=iI( S r SSKrSSKJrJr SSKJrJr SSKJ r J
r
J r J r J
r
SSKJr SSKJrJr SSKJr SS KJr SS
KJrJr SS KJrJ r \\\ /\ 4 r S r S
r Sr!Sr"Sr#Sr$ " S S\5 r%g)zSummarization middleware. N)CallableIterable)Anycast) AIMessage
AnyMessageMessageLikeRepresentation
RemoveMessage ToolMessage HumanMessage)count_tokens_approximately
trim_messages)REMOVE_ALL_MESSAGES)Runtime)AgentMiddleware
AgentState)
BaseChatModelinit_chat_modela
Context Extraction Assistant
Your sole objective in this task is to extract the highest quality/most relevant context from the conversation history below.
You're nearing the total number of input tokens you can accept, so you must extract the highest quality/most relevant pieces of information from your conversation history.
This context will then overwrite the conversation history presented below. Because of this, ensure the context you extract is only the most important information to your overall goal.
The conversation history below will be replaced with the context you extract in this step. Because of this, you must do your very best to extract and record all of the most important context from the conversation history.
You want to ensure that you don't repeat any actions you've already completed, so the context you extract from the conversation history should be focused on the most important information to your overall goal.
The user will message you with the full message history you'll be extracting context from, to then replace. Carefully read over it all, and think deeply about what information is most important to your overall goal that should be saved:
With all of this in mind, please carefully read over the entire conversation history, and extract the most important and relevant context to replace it so that you can free up space in the conversation history.
Respond ONLY with the extracted context. Do not include any additional information, or text before or after the extracted context.
Messages to summarize:
{messages}
z!## Previous conversation summary: i c ^ \ rS rSrSrS\\\\4S\ \
- S\ S- S\ S\ S\ S \ S
S4U 4S jjjr
S \S
\S
\\ \4 S- 4S jrS\ S
\\ 4S jrS\\ S
S4S jrS\\ S\ S
\\\ \\ 4 4S jrS\\ S
\ 4S jrS\\ S\ S
\4S jr S\S
\4S jr S\ S
\ \ 4S jr S\\ S \ S\ S \ \ S
\4
S jr!S \\ S
\ 4S jr"S\\ S
\\ 4S! jr#S"r$U =r%$ )#SummarizationMiddleware> a+ Summarizes conversation history when token limits are approached.
This middleware monitors message token counts and automatically summarizes older
messages when a threshold is reached, preserving recent messages and maintaining
context continuity by ensuring AI/Tool message pairs remain together.
Nmodelmax_tokens_before_summarymessages_to_keep
token_countersummary_promptsummary_prefixreturnc > [ TU ] 5 [ U[ 5 ( a [ U5 nXl X l X0l X@l XPl X`l
g)a Initialize the summarization middleware.
Args:
model: The language model to use for generating summaries.
max_tokens_before_summary: Token threshold to trigger summarization.
If `None`, summarization is disabled.
messages_to_keep: Number of recent messages to preserve after summarization.
token_counter: Function to count tokens in messages.
summary_prompt: Prompt template for generating summaries.
summary_prefix: Prefix added to system message when including summary.
N) super__init__
isinstancestrr r r r r r r! )selfr r r r r r! __class__s y/home/dmtnaga/Documents/work/airagagent/rag_env/lib/python3.13/site-packages/langchain/agents/middleware/summarization.pyr% SummarizationMiddleware.__init__F sJ ( eS ! !#E*E
)B& 0 * , , stateruntimec L US nU R U5 U R U5 nU R b X@R : a gU R U5 nUS:: a gU R X55 u pgU R U5 nU R
U5 n S[ [ S9/U QUQ0$ )zOProcess messages before model invocation, potentially triggering summarization.messagesNr )id) _ensure_message_idsr r _find_safe_cutoff_partition_messages_create_summary_build_new_messagesr
r )
r( r- r. r0 total_tokens cutoff_indexmessages_to_summarizepreserved_messagessummary new_messagess
r* before_model$SummarizationMiddleware.before_modelf s $ *))(3 * * 6 = ==--h7 1 484L4LX4d1 &&'<=//8
!45 $
r, r; c [ SU 3S9/$ )Nz0Here is a summary of the conversation to date:
)contentr )r( r; s r* r6 +SummarizationMiddleware._build_new_messages s #UV]U^!_ `
r, r0 c z U H5 nUR b M [ [ R " 5 5 Ul M7 g)zAEnsure all messages have unique IDs for the add_messages reducer.N)r1 r' uuiduuid4)r( r0 msgs r* r2 +SummarizationMiddleware._ensure_message_ids s' Cvv~ TZZ\* r, conversation_messagesr8 c USU nXS nX44$ )zAPartition messages into those to summarize and those to preserve.N )r( rG r8 r9 r: s r* r4 +SummarizationMiddleware._partition_messages s% !6m| D 2= A$88r, c [ U5 U R :: a g[ U5 U R -
n[ USS5 H nU R X5 ( d M Us $ g)zFind safe cutoff point that preserves AI/Tool message pairs.
Returns the index where messages can be safely cut without separating
related AI and Tool messages. Returns 0 if no safe cutoff is found.
r )lenr range_is_safe_cutoff_point)r( r0
target_cutoffis r* r3 )SummarizationMiddleware._find_safe_cutoff s[ x=D 1 1 1H
(=(==
}b"-A))(66 . r, c F U[ U5 : a g[ SU[ -
5 n[ [ U5 U[ - 5 n[ X45 HS nU R X 5 ( d M U R
[ SX 5 5 nU R XX&5 ( d MS g g)z?Check if cutting at index would separate AI/Tool message pairs.Tr r F) rM max _SEARCH_RANGE_FOR_TOOL_PAIRSminrN _has_tool_calls_extract_tool_call_idsr _cutoff_separates_tool_pair)r( r0 r8 search_start
search_endrQ
tool_call_idss r* rO -SummarizationMiddleware._is_safe_cutoff_point s 3x= (1l-I IJ X 7S(ST
|0A'' 44 7 7[(+8V WM//\YY
1 r, messagec n [ U[ 5 =( a [ US5 =( a UR $ )z2Check if message is an AI message with tool calls.
tool_calls)r& r hasattrr` )r( r^ s r* rW 'SummarizationMiddleware._has_tool_calls s+
w * dww /M dRYRdRd
r,
ai_messagec [ 5 nUR HL n[ U[ 5 ( a UR S5 O [ USS5 nUc M; UR
U5 MN U$ )z)Extract tool call IDs from an AI message.r1 N)setr` r& dictgetgetattradd)r( rc r\ tccall_ids r* rX .SummarizationMiddleware._extract_tool_call_ids sZ
''B&0T&:&:bffTlDRV@WG" !!'* ( r, ai_message_indexr\ c [ US- [ U5 5 H? nX n[ U[ 5 ( d M UR U; d M0 X#: nXS: nXx:w d M? g g)zMCheck if cutoff separates an AI message from its corresponding tool messages. TF)rN rM r& r tool_call_id) r( r0 rm r8 r\ jr^ ai_before_cutofftool_before_cutoffs r* rY 3SummarizationMiddleware._cutoff_separates_tool_pair s` '!+S];A kG';//G4H4HM4Y#3#B %&%5"#9
< r, r9 c , U( d gU R U5 nU( d g U R R U R R US95 n[ SUR 5 R 5 $ ! [ a nSU< 3s SnA$ SnAff = f)z(Generate summary for the given messages.z!No previous conversation history.z0Previous conversation was too long to summarize.)r0 r' zError generating summary: N) _trim_messages_for_summaryr invoker formatr r@ strip Exception)r( r9 trimmed_messagesresponsees r* r5 'SummarizationMiddleware._create_summary s $6 ::;PQ E 6zz(()<)<)C)CM])C)^_Hx / /0668 8 6/u5 5 6s AA9 9
BBBBc
r [ U[ U R SSSSS9$ ! [ a
U[ * S s $ f = f)z6Trim messages to fit within summary generation limits.humanlastT)
max_tokensr start_onstrategy
allow_partialinclude_systemN)r _DEFAULT_TRIM_TOKEN_LIMITr rz _DEFAULT_FALLBACK_MESSAGE_COUNT)r( r0 s r* rv 2SummarizationMiddleware._trim_messages_for_summary sR ? 4 " 0 0 " #
? < < => > ?s 66)r r r r! r r )&__name__
__module__ __qualname____firstlineno____doc___DEFAULT_MESSAGES_TO_KEEPr DEFAULT_SUMMARY_PROMPTSUMMARY_PREFIXr' r int TokenCounterr% r r rf r r= listr
r6 r r2 tupler4 r3 boolrO rW r re rX rY r5 rv __static_attributes__
__classcell__)r) s @r* r r > s 15 9&@ 4 , -]" - $': - -
$ -
- -
- -@
*
w
4S>TXCX
<
3
4 3E
+D,< + + 9 #J / 9 9
tJ j!11 2 9$z*: s "d:.> c VZ $
z
d
s3x z"
3x
"
6T*5E
6#
6
?4
3C
?ZHX
?
?r, r )&r rC collections.abcr r typingr r langchain_core.messagesr r r r
r langchain_core.messages.humanr
langchain_core.messages.utilsr r langgraph.graph.messager langgraph.runtimer !langchain.agents.middleware.typesr r langchain.chat_modelsr r r r r r r r r rU r rI r, r* r s . 7 S & I @";<=sBC 8 5 "$ {?o {?r,