR (a[ UR$[UR$![an[SUR35UeSnAff=f)z&The file format of the binary content.Unknown media type: N) rVrrr0rr^rrrr)rrs rrBinaryContent.formats N}}+DOO<<+DOO<<+DOO<<.t?? N3DOO3DEFA M Ns(#B#B #B.B B) B$$B)r)r&r%rr'rrrrrr)rrrr)r2r$rzBinaryContent | BinaryImage)r>rrr$)rHz PathLike[str]rr$rr)rrrrrrrr rrr staticmethodr3 classmethodrArIrrrrr>r rVr0r^rarr'rrrrrrr$r$9s7 K JII,-1O*1jotjKf'D &M"&15"*"&  N    /          X XSS$  N N>>44444444:: N N22Hrr$cP\rSrSrSrSSSSS.S SjjrSrSrg) r1iz0Binary content that's guaranteed to be an image.Nr(r*cgr}rr,s rrBinaryImage.__init__r.rc<UR(d [S5eg)Nz>`BinaryImage` must have a media type that starts with "image/")r0rrs r __post_init__BinaryImage.__post_init__}}]^ ^rr)r&r%rzImageMediaType | strrrrrrr)rrrrrrrrrrrlrrrrr1r1sd;"&15"*"&  )    /       _rr1cP\rSrSrSrSSSSS.S SjjrSrSrg) BinaryAudioiz-Binary content that's guaranteed to be audio.Nr(r*cgr}rr,s rrBinaryAudio.__init__r.rc<UR(d [S5eg)Nz>`BinaryAudio` must have a media type that starts with "audio/")rVrrs rrlBinaryAudio.__post_init__rnrr)r&r%rzAudioMediaType | strrrrrrr)rrrrrorrrrqrqsd8"&15"*"&  )    /       _rrqc8\rSrSr%SrSrS\S'SrS\S'S rg ) CachePointiaA cache point marker for prompt caching. Can be inserted into UserPromptPart.content to mark cache boundaries. Models that don't support caching will filter these out. Supported by: - Anthropic - Amazon Bedrock (Converse API) - OpenAI (GPT-5.6 models) - OpenRouter (Anthropic and Gemini models via `OpenRouterModel`, plus OpenAI GPT-5.6 models when using `OpenAIChatModel` or `OpenAIResponsesModel` with `OpenRouterProvider`) z cache-pointzLiteral['cache-point']r5mzLiteral['5m', '1h']ttlrN) rrrrrrrryrrrrrwrws' $1D 0M#C #rrw) anthropicopenaigooglez google-cloudz google-glaz google-vertexbedrockxaiUploadedFileProviderNamecZ\rSrSr%SrS\S'S\S'S\S'S rS \S '\"S S S 9rS\S'\"S S S 9r S\S'Sr S\S'S S S SS S S.SSjjr \ R\SSj55r\ R\SSj55r\SSj5r\R&rSrg ) UploadedFileiFaA reference to a file uploaded to a provider's file storage by ID. This allows referencing files that have been uploaded via provider-specific file APIs rather than providing the file content directly. Supported by: - [`AnthropicModel`][pydantic_ai.models.anthropic.AnthropicModel] - [`OpenAIChatModel`][pydantic_ai.models.openai.OpenAIChatModel] - [`OpenAIResponsesModel`][pydantic_ai.models.openai.OpenAIResponsesModel] - [`BedrockConverseModel`][pydantic_ai.models.bedrock.BedrockConverseModel] - [`GoogleModel`][pydantic_ai.models.google.GoogleModel] (Gemini API: [Files API](https://ai.google.dev/gemini-api/docs/files) URIs, Google Cloud: GCS `gs://` URIs) - [`XaiModel`][pydantic_ai.models.xai.XaiModel] rfile_idr provider_namer rNrrFrrrrrz uploaded-fileLiteral['uploaded-file']r)rrrrrrcgr}r) rrrrrrrrrs rrUploadedFile.__init__rrcURb UR$[UR5n[R UR 5up#U=(d S$)aReturn the media type of the file, inferred from `file_id` if not explicitly provided. Note: Inference relies on the file extension in `file_id`. For opaque file IDs (e.g., `'file-abc123'`), the media type will default to `'application/octet-stream'`. Inference relies on Python's `mimetypes` module, whose results may vary across platforms. Required by some providers (e.g., Bedrock) for certain file types. rD)rrrrrrH)rparsedrrs rrUploadedFile.media_typesL    '## #$,,'"--fkk: 666rcRUR=(d [UR5$)aJThe identifier of the file, such as a unique ID. This identifier can be provided to the model in a message to allow it to refer to this file in a tool call argument, and the tool can look up the file in question by iterating over the message history and finding the matching `UploadedFile`. This identifier is only automatically passed to the model when the `UploadedFile` is returned by a tool. If you're passing the `UploadedFile` as a user message, it's up to you to include a separate text part with the identifier, e.g. "This is file :" preceding the `UploadedFile`. )rrrrs rrUploadedFile.identifiersP#B4<<#PPrc*URnURS5(a [U$URS5(a [U$URS5(a [U$[ U$![ an[SU35UeSnAff=f)zA general-purpose media-type-to-format mapping. Maps media types to format strings (e.g. `'image/png'` -> `'png'`). Covers image, video, audio, and document types. Currently used by Bedrock, which requires explicit format strings. rZr]rTrdN)rr9rrrrrrrs rrUploadedFile.formats__  I$$X..+J77&&x00+J77&&x00+J77.z:: I3J<@Aq H Is(A4A4 A4+A44 B>B  Br)rrrrrrrrrrrrrrrrrrr)rrrrrrrr rrrrrrrrrrr'rrrrrrrrFsY L,+J-1O*1jotjKfjotjKf&5D "4M"&15!%)8"&"&  0    /  '         7 7  Q QII&22Hrrrztuple[type, ...]MULTI_MODAL_CONTENT_TYPESztuple[str, ...]_FILE_URL_KINDSc"[U[5$)zBCheck if obj is a MultiModalContent type, enabling type narrowing.)rr)objs ris_multi_modal_contentrs c4 55r UserContent_USER_CONTENT_TYPESzSerialize the value yourself before passing it, e.g. with Pydantic (`pydantic_core.to_json()`) or `pydantic_ai.format_as_xml()`._ToolReturnValueT)rc\rSrSr%SrS\S'S\S'SrS\S 'SrS \S 'SrS \S 'Sr S\S'\ Rr Sr g) ToolReturniuA structured tool return that separates the tool result from additional content sent to the model. Can be parameterized with a type to enable return schema generation: - `ToolReturn[User]` — generates a return schema for `User` - `ToolReturn` (bare) — no return schema generated ToolReturnContent return_valuer rN"str | Sequence[UserContent] | Nonerrrzlist[str] | Nonetools tool-returnLiteral['tool-return']rr)rrrrrrrrrrr'rrrrrrrrs^$#;J26G /6HcP"E "$1D 022Hrrr`rar[r]rcr^r_r\rbzdict[str, DocumentFormat]rrPrOrRrQrSrT)rIrHrFrGrErDzdict[str, AudioFormat]rrVrWrXrYzdict[str, ImageFormat]rrerfrgrhrirjrlrmzdict[str, VideoFormat]rimageaudiovideo)rrrz-dict[str, Literal['image', 'audio', 'video']]_kind_to_modality_lookupcURS5(agURS5(agURS5(agg)zCInfer modality from media type for OTel GenAI semantic conventions.rZrrTrr]rN)r9)rs r_infer_modality_from_media_typerFsAX&&   x ( (   x ( ( rcRURS:aU[R"SUS9n[U5nUbXCS'UR(aUR (a U"5US'U$[R "SUS9nUR(aUR (a U"5US'U$)z=Convert binary content to OTel message part based on version.blob)rrmodalityrr()rr)versionr&BlobPartrrinclude_binary_contentBinaryDataPart)rbase64_contentr blob_partrconverted_parts r_convert_binary_to_otel_partrQs1"++:N 2:>  $,j !  # #(G(G#1#3Ii '66HQ[\  # #(G(G(6(8N9 %rc\rSrSr%SrS\S'S\S'\"\S9rS\S 'S r S \S 'SS jr SSjr \ RrSrg)UserPromptPartidaA user prompt, generally written by the end user. Content comes from the `user_prompt` parameter of [`Agent.run`][pydantic_ai.agent.AbstractAgent.run], [`Agent.run_sync`][pydantic_ai.agent.AbstractAgent.run_sync], and [`Agent.run_stream`][pydantic_ai.agent.AbstractAgent.run_stream]. zstr | Sequence[UserContent]rr rrrr user-promptzLiteral['user-prompt']rc URn[U[5(ag[U[5(a[U[[ -5(a([ S[U5RS[35e[U5HEup#[U[5(aM[ SUS[U5RS[35e g)NzT`UserPromptPart.content` must be a `str` or a sequence of `UserContent` items, got `z`. z`UserPromptPart.content[z&]` must be a `UserContent` item, got `) rrrr r% bytearrayrrr_NOT_USER_CONTENT enumerater)rrindexitems rrlUserPromptPart.__post_init__ws,, gs # # '8,, 7EIDU0V0VW ../s3D2EG %W-KEd$788 .ug6 J//04E3FH.rc /n[UR[5(a UR/O URnUGHn[U[[-5(aa[U[5(aUO URnUR [ R "SSS0UR(aSU0O0D65 M[U[[-[-[-5(aURS:a~[ R"SS9n[RUR 5nUbXvS'UR"US'UR(aUR&US'UR U5 GM9UR [ R("SSUR 0UR(aS UR&0O0D65 GM[U[*5(a-UR [-UR"U4S jU55 GM[U[.5(a}UR"R1S S 5S nUS;aUnOSn[ R2"SXtR"S9n UR(aUR4U S'UR U 5 GM`[U[65(aGMxUR SUR 05 GM U$![$a GNf=f)Nrrrruri)rrrrcUR$r}r ps r3UserPromptPart.otel_message_parts..s Z[ZbZbr/r%r)rrrdocumentfile)rrrrr)rrrrappendr&rrrrrrrUriPartrgetrrrr MediaUrlPartr$rrr;FilePartrrw) rrpartsrpart content_struri_partrcategory file_parts rr!UserPromptPart.otel_message_partssX24;EdllTW;X;X$,,^b^j^jD$k 122&0s&;&;d  "++#DLD\D\K(@bd D(X"5 "Ch"NOO##q(-555AH7;;DIIFH+/7,04- //*.((LL*LL&33!%3;3K3Kudhh/QS D-00 9$//TXKbdlmnD,// ??00a8;::'H)H*33(^m^mn +++/<>%11<>5'33@@1!--@@I% A?  rrN)rrrrrr)rr) rrrrrrgrrfrrrrrrrsH&P  (E  !    rrrr) union_mode)rz'pydantic.TypeAdapter[ToolReturnContent]tool_return_content_ta)z tool-searchzcapability-load ToolPartKindztuple[ToolPartKind, ...]_TOOL_PART_KINDSc8^[U4Sj[5S5$)a:Return `value` if it's a known [`ToolPartKind`][pydantic_ai.messages.ToolPartKind], else `None`. UI adapters call this at the wire boundary to validate an untrusted client-supplied `tool_kind` string before setting it on a part, so an unknown value degrades to `None` rather than asserting a bogus discriminator. c36># UHoT:XdM Uv M g7fr}r).0rrs r "parse_tool_kind..rsD"2$em"2s  N)nextrrs`rparse_tool_kindrks D"2Dd KKrz;The tool call was interrupted before a result was produced.cS[R"U5S[R"U5S[R"U5S3S4$)uuThe open and close tags framing one tool-produced file that has to travel on the user channel. A provider whose tool result channel is text-only can only deliver a tool's multimodal output on the user channel — the same one the end user's own uploads travel on — which leaves the model unable to tell a tool attachment from something the person it is talking to attached. The tags name the call the media came from, so tool output, which may be attacker-influenced, is no longer presented at user trust level. `Model.prepare_messages` frames a `SystemPromptPart` a provider can't send natively as `...` for the same reason. Each file is framed on its own so that `file_id` carries the same identifier the tool result text cross-references as "See file .", which one set of tags around a whole call's files could not name. They identify rather than prove. This is prompt text like any other, so a tool can emit the closing tag and a user can type the opening one; the attribute values are escaped, and a provider that accepts media in its tool result channel sends it there and never renders these at all. zz)r^escape) tool_name tool_call_idfile_identifiers r_tool_result_provenance_tagsr{sP& #4;;y#9"::J4;;WcKdJef[[12" 6 rc\rSrSr%SrS\S'S\S'\"\S9rS\S'S \S 'S r S \S 'S r S\S'\"\ S9r S\S'Sr S\S'S(SjrS)Sjr\S*Sj5r\SS.S+Sjj5r\SS.S,Sjj5r\SS.S-Sjj5rSSS.S.S jjrSS.S/S!jjrSS.S0S"jjrS1S#jrSS.S2S$jjrS3S%jrS4S&jr\R6rS'rg )5BaseToolReturnPartiz!Base class for tool return parts.rr rrrr r rNToolPartKind | None tool_kindrrrrsuccessz5Literal['success', 'failed', 'denied', 'interrupted']outcomecR[UR5(a/UR/S4$[UR[5(aO/n/nURH7n[U5(aUR U5 M&UR U5 M9 XS4$UR//S4$)zSplit content into non-file and file parts. Returns: A 3-tuple of (`data_parts`, `file_parts`, `was_list`) where `was_list` indicates whether the original content was a list. FT)rrrlistr)r non_filesfilesrs r_split_content!BaseToolReturnPart._split_contents "$,, / / ~u, ,  d + +#%I-/E\\)!,,LLO$$Q' " T) ) ~r5((rcUR5upnU(dSU4$[U5S:XaU(a[U5(aUSU4$X4$)zSplit content and unwrap single-item data lists. Returns the unwrapped data value (or None if empty) and the file parts. Single-item lists are unwrapped when content was scalar or when files were filtered out. Nr%r)rr:r)rr&rwas_lists r _unwrap_dataBaseToolReturnPart._unwrap_datasQ !% 3 3 5X;  t9>8tE{{7E> !{rc,UR5upnU$)znThe multimodal file parts from `content` (`ImageUrl`, `AudioUrl`, `DocumentUrl`, `VideoUrl`, `BinaryContent`).)r)rrrs rrBaseToolReturnPart.filess))+ ! rrawrcgr}r)rrs r content_items BaseToolReturnPart.content_itemssY\rT wrap_if_errorcgr}rrrr&s rr#r$ssvrcgr}rr(s rr#r$s),rrr&c [UR[5(a URnO UR/nUS:XaU$U(aEURS:Xa5US:XaUR 5OUR 5nU/UR Q$/nUHn[U5(aURU5 M&[U[5(aURU5 MNUS:Xa3UR[RUSS9R55 MUR[RUSSS95 M U$)aReturn content as a flat list for iteration, with optional serialization. Args: mode: Controls serialization of non-file items: - `'raw'`: No serialization. Returns items as-is. - `'str'`: Non-file items are serialized to strings via `tool_return_ta`. File items (`MultiModalContent`) pass through unchanged. - `'jsonable'`: Non-file items are serialized to JSON-compatible Python objects via `tool_return_ta`. File items pass through unchanged. wrap_if_error: Whether to wrap failed tool returns in an `{"error": ...}` object (ignored in `'raw'` mode). When `True` (the default), a failed return's non-file data collapses into a single wrapped error item so providers without a native error channel still see the failure explicitly; files pass through unchanged. Set this to `False` when the provider has a native error channel (e.g. Anthropic `is_error`) and should receive the content unwrapped. r!failedrTby_aliasrrr.)rrrrmodel_response_strmodel_response_objectrrrrr dump_jsonrQ dump_python)rrr&itemswrappedresultrs rr#r$s& dllD ) )LLE\\NE 5=L T\\X5375=d--/dF`F`FbG)djj) )PRD%d++ d#D#&& d# n66td6KRRTU n88FUY8Z[ rc"UR5up#UcSnO:[U[5(aUnO"[R USS9R 5nU(a5UR S:Xa%[R SU05R 5$U$)a1Return a string representation of the data content for the model. This excludes multimodal files - use `.files` to get those separately. Args: wrap_if_error: Whether to wrap failed tool returns in an `{"error": ...}` object. Set this to `False` when the provider has a native error channel. Tr-r,rs)rrrrr2rQr)rr&rrresponses rr0%BaseToolReturnPart.model_response_str,s$$& =H s # #H%///ELLNH T\\X5!++Wh,?@GGI IrcU(a!URS:XaSURSS90$UR5up#Uc0$[R USSS9n[ R "U5(aU$[U0$)aReturn a dictionary representation of the data content, wrapping non-dict types appropriately. This excludes multimodal files - use `.files` to get those separately. Gemini supports JSON dict return values, but no other JSON types, hence we wrap anything else in a dict. Args: wrap_if_error: Whether to wrap failed tool returns in an `{"error": ...}` object. Set this to `False` when the provider has a native error channel. r,rsFr%rTr/)rr0rrr3r' is_str_dictRETURN_VALUE_KEY)rr&rr json_contents rr1(BaseToolReturnPart.model_response_objectAs~ T\\X5T4454IJ J$$& =I%11%ft1T   l + + $l3 3rcURn[U[5(a[R"U5n[U[ 5(a [SU5$[U[5(a [SU5$g![ a gf=f)aReturn `content` as structured JSON data (a `dict` or `list`), or `None` if it has none. A JSON string is parsed; already-structured content is returned as-is; a plain/non-JSON string, scalar, or multimodal content yields `None` (there is no structured payload). A read-side companion to [`files`][pydantic_ai.messages.BaseToolReturnPart.files] and [`model_response_object`][pydantic_ai.messages.BaseToolReturnPart.model_response_object]; some UI wire formats (e.g. AG-UI) transmit tool results as JSON strings, so [`narrow_type`][pydantic_ai.messages.ToolReturnPart.narrow_type] uses it to recover the structured payload a typed return subclass expects. Ndict[str, Any]z list[Any]) rrrr from_jsonrrrr)rrs rstructured_content%BaseToolReturnPart.structured_contentWs},, gs # # '11': gt $ $('2 2 gt $ $ W- -   sA<< B B cUR5up#nU(dURUS9/4$/n/nURSSS9Hn[U5(aaUR SUR S35 [ URURUR 5upURXU /5 Mt[U[5(dMUR U5 M U(aiURS:XaYSURSS90n UV s/sHn SU R S3PM n n [RU /U Q5R5U4$U(a%[RU5R5U4$US U4$s sn f) aZBuild a text-only tool result with multimodal files extracted for a trailing user message. For providers whose tool result API only accepts text. Multimodal files are referenced by identifier in the tool result text ('See file {id}.') and included in full in the returned file content list. Each file is framed by `_tool_result_provenance_tags` so the model can tell it from the user's own uploads, which travel on the same channel. Args: wrap_if_error: Whether to wrap failed tool returns in an `{"error": ...}` object. Set this to `False` when the provider has a native error channel. r%rFr*z See file .r,rsr)rr0r#rrrrr r extendrrrrr2rQ) rr&rrrtool_content_parts file_contentropen_tag close_tagrsrfile_referencess r#model_response_str_and_user_content6BaseToolReturnPart.model_response_str_and_user_contentns~"002(***GK K(**, &&E&GD%d++"))Idoo5Fa*HI&B4>>SWSdSdfjfufu&v###XY$?@D#&&"))$/ H T\\X5d55E5JKEJOP%$4??*;1=%OP!++U,E_,EFMMOQ]] ] !++,>?FFH,V V"!$l22 Qs Fc[R"SURURS9nUR(a/UR b"[ [UR U55US'U/$)Ntool_call_responseridnamer6)r&ToolCallResponsePartr r rrr3r2rrrs rr%BaseToolReturnPart.otel_message_partssZ22%     # # (@*+@x+XYDNv rcURSL$)z-Return `True` if the tool return has content.Nrrs r has_contentBaseToolReturnPart.has_contents||4''rr)rz/tuple[list[Any], list[MultiModalContent], bool])rz#tuple[Any, list[MultiModalContent]])rzlist[MultiModalContent])rzLiteral['raw']rzlist[ToolReturnContent])rzLiteral['str']r&rrzlist[str | MultiModalContent])rzLiteral['jsonable']r&rrzlist[Any | MultiModalContent])rz!Literal['raw', 'str', 'jsonable']r&rrzWlist[ToolReturnContent] | list[str | MultiModalContent] | list[Any | MultiModalContent])r&rrr)r&rrrA)rz!dict[str, Any] | list[Any] | None)r&rrztuple[str, list[UserContent]]rr)rrrrrrr _generate_tool_call_idr rrrrrrrrrrr#r0r1rCrMrrYr'rrrrrrrrsG+N/ F.DEL#E J%)I") HcP9Ix90ENG BN"),  6;\\ KOvv BF,*,;?, &,, B4,.LP%3N (22HrrcP\rSrSr%SrS\S'SrS\S'\SS .S S jj5rS r g) ToolReturnPartizAA tool return message, this encodes the result of running a tool.r rrrrNrc$[U[U5$)u#Promote a base `ToolReturnPart` to its typed subclass when its `tool_kind` is registered. Best-effort: returns the part unchanged when the `tool_kind` (kwarg or on the part) resolves to no registered subclass, and strips an unsubstantiated `tool_kind` when the part's data doesn't validate against that subclass — keeping it on a base part would break a [`ModelMessagesTypeAdapter`][pydantic_ai.messages.ModelMessagesTypeAdapter] round-trip. For direct construction; Pydantic deserialization promotes automatically via the discriminated union. )_narrow_return_TOOL_RETURN_NARROWERSrrs rr3ToolReturnPart.narrow_typesd$:IFFrr)rr]rrrr] rrrrrrrrfr3rrrrr]r]s/KJ(5I%5RNR G Grr]cp\rSrSr%SrS\S'SrS\S'SrS\S 'S rS \S '\ SS .SSjj5r Sr g)NativeToolReturnPartia]A tool return message from a native tool. For native tools with a stable cross-provider shape (currently `tool_search`), a `NativeToolReturnPart` may be promoted to a typed subclass like [`NativeToolSearchReturnPart`][pydantic_ai.messages.NativeToolSearchReturnPart] with a narrowed `content` `TypedDict`. See `NativeToolCallPart` for the pattern. r rNrrrprovider_detailsbuiltin-tool-returnzLiteral['builtin-tool-return']rr^c$[U[U5$)u)Promote a base `NativeToolReturnPart` to its typed subclass when its `tool_kind` is registered. Best-effort: returns the part unchanged when the `tool_kind` (kwarg or on the part) resolves to no registered subclass, and strips an unsubstantiated `tool_kind` when the part's data doesn't validate against that subclass — keeping it on a base part would break a [`ModelMessagesTypeAdapter`][pydantic_ai.messages.ModelMessagesTypeAdapter] round-trip. For direct construction; Pydantic deserialization promotes automatically via the discriminated union. )r`_NATIVE_RETURN_NARROWERSrbs rr3 NativeToolReturnPart.narrow_typesd$!>?? LLUEL B  78  ,  rc [UR[5(a*URcSUR3nOURnOURcO[ UR5VVs0sH,up#U[ UR SS55S::aSS1OS1_M. nnnOSS10n[RURUSS 9n[UR[5=(a [ UR5S:gn[ UR5S U(aS OS S UR5S3nUS3$s snnf)z>Return a string message describing why the retry is requested.zValidation feedback: locrr%ctxinput__all__)excludeindentz validation errorsr8z : ```json z ```z Fix the errors and try again.) rrrr rr:rerror_details_tar2rrQ)r descriptionirr} json_errorsplurals rmodel_responseRetryPromptPart.model_response(s/ dllC ( (~~% 6t||nE "ll ~~%_himiuiu_v_vW[WXA3quuUB/?+@A+Ew'E7R_v%ug.*44T\\7[\4]K d3NDLL8IQ8NFt||$%%6fs"6M\ZeZlZlZnYootu ?@@s/3D=c<URc?[R"SSS0UR(aSUR 50D6/$0D6/$[R "SUR URS9nUR(aUR 5US'U/$)NrrrrPrQr6r)r r&rrrrTr rUs rr"RetryPromptPart.otel_message_parts@s >> !''JRJbJbY0C0C0E$F hj  "66)$$^^D ''!%!4!4!6X6Mrr)rsz*pydantic_core.ValidationError | ModelRetryr rr rrrmrr)rrrrrrr r r[r rrrrgrurrr'rrrrrrrmrms 43 J Iz 7.DEL#E  9Ix96)7I&7R !%#' 9  !   ,A0&22Hrrm)frozenrc:\rSrSrSrSSjr\RrSr g)AgentInstructionSourcei\zYThe agent's own instructions. There is exactly one agent in scope, so it carries no id. cg)Nagentrrs r__str__AgentInstructionSource.__str__csrrNr) rrrrrrr'rrrrrrrr\s 22HrrcF\rSrSr%SrS\S'SSjr\Rr Sr g) ToolsetInstructionSourceiizA toolset with an `id`.rrRc SUR3$)Nztoolset:rRrs rr ToolsetInstructionSource.__str__os$''##rrNr rrrrrrrr'rrrrrrrris! G$22HrrcF\rSrSr%SrS\S'SSjr\Rr Sr g) CapabilityInstructionSourceiuzA capability with an `id`.rrRc SUR3$)Nz capability:rrs rr#CapabilityInstructionSource.__str__{sTWWI&&rrNrrrrrrrus$ G'22HrrInstructionSourcecb\rSrSr%SrS\S'S\S'SrS\S 'S S jr\Rr S r g) InstructionIdiz`The key an instruction part is addressed by: who contributed it, and which of their parts it is.rsourcer rNrrSc|URbURSUR3$[UR5$)N:)rSrrrs rrInstructionId.__str__s2/3yy/D$++a {+Z#dkkJZZrrr) rrrrrrrSrr'rrrrrrrrs3j dJD*g[22Hrrc[U[5(dU$US:Xa[[55$UR S5(a![[5UR S5S9$S[ 4S[44H^upUS3nUR U5(dM"UR U5RS5upEn[U"U5U(aUOSS9s $ g)zRebuild a key from the string it renders to. A namespace this version doesn't know can only come from a newer one, whose keys it has no way to address anyway, so it reads as unaddressable rather than failing the message it arrived on. rzagent:rStoolset capabilityrN) rrrrr9 removeprefixrr partition)r namespacerr? source_id separatorrSs r_deserialize_instruction_idrs eS ! !  3566 !!35E.s4e99eN)joinstriprs rrInstructionPart.joins'{{4e44::<DDrc[USS9$)z_Sort instruction parts with static (`dynamic=False`) before dynamic, preserving relative order.cUR$r})rrs rr(InstructionPart.sorted..s199rkey)sortedrs rrInstructionPart.sortedse!455rr)rSequence[InstructionPart]rr)rrrzlist[InstructionPart])rrrrrrrrSrRrrfrrr'rrrrrrrrsL4JGTD* #'B&&)6I%5LEE6622Hrr)rkw_onlycv\rSrSr%Sr\"SS9rS\S'SrS\S 'S r S \S 'SS jr \ Rr Srg)ToolAvailabilityDeltaPartiu?Records that the set of tools available to the model changed at this point. Additions only. Withdrawing a tool is not supported yet, because no provider can be told about one without also invalidating the prompt cache this part exists to protect: Anthropic rejects a reference to a tool the request doesn't declare, so a withdrawn tool has to leave the `tools` array, and that is itself the invalidation. The name says *availability* rather than *addition* so removals can join once they can be done cache-safely — see https://github.com/pydantic/pydantic-ai/issues/6985. c/$r}rrrrr"ToolAvailabilityDeltaPart.sbrrzdAnnotated[list[str], pydantic.Field(validation_alias=pydantic.AliasChoices('tools_added', 'added'))] tools_addedNrr tool-availability-deltaz"Literal['tool-availability-delta']rcxSRSUR55n[R"SSU3S9/$)aRender the change as trace content. Tool names are recorded regardless of `include_content`: they aren't user content, they're already visible in the request's tool definitions, and a run where the model suddenly can call something is unreadable without them. z, c3,# UH nSU3v M g7f)+Nr)rrSs rr?ToolAvailabilityDeltaPart.otel_message_parts..$sD3C4avJ3CrzTool availability changed: )rr)rrr&r)rrchangess rr,ToolAvailabilityDeltaPart.otel_message_partss>))D43C3CDD''V?Z[bZc=deffrrr)rrrrrr rrr rrr'rrrrrrrrsU j)* $L*#;4MI1MRg22Hrrc\rSrSr%SrS\S'S\S'SrS\S 'SrS \S 'S rS \S'Sr S \S'Sr S \S'Sr S\S'Sr S\S'SSjr \SS.SSjj5r\R"rSrg) ModelRequesti*zmA request generated by Pydantic AI and sent to a model, e.g. a message from the Pydantic AI app to the model.Sequence[ModelRequestPart]rr rNzdatetime | Nonerr instructionsrequestzLiteral['request']rrun_idconversation_idrrrurzstatecURHCn[U[5(dMURS:wdM,[ SUR<35e g)NuserzE`SpeechPart` in `ModelRequest.parts` must have `speaker='user'`, got rr SpeechPartspeakerrrrs rrlModelRequest.__post_init__TsIJJD$ ++ 0F![\`\h\h[kl r)rc$U"[U5/US9$)z:Create a `ModelRequest` with a single user prompt as text.)rr)r)r= user_promptrs ruser_text_promptModelRequest.user_text_prompt]s.56\RRrrr)rrrrrr)rrrrrrrrrrrrrrlrgrr'rrrrrrrr*sw %%(J"&I%?#L*#_(D (UFJR"&OZ& '+H#*n)E )NRSS22Hrrc\rSrSr%SrS\S'S\S'SrS\S 'SrS\S 'SrS \S 'S r S\S'SSjr \ Rr Srg)riez#A plain text response from a model.rrr rNrrRrrrgrLiteral['text']rc,[UR5$)z/Return `True` if the text content is non-empty.rrrs rrYTextPart.has_contentDLL!!rrrrrrrrrrRrrgrrYr'rrrrrrrresf- L+JB  !%M:$ /3+2 "(I'R"22Hrrc\rSrSr%SrS\S'S\S'SrS\S 'SrS\S 'SrS\S 'Sr S \S 'Sr S\S'SSjr \ RrSrg) ThinkingPartiz!A thinking response from a model.rrr rNrrR signaturerrrgthinkingLiteral['thinking']rc,[UR5$)z3Return `True` if the thinking content is non-empty.rrs rrYThinkingPart.has_contentrrrr)rrrrrrrRrrrgrrYr'rrrrrrrrsx+ L/JB  !Iz  !%M:$ /3+2 &0I"/R"22Hrr#pydantic_ai_standing_prompt_plantedc\rSrSr%SrSrS\S'S\S'SrS\S'SrS\S 'Sr S \S 'S r S \S'SSjr \ RrSrg)CompactionPartiaA compaction part that summarizes previous conversation history. Compaction parts contain an opaque or readable summary of prior messages, produced by provider-specific compaction mechanisms. They must be round-tripped back to the same provider in subsequent requests. For Anthropic, `content` contains a readable text summary. For OpenAI, `content` is `None` and the encrypted data is stored in `provider_details`. Nrrr rrRrrrg compactionzLiteral['compaction']rc,[UR5$)z5Return `True` if the compaction content is non-empty.rrs rrYCompactionPart.has_contentrrrr)rrrrrrrrRrrgrrYr'rrrrrrrrstGZ JB  !%M:$ /3+2 (4I$3R"22Hrrc\rSrSr%SrS\S'S\S'SrS\S 'SrS\S 'SrS \S 'S r S\S'SSjr \ Rr Srg)rizA file response from a model.zLAnnotated[BinaryContent, pydantic.AfterValidator(BinaryContent.narrow_type)]rr rNrrRrrrgrzLiteral['file']rc@[URR5$)z/Return `True` if the file content is non-empty.)rrr&rs rrYFilePart.has_content sDLL%%&&rrrrrrrrrsg' YY+JB  !%M:$ /3+2 "(I'R'22Hrrc\rSrSr%SrS\S'SrS\S'SrS\S 'SrS \S 'Sr S\S 'Sr S\S 'Sr S\S'Sr S\S'\ SSj5rSSjrSSjr\R$rSrg)ri! aSpoken audio exchanged during a realtime session, paired with its transcript. This part is a member of both [`ModelRequestPart`][pydantic_ai.messages.ModelRequestPart] and [`ModelResponsePart`][pydantic_ai.messages.ModelResponsePart], distinguished by `speaker`: in `ModelRequest.parts` the speaker is always `'user'`; in `ModelResponse.parts` it is always `'assistant'`. This invariant is enforced at runtime when a message is constructed. Standard (non-realtime) models can't consume this part directly; when history containing it is used in an agent run, [`Model.prepare_messages`][pydantic_ai.models.Model.prepare_messages] converts user-speaker parts to [`UserPromptPart`][pydantic_ai.messages.UserPromptPart]s and assistant-speaker parts to [`TextPart`][pydantic_ai.messages.TextPart]s. zLiteral['user', 'assistant']rNr transcriptzBinaryContent | Noner int | Noneinterrupted_at_msrRrrrgspeechLiteral['speech']rc,UR=(d S$)zThe transcript, or an empty string if transcription was unavailable. Mirrors [`TextPart.content`][pydantic_ai.messages.TextPart.content] so code that renders message parts generically can treat spoken content like text. r8)rrs rrSpeechPart.contentX s$"$rc$^/nURbFUR[R"SSS0UR(aSUR0O0D65 UR =mb*UR[ TRU4SjU55 U$)Nrrrc>TR$r}r)rsrr/SpeechPart.otel_message_parts..j s PUP\P\rr)rrr&rrrrr)rrrrs @rrSpeechPart.otel_message_partsa s24 ?? & LL''DLD\D\Y$@bd  ZZ E , LL5e6F6FH\^fg h rcV[UR5=(d URSL$)z=Return `True` if the part has a transcript or retained audio.N)rrrrs rrYSpeechPart.has_contentm sDOO$> $(>>rrrrr)rrrrrrrrrrRrrgrrrrrYr'rrrrrrrr! s *)G!J !O"&E & %)z(B a $M:$ /3+2 $,I +R %% ?22Hrrc\rSrSr%SrS\S'SrS\S'\"\S9r S\S 'S \S 'Sr S \S 'Sr S\S'Sr S\S'Sr S\S'SSjrSS.SSjjrSSjrSSjr\R&rSrg)BaseToolCallPartit A tool call from a model.rr Nstr | dict[str, Any] | Noneargsrr r rrrrrRrrrgcSUlgr} otel_metadatars rrlBaseToolCallPart.__post_init__ s NRrF)raise_if_invalidcUR(d0$[UR[5(a UR$[R"UR5n[U[5(dS5e[ [[ [4U5$![[4a U(ae[UR0s$f=f)aReturn the arguments as a Python dictionary. This is just for convenience with models that require dicts as input. Args: raise_if_invalid: If `True`, a `ValueError` or `AssertionError` caused by malformed or non-object JSON in `args` will be re-raised. When `False` (the default), such JSON is handled gracefully by returning `{'INVALID_JSON': ''}` so that the value can still be sent to a model API (e.g. during a retry flow) without crashing. zargs should be a dict) r rrrrBrrrrAssertionErrorINVALID_JSON_KEY)rrr s r args_as_dictBaseToolCallPart.args_as_dict syyI dii & &99  1 **4995DdD)) B+B B)S#X- -N+ 1$dii0 0 1sAB*CCcjUR(dg[UR[5(aA[[R"UR5[ 5(a UR$[R"UR55R5$![ a N>f=f)aReturn the arguments as a JSON string. This is just for convenience with models that require JSON strings as input. JSON that's malformed or doesn't represent an object is handled gracefully by returning `'{"INVALID_JSON":""}'`, matching [`args_as_dict`][pydantic_ai.messages.BaseToolCallPart.args_as_dict], so that the value can still be sent to a model API (e.g. during a retry flow) instead of being rejected by one that requires an object. Because of that, this is not the way to render args that are still streaming in: a partial fragment that only becomes valid JSON once the following deltas are concatenated would be degraded to the wrapper. Emit those verbatim instead, as the UI event streams do. z{}) r rrrrBrrto_jsonrrQrs rargs_as_json_str!BaseToolCallPart.args_as_json_str syy dii % % m55dii@$GG 99$H $$T%6%6%89@@BB  s>B%% B21B2c&URS0S4;$)z+Return `True` if the tool call has content.r8Nr rs rrYBaseToolCallPart.has_content syyR..rrr)rrrrArr)rrrrrrr r r[r rrRrrgrlrrrYr'rrrrrrr r t s#N'(,D %, .DEL#E J%)I") B  !%M:$ /3+2 R8=14C4/22Hrr cP\rSrSr%SrS\S'SrS\S'\SS .S S jj5rS r g) ToolCallParti r r r tool-callzLiteral['tool-call']rNr^c$[U[U5$)u!Promote a base `ToolCallPart` to its typed subclass when its `tool_kind` is registered. Best-effort: returns the part unchanged when the `tool_kind` (kwarg or on the part) resolves to no registered subclass, and strips an unsubstantiated `tool_kind` when the part's data doesn't validate against that subclass — keeping it on a base part would break a [`ModelMessagesTypeAdapter`][pydantic_ai.messages.ModelMessagesTypeAdapter] round-trip. For direct construction; Pydantic deserialization promotes automatically via the discriminated union. ) _narrow_call_TOOL_CALL_NARROWERSrbs rr3ToolCallPart.narrow_type sD"6 BBrr)rr rrrr rdrrrr r  s2#J&1I#1YLP C Crr cP\rSrSr%SrS\S'SrS\S'\SS .S S jj5rS r g) NativeToolCallParti aA tool call to a native tool. For native tools with a stable cross-provider shape (currently `tool_search`), this base class can be promoted to a typed subclass with a narrowed `args` `TypedDict`. See [`NativeToolSearchCallPart`][pydantic_ai.messages.NativeToolSearchCallPart] for the canonical example. Adding a typed subclass for a future native tool (see `pydantic_ai._tool_search` for a worked example): 1. Add a sibling `pydantic_ai/_.py` module that defines the cross-provider `TypedDict`s, the `NativeToolCallPart` / `NativeToolReturnPart` subclasses, and registers their narrowers into `_NATIVE_CALL_NARROWERS` / `_NATIVE_RETURN_NARROWERS` keyed by `tool_kind`. Subclass overrides `tool_kind: Literal['']` to match the emitting [`AbstractNativeTool.kind`][pydantic_ai.native_tools.AbstractNativeTool.kind], and shadows `args` / `content` with a narrower type. 2. Late-import the new module from this file (alongside the existing tool-search import) so registration runs whenever `pydantic_ai.messages` is imported. 3. Add the subclass to `ModelResponsePart`'s discriminated union and to `_model_response_part_discriminator` so Pydantic deserialization auto-promotes on `model_validate` / `model_validate_json`. Dispatch is by `tool_kind`, not `tool_name`. This protects users whose tools happen to share a name with one of ours from accidentally getting their parts promoted (and failing shape validation against the typed `args`/`content`). The `provider_details` field carries genuinely non-portable provider extras (e.g. Anthropic's `strategy: 'bm25' | 'regex'` for tool search). Promote a field to a typed slot in `args` / `content` only when at least two of OpenAI, Anthropic, and Google support it (cf. [issue #3885](https://github.com/pydantic/pydantic-ai/issues/3885)). MCP server tools land here with `tool_kind='mcp_server'` (label stays in `tool_name='mcp_server: