g3fi$ $UdZddlmZddlZddlZddlZddlZddlZddlm Z m Z m Z ddl m Z mZddlmZmZmZmZmZmZmZmZmZmZddlmZmZmZddlmZm Z dd l!m"Z"m#Z#dd l$m%Z%m&Z&dd l'm(Z(dd l)m*Z*m+Z+dd l,m-Z-ddl.m/Z/m0Z0ddl1m2Z2m3Z3ddl4m5Z5ddl6m7Z7m8Z8ddl9m:Z:m;Z;mZ>ddl?m@Z@ddlAmBZB ddlCmDZDdZEejeHZIdWdZJeee"edfee2edfzee*edfzee7edfzee/edfzee;ed fzee#ed!fzee3ed"fzee+ed#fzee8ed$fzee0ed%fzee<ed&fzeeeJ'fZK dX dYd)ZLdZd*ZMd[d+ZNd\d,ZOe%ePeQzeReQeQfzeQzeSeQefzZT d] d^d-ZUd_d.ZV d`d/ZWed0ZXed1d2ZYGd3d4eeXeYfZZ dad5Z[e[dddddddd6 dbd7Z\e[d(d8 dcd9Z]e[d:dddddd; ddd<Z^e%eQzeSeQefzZ_ed=e_>Z`e e`Zaed?ddd@ dedAZbed?ddd@ dfdBZbd?ddd@ dgdCZbdddD dhdEZcdddddF didGZde2e3e"e#e7e8e;e|jr||jz }n!d|jvr||jdz }|j||j|S)aNConvert a sequence of messages to strings and concatenate them into one string. Args: messages: Messages to be converted to strings. human_prefix: The prefix to prepend to contents of `HumanMessage`s. ai_prefix: The prefix to prepend to contents of `AIMessage`. message_separator: The separator to use between messages. Returns: A single string concatenation of all input messages. Raises: ValueError: If an unsupported message type is encountered. Note: If a message is an `AIMessage` and contains both tool calls under `tool_calls` and a function call under `additional_kwargs["function_call"]`, only the tool calls will be appended to the string representation. Example: ```python from langchain_core import AIMessage, HumanMessage messages = [ HumanMessage(content="Hi, how are you?"), AIMessage(content="Good, how are you?"), ] get_buffer_string(messages) # -> "Human: Hi, how are you?\nAI: Good, how are you?" ``` SystemFunctionToolzGot unsupported message type: z: function_call)r2r"rr%r r(rrole ValueErrortext tool_callsadditional_kwargsappendjoin) messages human_prefix ai_prefixmessage_separatorstring_messagesmrLr:messages r;get_buffer_stringrZes JO ( a &D 9 %D = )D ? +D ; 'D ; '66D21#6CS/ !F"QVVH% a #||all^, A$7$77a11/BCDw'5(8  ! !/ 22r=c|d}|dk(rtdi|dS|dk(rtdi|dS|dk(rtdi|dS|dk(rtdi|dS|dk(rt di|dS|dk(rt di|dS|d k(rt di|dS|d k(rtdi|dS|d k(rtdi|dS|d k(rtdi|dS|d k(rtdi|dS|dk(rtdi|dS|dk(rtdi|dSd|}t|)Nr0r@datar>rBrArCrDremoverr#r!r)r&rzGot unexpected message type: )r"rr%rr r(r$rr#r!r)r&rrM)rYtype_r:s r;_message_from_dictr`sv FOE .gfo.. }+76?++ /wv// -WV_-- 111 -WV_-- /wv//   000 ## 376?33 &&#6gfo66 ""2'&/22 $$!4GFO44 ""2'&/22 )% 1C S/r=c>|Dcgc] }t|c}Scc}w)zConvert a sequence of messages from dicts to `Message` objects. Args: messages: Sequence of messages (as dicts) to convert. Returns: list of messages (BaseMessages). )r`)rSrXs r;messages_from_dictrbs,4 4a q ! 44 4sc 2t|ts|Sdg}t|tr|jddgt d|j j ddi|jjDcic] \}}||vs ||c}}Scc}}w)zConvert a message chunk to a `Message`. Args: chunk: Message chunk to convert. Returns: Message. r0tool_call_chunkschunk_positionrr^) r2rrextendr __class____mro____dict__items)chunk ignore_keyskr8s r;message_chunk_to_messageros e- . (K%(.0@AB "" % 4 4 6O1!;:Nq!tO  Ps 7 BBc fi}|||d<|||d<|r?|jddx}r||d<||d<|j|jdi|||d<|vg|d<|D]l} d| vrR| dd } t| trt j | d } |dj | dd| | dd d Y|dj | n|dvr8|jdijdd x} r| |d<tdd|i|} | S|dvr8|jdijdd x} r| |d<tdd|i|} | S|dvr4|dk(r |jdxsi|d<d|dd<tdd|i|} | S|dk(rtdd|i|} | S|dk(r[|jdijdd} |jdijdd}|||d<td|| d|} | S|dk(r tdi|} | Sd|d}t|tj }t#|)aCreate a message from a `Message` type and content string. Args: message_type: the type of the message (e.g., `'human'`, `'ai'`, etc.). content: the content string. name: the name of the message. tool_call_id: the tool call id. tool_calls: the tool calls. id: the id of the message. additional_kwargs: additional keyword arguments. Returns: a message of the appropriate type. Raises: ValueError: if the message type is not one of `'human'`, `'user'`, `'ai'`, `'assistant'`, `'function'`, `'tool'`, `'system'`, or `'developer'`. Nname tool_call_idresponse_metadatarPidrOrC argumentsF)strict tool_call)rqargsrtr0>userr@examplecontent>r> assistant>rB developerr}__openai_role__rDartifactstatus)r{rr]zUnexpected message type: 'z_'. Use one of 'human', 'user', 'ai', 'assistant', 'function', 'tool', 'system', or 'developer'.rY error_coder^)popupdater2r6jsonloadsrQgetr"rr%r r(r$rrMESSAGE_COERCION_FAILURErM) message_typer{rqrrrOrtrPkwargsrsrwrxrzrYrrr:s r;!_create_message_from_message_typers8 F v!-~ 1 5 56I4 P P  P*;F& '&7"#  !2!6!67JB!OP ~t !|# 7IY& ,[9dC(::d59D|$++ )* 5f = $'o + |$++I6 7 ((jj!4b9==iO O7 O 'F9 +FGFvF6 N5 , ,jj!4b9==iO O7 O 'F9 6G6v6. N- 0 0 ; &*0**5H*I*ORF& '=HF& '(9 :::6:$ N#  #!<' O; 6-4* (99I8TH0 O/ GT "\\^  ) 2%>>&1%..39rK5 k %/  O +4=/:SY5W5WX!#&&3 6M)#.A5 6 2%>>&1 2 )MgYWC  (J(JCS/q (  )sNC"9D D!" C>+C99C>DD!DD!! E*,EEcddlm}t||r|jS|Dcgc] }t |c}Scc}w)zConvert a sequence of messages to a list of messages. Args: messages: Sequence of messages to convert. Returns: list of messages (BaseMessages). rr+)langchain_core.prompt_valuesr,r2 to_messagesr)rSr,rXs r;convert_to_messagesrs:9(K(##%%,4 5q  " 55 5s<_P_R_co) covariantcleZdZe d ddZe ddZ d ddZy) _RunnableSupportCallableNcyNr^selfrSrxrs r;__call__z!_RunnableSupportCallable.__call__s @Cr=cyrr^rs r;rz!_RunnableSupportCallable.__call__s r=cyrr^rs r;rz!_RunnableSupportCallable.__call__s HKr=r)rSNonerx_P.argsr _P.kwargsreturnz4Runnable[Sequence[MessageLikeRepresentation], _R_co])rSz1Sequence[MessageLikeRepresentation] | PromptValuerxrrrrrrSz8Sequence[MessageLikeRepresentation] | PromptValue | Nonerxrrrrz<_R_co | Runnable[Sequence[MessageLikeRepresentation], _R_co])r7 __module__ __qualname__rrr^r=r;rrs CCC C > CCC   NRKJKK K F Kr=rcVt d dfd }td|S)Nchddlm}| |g|i|S|tfi|jS)Nr)RunnableLambda)rq)langchain_core.runnables.baserrr7)rSrxrrfuncs r;wrappedz"_runnable_support..wrappeds? A  24262 2gd5f5DMMJJr=#_RunnableSupportCallable[_P, _R_co]rr)rr)rrs` r;_runnable_supportrsY  4[MQ KJ K K K F K K 5w ??r=) include_names exclude_names include_types exclude_types include_ids exclude_idsexclude_tool_callsct|}g}|D]} |r| j|vs|r t| |s|r| j|vr3|dur-t | t r | j st | trdt |tttfrt | t r| j r| j D cgc] } | d|vr|  } } | s| j} t | jtrP| jD cgc];} t | tr'| jddk7s| jd|vr| =} } | j| | d} n t | tr| j|vrj|s|s|r2|r| j|vs"|r t| |s|s| j|vs|j!| |Scc} wcc} w)a Filter messages based on `name`, `type` or `id`. Args: messages: Sequence Message-like objects to filter. include_names: Message names to include. exclude_names: Messages names to exclude. include_types: Message types to include. Can be specified as string names (e.g. `'system'`, `'human'`, `'ai'`, ...) or as `BaseMessage` classes (e.g. `SystemMessage`, `HumanMessage`, `AIMessage`, ...). exclude_types: Message types to exclude. Can be specified as string names (e.g. `'system'`, `'human'`, `'ai'`, ...) or as `BaseMessage` classes (e.g. `SystemMessage`, `HumanMessage`, `AIMessage`, ...). include_ids: Message IDs to include. exclude_ids: Message IDs to exclude. exclude_tool_calls: Tool call IDs to exclude. Can be one of the following: - `True`: All `AIMessage` objects with tool calls and all `ToolMessage` objects will be excluded. - a sequence of tool call IDs to exclude: - `ToolMessage` objects with the corresponding tool call ID will be excluded. - The `tool_calls` in the AIMessage will be updated to exclude matching tool calls. If all `tool_calls` are filtered from an AIMessage, the whole message is excluded. Returns: A list of Messages that meets at least one of the `incl_*` conditions and none of the `excl_*` conditions. If not `incl_*` conditions are specified then anything that is not explicitly excluded will be included. Raises: ValueError: If two incompatible arguments are provided. Example: ```python from langchain_core.messages import ( filter_messages, AIMessage, HumanMessage, SystemMessage, ) messages = [ SystemMessage("you're a good assistant."), HumanMessage("what's your name", id="foo", name="example_user"), AIMessage("steve-o", id="bar", name="example_assistant"), HumanMessage( "what's your favorite color", id="baz", ), AIMessage( "silicon blue", id="blah", ), ] filter_messages( messages, incl_names=("example_user", "example_assistant"), incl_types=("system",), excl_ids=("bar",), ) ``` ```python [ SystemMessage("you're a good assistant."), HumanMessage("what's your name", id="foo", name="example_user"), ] ``` Trtr0tool_use)rOr{)r)rrq_is_message_typertr2rrOr(listtuplesetr{r3r model_copyrrrQ)rSrrrrrrrfilteredr:rwrOr{ content_blocks r;filter_messagesrsj#8,H"$H4! sxx=8"23 "F+ 5   % Y 'CNN#{+  (4*< =#y)cnn&)^^! .@@  "++ckk40.1[[) *=$ ?,008JF,006>PP &Gnn*4I%3 ,1A1AEW1W+#((m";"23 "F+ 5 OOC i4!l OMs 9GAG )chunk_separatorcr|sgSt|}g}|D]}|r|jnd}|s|j|,t|tst||j s|j ||gft|}t|}|jr|jjt|jtrGt|jtr-|jr!|jr|xj|z c_ |jt||z"|S)a Merge consecutive Messages of the same type. !!! note `ToolMessage` objects are not merged, as each has a distinct tool call id that can't be merged. Args: messages: Sequence Message-like objects to merge. chunk_separator: Specify the string to be inserted between message chunks. Returns: list of BaseMessages with consecutive runs of message types merged into single messages. By default, if two messages being merged both have string contents, the merged content is a concatenation of the two strings with a new-line separator. The separator inserted between message chunks can be controlled by specifying any string with `chunk_separator`. If at least one of the messages has a list of content blocks, the merged content is a list of content blocks. Example: ```python from langchain_core.messages import ( merge_message_runs, AIMessage, HumanMessage, SystemMessage, ToolCall, ) messages = [ SystemMessage("you're a good assistant."), HumanMessage( "what's your favorite color", id="foo", ), HumanMessage( "wait your favorite food", id="bar", ), AIMessage( "my favorite colo", tool_calls=[ ToolCall( name="blah_tool", args={"x": 2}, id="123", type="tool_call" ) ], id="baz", ), AIMessage( [{"type": "text", "text": "my favorite dish is lasagna"}], tool_calls=[ ToolCall( name="blah_tool", args={"x": -10}, id="456", type="tool_call", ) ], id="blur", ), ] merge_message_runs(messages) ``` ```python [ SystemMessage("you're a good assistant."), HumanMessage( "what's your favorite color\\n" "wait your favorite food", id="foo", ), AIMessage( [ "my favorite colo", {"type": "text", "text": "my favorite dish is lasagna"} ], tool_calls=[ ToolCall({ "name": "blah_tool", "args": {"x": 2}, "id": "123", "type": "tool_call" }), ToolCall({ "name": "blah_tool", "args": {"x": -10}, "id": "456", "type": "tool_call" }) ] id="baz" ), ] ``` N) rrrQr2r(rhrg _msg_to_chunkrsclearr{r6 _chunk_to_msg)rSrmergedr:last last_chunk curr_chunks r;merge_message_runsrasN  "8,H "FB%vzz|4 MM#  [ )C1P MM4+ &&t,J&s+J++,,224:--s3z1137&&&&""o5" MM- Z(?@ A%B& Mr=r)strategy allow_partialend_onstart_oninclude_system text_splitterc B |r|dk(r d} t| |r|dk(r d} t| t|}t|trB|tvr t| n2dj dtD} d|d| d} t| | t d r j} n~t rYtttj jjjt urd fd } n } nd t# d} t| t$rt|t&r |j(} n|r t+d |} nt,} |dk(rt/||| | |rd|Sd |S|dk(rt1||| ||||| Sd|d} t| )aM.Trim messages to be below a token count. `trim_messages` can be used to reduce the size of a chat history to a specified token or message count. In either case, if passing the trimmed chat history back into a chat model directly, the resulting chat history should usually satisfy the following properties: 1. The resulting chat history should be valid. Most chat models expect that chat history starts with either (1) a `HumanMessage` or (2) a `SystemMessage` followed by a `HumanMessage`. To achieve this, set `start_on='human'`. In addition, generally a `ToolMessage` can only appear after an `AIMessage` that involved a tool call. 2. It includes recent messages and drops old messages in the chat history. To achieve this set the `strategy='last'`. 3. Usually, the new chat history should include the `SystemMessage` if it was present in the original chat history since the `SystemMessage` includes special instructions to the chat model. The `SystemMessage` is almost always the first message in the history if present. To achieve this set the `include_system=True`. !!! note The examples below show how to configure `trim_messages` to achieve a behavior consistent with the above properties. Args: messages: Sequence of Message-like objects to trim. max_tokens: Max token count of trimmed messages. token_counter: Function or llm for counting tokens in a `BaseMessage` or a list of `BaseMessage`. If a `BaseLanguageModel` is passed in then `BaseLanguageModel.get_num_tokens_from_messages()` will be used. Set to `len` to count the number of **messages** in the chat history. You can also use string shortcuts for convenience: - `'approximate'`: Uses `count_tokens_approximately` for fast, approximate token counts. !!! note `count_tokens_approximately` (or the shortcut `'approximate'`) is recommended for using `trim_messages` on the hot path, where exact token counting is not necessary. strategy: Strategy for trimming. - `'first'`: Keep the first `<= n_count` tokens of the messages. - `'last'`: Keep the last `<= n_count` tokens of the messages. allow_partial: Whether to split a message if only part of the message can be included. If `strategy='last'` then the last partial contents of a message are included. If `strategy='first'` then the first partial contents of a message are included. end_on: The message type to end on. If specified then every message after the last occurrence of this type is ignored. If `strategy='last'` then this is done before we attempt to get the last `max_tokens`. If `strategy='first'` then this is done after we get the first `max_tokens`. Can be specified as string names (e.g. `'system'`, `'human'`, `'ai'`, ...) or as `BaseMessage` classes (e.g. `SystemMessage`, `HumanMessage`, `AIMessage`, ...). Can be a single type or a list of types. start_on: The message type to start on. Should only be specified if `strategy='last'`. If specified then every message before the first occurrence of this type is ignored. This is done after we trim the initial messages to the last `max_tokens`. Does not apply to a `SystemMessage` at index 0 if `include_system=True`. Can be specified as string names (e.g. `'system'`, `'human'`, `'ai'`, ...) or as `BaseMessage` classes (e.g. `SystemMessage`, `HumanMessage`, `AIMessage`, ...). Can be a single type or a list of types. include_system: Whether to keep the `SystemMessage` if there is one at index `0`. Should only be specified if `strategy="last"`. text_splitter: Function or `langchain_text_splitters.TextSplitter` for splitting the string contents of a message. Only used if `allow_partial=True`. If `strategy='last'` then the last split tokens from a partial message will be included. if `strategy='first'` then the first split tokens from a partial message will be included. Token splitter assumes that separators are kept, so that split contents can be directly concatenated to recreate the original text. Defaults to splitting on newlines. Returns: List of trimmed `BaseMessage`. Raises: ValueError: if two incompatible arguments are specified or an unrecognized `strategy` is specified. Example: Trim chat history based on token count, keeping the `SystemMessage` if present, and ensuring that the chat history starts with a `HumanMessage` (or a `SystemMessage` followed by a `HumanMessage`). ```python from langchain_core.messages import ( AIMessage, HumanMessage, BaseMessage, SystemMessage, trim_messages, ) messages = [ SystemMessage("you're a good assistant, you always respond with a joke."), HumanMessage("i wonder why it's called langchain"), AIMessage( 'Well, I guess they thought "WordRope" and "SentenceString" just ' "didn't have the same ring to it!" ), HumanMessage("and who is harrison chasing anyways"), AIMessage( "Hmmm let me think.\n\nWhy, he's probably chasing after the last " "cup of coffee in the office!" ), HumanMessage("what do you call a speechless parrot"), ] trim_messages( messages, max_tokens=45, strategy="last", token_counter=ChatOpenAI(model="gpt-4o"), # Most chat models expect that chat history starts with either: # (1) a HumanMessage or # (2) a SystemMessage followed by a HumanMessage start_on="human", # Usually, we want to keep the SystemMessage # if it's present in the original history. # The SystemMessage has special instructions for the model. include_system=True, allow_partial=False, ) ``` ```python [ SystemMessage( content="you're a good assistant, you always respond with a joke." ), HumanMessage(content="what do you call a speechless parrot"), ] ``` Trim chat history using approximate token counting with `'approximate'`: ```python trim_messages( messages, max_tokens=45, strategy="last", # Using the "approximate" shortcut for fast token counting token_counter="approximate", start_on="human", include_system=True, ) # This is equivalent to using `count_tokens_approximately` directly from langchain_core.messages.utils import count_tokens_approximately trim_messages( messages, max_tokens=45, strategy="last", token_counter=count_tokens_approximately, start_on="human", include_system=True, ) ``` Trim chat history based on the message count, keeping the `SystemMessage` if present, and ensuring that the chat history starts with a HumanMessage ( or a `SystemMessage` followed by a `HumanMessage`). trim_messages( messages, # When `len` is passed in as the token counter function, # max_tokens will count the number of messages in the chat history. max_tokens=4, strategy="last", # Passing in `len` as a token counter function will # count the number of messages in the chat history. token_counter=len, # Most chat models expect that chat history starts with either: # (1) a HumanMessage or # (2) a SystemMessage followed by a HumanMessage start_on="human", # Usually, we want to keep the SystemMessage # if it's present in the original history. # The SystemMessage has special instructions for the model. include_system=True, allow_partial=False, ) ```python [ SystemMessage( content="you're a good assistant, you always respond with a joke." ), HumanMessage(content="and who is harrison chasing anyways"), AIMessage( content="Hmmm let me think.\n\nWhy, he's probably chasing after " "the last cup of coffee in the office!" ), HumanMessage(content="what do you call a speechless parrot"), ] ``` Trim chat history using a custom token counter function that counts the number of tokens in each message. ```python messages = [ SystemMessage("This is a 4 token text. The full message is 10 tokens."), HumanMessage( "This is a 4 token text. The full message is 10 tokens.", id="first" ), AIMessage( [ {"type": "text", "text": "This is the FIRST 4 token block."}, {"type": "text", "text": "This is the SECOND 4 token block."}, ], id="second", ), HumanMessage( "This is a 4 token text. The full message is 10 tokens.", id="third" ), AIMessage( "This is a 4 token text. The full message is 10 tokens.", id="fourth", ), ] def dummy_token_counter(messages: list[BaseMessage]) -> int: # treat each message like it adds 3 default tokens at the beginning # of the message and at the end of the message. 3 + 4 + 3 = 10 tokens # per message. default_content_len = 4 default_msg_prefix_len = 3 default_msg_suffix_len = 3 count = 0 for msg in messages: if isinstance(msg.content, str): count += ( default_msg_prefix_len + default_content_len + default_msg_suffix_len ) if isinstance(msg.content, list): count += ( default_msg_prefix_len + len(msg.content) * default_content_len + default_msg_suffix_len ) return count ``` First 30 tokens, allowing partial messages: ```python trim_messages( messages, max_tokens=30, token_counter=dummy_token_counter, strategy="first", allow_partial=True, ) ``` ```python [ SystemMessage("This is a 4 token text. The full message is 10 tokens."), HumanMessage( "This is a 4 token text. The full message is 10 tokens.", id="first", ), AIMessage( [{"type": "text", "text": "This is the FIRST 4 token block."}], id="second", ), ] ``` firstz5start_on parameter is only valid with strategy='last'z;include_system parameter is only valid with strategy='last'z, c3(K|] }d|d yw)'Nr^).0keys r; z trim_messages..(s,"!C5 ,sz Invalid token_counter shortcut 'z'. Available shortcuts: r1get_num_tokens_from_messagesc,tfd|DS)Nc3.K|] }|ywrr^)rr:actual_token_counters r;rz.list_token_counter..?sI/4Is)sum)rSrs r;list_token_counterz)trim_messages..list_token_counter>sIIIIr=z'token_counter' expected to be a model that implements 'get_num_tokens_from_messages()' or a function. Received object of type rN max_tokens token_counterrpartial_strategyrr)rrrrrrrzUnrecognized strategy=z.. Supported strategies are 'last' and 'first'.rSSequence[BaseMessage]rint)rMrr2r6_TOKEN_COUNTER_SHORTCUTSrRr4rcallablenextiterinspect signature parametersvalues annotationrr0_HAS_LANGCHAIN_TEXT_SPLITTERSr. split_textr_default_text_splitter_first_max_tokens_last_max_tokens)rSrrrrrrrrr:available_shortcutsrtext_splitter_fnrs @r; trim_messagesrsl H'Eo(g-Ko"8,H-% 4 4#;M#J "&)),&>,# 3=/B((;'t7d6|jDsd?D cgc]} | vr|  c} x}rd|d| d4|d}t|| jdg| d<| dj)d7jd@jdjdAd;d<|s=| j)PjddBk(rvdCD cgc] } | vs|  c} x}rd|d| dD|d}t|tdEdFjdGrdHndIJ}| j=t?|g|KډjddLk(sdLvrFdLvrd|d| dM}t|| j)d t9j:dLd 8jddNk(sdNvrZdNvsd dNvrd|d| dO|d| dP }t|dNd }t|tr|d }| j)d |d jddQk(rydRD cgc] } | vs|  c} x}rd|d| dS|d}t|ddTvrdUdT}t|t3d(}| j)dd#d$dTd.|id7jddVvs|r| j)_d|d| dW}t||d k(r*t5dX| Dsdj'dY| D} | | dE<|j"r| dEs| r|j=|  P|j=| g|  f|r|dZS|Scc} wcc} wcc} wcc} wcc} wcc} wcc} wcc} w)[a Convert LangChain messages into OpenAI message dicts. Args: messages: Message-like object or iterable of objects whose contents are in OpenAI, Anthropic, Bedrock Converse, or VertexAI formats. text_format: How to format string or text block contents: - `'string'`: If a message has a string content, this is left as a string. If a message has content blocks that are all of type `'text'`, these are joined with a newline to make a single string. If a message has content blocks and at least one isn't of type `'text'`, then all blocks are left as dicts. - `'block'`: If a message has a string content, this is turned into a list with a single content block of type `'text'`. If a message has content blocks these are left as is. include_id: Whether to include message IDs in the openai messages, if they are present in the source messages. pass_through_unknown_blocks: Whether to include content blocks with unknown formats in the output. If `False`, an error is raised if an unknown content block is encountered. Raises: ValueError: if an unrecognized `text_format` is specified, or if a message content block is missing expected keys. Returns: The return type depends on the input type: - dict: If a single message-like object is passed in, a single OpenAI message dict is returned. - list[dict]: If a sequence of message-like objects are passed in, a list of OpenAI message dicts is returned. Example: ```python from langchain_core.messages import ( convert_to_openai_messages, AIMessage, SystemMessage, ToolMessage, ) messages = [ SystemMessage([{"type": "text", "text": "foo"}]), { "role": "user", "content": [ {"type": "text", "text": "whats in this"}, { "type": "image_url", "image_url": {"url": "data:image/png;base64,'/9j/4AAQSk'"}, }, ], }, AIMessage( "", tool_calls=[ { "name": "analyze", "args": {"baz": "buz"}, "id": "1", "type": "tool_call", } ], ), ToolMessage("foobar", tool_call_id="1", name="bar"), {"role": "assistant", "content": "thats nice"}, ] oai_messages = convert_to_openai_messages(messages) # -> [ # {'role': 'system', 'content': 'foo'}, # {'role': 'user', 'content': [{'type': 'text', 'text': 'whats in this'}, {'type': 'image_url', 'image_url': {'url': "data:image/png;base64,'/9j/4AAQSk'"}}]}, # {'role': 'assistant', 'tool_calls': [{'type': 'function', 'id': '1','function': {'name': 'analyze', 'arguments': '{"baz": "buz"}'}}], 'content': ''}, # {'role': 'tool', 'name': 'bar', 'content': 'foobar'}, # {'role': 'assistant', 'content': 'thats nice'} # ] ``` !!! version-added "Added in `langchain-core` 0.3.11" >blockrzUnrecognized text_format=z&, expected one of 'string' or 'block'.rLrqrOrefusalrrrtrrrN)r0rNc3hK|]*}t|txs|jddk(,ywr0rNN)r2r6rrr s r;rz-convert_to_openai_messages..s5-  uc " Aeii&76&A A- s02rFc3JK|]}t|tr|n|dywrNN)r2r6r s r;rz-convert_to_openai_messages..s* $E3/U6]B s!#r0)rNz'Unrecognized content block at messages[z ].content[z4] has 'type': 'text' but is missing expected key(s) z. Full content block: image_url)rz9] has 'type': 'image_url' but is missing expected key(s) )r0rfilefilenamezGenerating a fallback filename.LC_AUTOGENERATEDimagesource) media_typer0r\z>] has 'type': 'image' but 'source' is missing expected key(s) urlzdata:r;,r\)rformatz:] has key 'image', but 'image' is missing expected key(s) bytesz data:image/rz;base64,zX] has 'type': 'image' but does not have a 'source' or 'image' key. Full content block: file_data input_audior)rtrqinputz9] has 'type': 'tool_use', but is missing expected key(s) c34K|]}|ddk(yw)rtNr^rrwr s r;rz-convert_to_openai_messages..s&%"$5;6srrCrF ensure_asciirqrur0rtrCrKc3LK|]}|djdk(yw)rtcall_idN)rr s r;rz-convert_to_openai_messages..s+%"$599Y+??s!$)r&rqrur&ru tool_result)r{ tool_use_idz<] has 'type': 'tool_result', but is missing expected key(s) r{r(is_errorerrorsuccess)rrr)rrzJ] has 'type': 'json' but does not have a 'json' key. Full content block: guard_contentz;] has 'type': 'guard_content' but does not have a messages[z5]['guard_content']['text'] key. Full content block: media) mime_typer\z/] has 'type': 'media' but does not have key(s) r.zaOpenAI messages can only support text and image data. Received content block with media of type: >thinking reasoningz_] does not match OpenAI, Anthropic, Bedrock Converse, or VertexAI format. Full content block: c3,K|] }|ddk7ywr r^r s r;rz-convert_to_openai_messages..s3,1f '3sc3&K|] }|d ywrr^r s r;rz-convert_to_openai_messages..s#GeE&M#Gsr) rMr2rr3r6r enumerate_get_message_openai_rolerqrrO_convert_to_openai_tool_callsrPrr(rrrtr{allrRrQrrloggerinfo_bytes_to_b64_stranyrrdumpsrgr)rSrrrerr oai_messages is_singleirYoai_msg tool_messagesr{jrnmissingformatted_blockrr b64_imager: tool_messagerNr s @r;rrs v--*k^+QRo!Lx+tS)ABByB:"8,H)r; 7!9'!BC  <<%llGFO gy )g.@.@$A'BTBT$UGL !  $ $ ( ( 3!(!:!:9!EGI  g{ +&-&:&:GN # '**#JJGDM'83bG  -h&!//$*GOODE H $-  - * ii $__ G G%goo6H *5eS)NNFE#BCYYv&&0.7"J1E>1"JJwJ(()s*QC8>&i'@I )o-NNE&M5=#QRYYv&+5.<"O1"OOwO(()s*QC8>&i'@I )o-NN$/).{);+51&B5&IO'++F3v="o5&of.EE $EF>P/ ;NN?3ii'72w%7G!&8!44v4'E'"#RXA'7#,,-3j*:!F6N;K)M." #())G"444';'"#q~A'7#,,-3j " 5= e|aA% $3 (J 6DC!GE CC#h-/  hsm++T 2}//T/:HFH,,-I6)#'1A1A(B#C 1i( #+#3#3CR#8  !<8DS>!<8!<=K:$3::H1HC'+$   $4$>#'1A1A(B#C #C=333>D(**C0''H,,d38H8H%--E!%-$!%.599V3D3N$yy0 '222=H+D1 %28DS>%B"#v-"&x '<"=K [!1e!$[!1!~.(Au}%H$:$3::H1HC s AQw/q0A6Jq   DS>r=)rrrrct|}t|dk(rgS|r:tt|D]#}t|d|s|j #nd} |r+t|dkDrt |dt r |d} |dd}|ddd} |} | r|| g} td|| z } t| | |||rdnd|} | ddd}| r| g|}|S)NrrHrfrr) rrLrKrrr2r%maxr)rSrrrrrrrrOsystem_messagereversed_messagesremaining_tokens system_tokensreversed_resultr9s r;rrs H~H 8} s8}% A#HRL&9   N#h-!+ 8A; 0V!!AB<!2"%~&67 q*}"<='####0d OTrT "F *6* Mr=z/dict[type[BaseMessage], type[BaseMessageChunk]]_MSG_CHUNK_MAPc|jtvr)t|jdi|jdhStjD],\}}t ||s|di|jdhcSd|jdt tj }t|tj}t|)Nr0excludezUnrecognized message class . Supported classes are rr^) rhrc model_dumprkr2rkeysrrrrM)rYmsg_cls chunk_clsr:s r;rrsN*g//0X73E3Evh3E3WXX,224E gw 'Dw116(1CD DE &g&7&7%88P ##% & ' ) 1S1S TC S/r=c |jtvr*t|jdi|jhdStjD]-\}}t ||s|di|jhdcSd|jdt tj }t|tj}t|)N>r0rerdrez!Unrecognized message chunk class rgrr^) rh_CHUNK_MSG_MAPrhrkr2rrirrrrM)rlrkrjr:s r;rrs .(eoo. 'UV  -224 7 eY '""J#  ,EOO+<. s@aJq$,?@s)r2r6r0r)rYr_typesrt types_str types_typess r;rrsf"%#t5UG5E!8qZ3%78I8@5@@K <<9 $ H 7K(HH9s A*A*cJtj|jdS)Nzutf-8)base64 b64encodedecode)bytes_s r;r9r9s   F # * *7 33r=ct|tryt|tryt|tryt|trQ|j j dd}t|ts#dt|j}t||St|tryt|tr |jSd|jd }t|) Nr|ryrDr~rBz,Expected '__openai_role__' to be a str, got rCzUnknown BaseMessage type r1)r2rr"r(r%rPrr6r0r7r5r rrLrhrM)rYrLr:s r;r4r4s'9%'<(';''=)((,,->I$$@dATAT@UVCC.  '?+';'|| %g&7&7%8 :C S/r=c v|Dcgc])}d|d|dtj|dddd+c}Scc}w) NrCrtrqrxFr!r#r$)rr;)rOrws r;r5r5'sR$  D/!&)!ZZ &(9N    s.6g@g@)chars_per_tokenextra_tokens_per_message count_namecd}t|D];}d}t|jtr|t |jz }n#t |j}|t |z }t|t rIt|jts/|jr#t |j}|t |z }t|tr|t |jz }t|} |t | z }|jr|r|t |jz }|tj||z z }||z }>tj|S)aApproximate the total number of tokens in messages. The token count includes stringified message content, role, and (optionally) name. - For AI messages, the token count also includes stringified tool calls. - For tool messages, the token count also includes the tool call ID. Args: messages: List of messages to count tokens for. chars_per_token: Number of characters per token to use for the approximation. One token corresponds to ~4 chars for common English text. You can also specify `float` values for more fine-grained control. [See more here](https://platform.openai.com/tokenizer). extra_tokens_per_message: Number of extra tokens to add per message, e.g. special tokens, including beginning/end of message. You can also specify `float` values for more fine-grained control. [See more here](https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb). count_name: Whether to include message names in the count. Returns: Approximate number of tokens in the messages. Note: This is a simple approximation that may not match the exact token count used by specific models. For accurate counts, use model-specific tokenizers. Warning: This function does not currently support counting image tokens. !!! version-added "Added in `langchain-core` 0.3.46" gr)rr2r{r6rLreprrrrOr(rrr4rqmathceil) rSrrr token_countrY message_charsr{tool_calls_contentrLs r;count_tokens_approximatelyr5s3RK&x0"0 goos + S1 1M7??+G S\ )M w *w5""!%g&8&8!9  S!34 4M g{ + S!5!56 6M'0T" <rs#   88$   /.?@FGSI:LPP@86*5$(!   8 $$ i&' cg../0 SV_,-. sx0012!445 6  SV_,- . (8 99: ;!3+>#??@A #*<"==>?"C,@$AAB C$c.D&EEF G #*<"==> ? i01 3 H  ! B3#B3B3B3 B3  B3J@ 50$s)eCHo-3d38nDI #.2 XX X X X , X  XXXv9x6?66(t_4(KxE 2K2@ @) @*+/*.>B>B(,(,6:L?L(L( L < L < L&L&L4LLL^ }?}} }}D*0QUSW FJB?BB B'BB OBQBBDBBBJ s"T#s(^3 T(RL /7(, ,  "&     /7(, ,  "&    /7(, \M\,\ \ "& \  \J 9=QUq#qq6 q . q 6 q Oqqt  SWQU4#446 4 . 4  44Q4 O44p# ~%!)! C?$2#7#7#9:41a!Q$:  *9 I I FI I4* "!&) O"1O"O"$ O"  O"  O"f0 -e:*$)!*v4;s;M>0 N >N N