import anthropic

from langchain.agents import create_agent
from langchain_anthropic import ChatAnthropic
from langchain_core.messages import HumanMessage

model = ChatAnthropic(
    model="claude-sonnet-4-6",
    betas=["compact-2026-01-12"],
    max_tokens=8000,
    context_management={
        "edits": [
            {
                "type": "compact_20260112",
                "trigger": {"type": "input_tokens", "value": 50000},
            }
        ]
    },
)

agent = create_agent(
    model=model,
)

client = anthropic.Anthropic()

SYSTEM_MESSAGE = "You are a helpful assistant that answers questions thoroughly and in great detail."

FILLER_PARAGRAPH = FILLER_PARAGRAPH = (
    "Brand24 is a social listening and media monitoring platform that helps businesses track "
    "online mentions of their brand, products, and competitors across social media, news sites, "
    "blogs, forums, podcasts, and more. The platform provides real-time alerts, sentiment analysis, "
    "influencer identification, and detailed analytics to help companies understand their online "
    "presence and make data-driven marketing decisions. With Brand24, users can monitor keywords, "
    "hashtags, and phrases across millions of online sources, gaining valuable insights into "
    "customer opinions, industry trends, and competitive landscape. The tool offers features like "
    "automated reports, discussion volume charts, mention analytics, and integration with popular "
    "business tools. Companies use Brand24 to protect their reputation, measure campaign "
    "effectiveness, find sales leads, and improve customer service by responding quickly to "
    "online conversations about their brand."
) * 5

N = 31

history = []
for i in range(N):
    history.append({"role": "user", "content": f"Question {i+1}: Tell me more about topic number {i+1}. {FILLER_PARAGRAPH}"})
    history.append({"role": "assistant", "content": f"Answer {i+1}: Here is a detailed response about topic {i+1}. {FILLER_PARAGRAPH}"})
history.append({"role": "user", "content": "Hi, my name is John"})

token_count = client.messages.count_tokens(
    model="claude-sonnet-4-6",
    system=SYSTEM_MESSAGE,
    messages=history,
)
assert token_count.input_tokens > 50_000, f"Expected >50k tokens, got {token_count.input_tokens:,}"
assert token_count.input_tokens < 60_000, f"Expected <60k tokens, got {token_count.input_tokens:,}"

# The first request triggers compaction. With print_mode="updates" everything is fine
result = agent.invoke({"messages": history}, print_mode="updates")
result2 = agent.invoke({"messages": result["messages"] + [HumanMessage(content="What is my name?")]}, print_mode="updates")

# Now we make the first request with print_mode="messages". The second request produce an error
result = agent.invoke({"messages": history}, print_mode="messages")  # Works fine
result2 = agent.invoke({"messages": result["messages"] + [HumanMessage(content="What is my name?")]}, print_mode=("messages"))  # Produces an error

---------------------------------------------------------------------------
BadRequestError                           Traceback (most recent call last)
Cell In[20], line 3
      1 from langchain_core.messages import HumanMessage
      2 
----> 3 result2 = agent.invoke({"messages": result["messages"] + [HumanMessage(content="What is my name?")]}, print_mode=("messages"))

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/pregel/main.py:3365, in Pregel.invoke(self, input, config, context, stream_mode, print_mode, output_keys, interrupt_before, interrupt_after, durability, version, **kwargs)
   3362             chunks.append(chunk)
   3363 else:
   3364     # v1: collect interrupts from updates stream
-> 3365     for chunk in self.stream(
   3366         input,
   3367         config,
   3368         context=context,
   3369         stream_mode=(
   3370             ["updates", "values"] if stream_mode == "values" else stream_mode
   3371         ),
   3372         print_mode=print_mode,
   3373         output_keys=output_keys,
   3374         interrupt_before=interrupt_before,
   3375         interrupt_after=interrupt_after,
   3376         durability=durability,
   3377         **kwargs,
   3378     ):
   3379         if stream_mode == "values":
   3380             if len(chunk) == 2:

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/pregel/main.py:2759, in Pregel.stream(self, input, config, context, stream_mode, print_mode, output_keys, interrupt_before, interrupt_after, durability, subgraphs, debug, version, **kwargs)
   2757 for task in loop.match_cached_writes():
   2758     loop.output_writes(task.id, task.writes, cached=True)
-> 2759 for _ in runner.tick(
   2760     [t for t in loop.tasks.values() if not t.writes],
   2761     timeout=self.step_timeout,
   2762     get_waiter=get_waiter,
   2763     schedule_task=loop.accept_push,
   2764 ):
   2765     # emit output
   2766     yield from _output(
   2767         stream_mode,
   2768         print_mode,
   (...)   2774         _state_mapper,
   2775     )
   2776 loop.after_tick()

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/pregel/_runner.py:167, in PregelRunner.tick(self, tasks, reraise, timeout, retry_policy, get_waiter, schedule_task)
    165 t = tasks[0]
    166 try:
--> 167     run_with_retry(
    168         t,
    169         retry_policy,
    170         configurable={
    171             CONFIG_KEY_CALL: partial(
    172                 _call,
    173                 weakref.ref(t),
    174                 retry_policy=retry_policy,
    175                 futures=weakref.ref(futures),
    176                 schedule_task=schedule_task,
    177                 submit=self.submit,
    178             ),
    179         },
    180     )
    181     self.commit(t, None)
    182 except Exception as exc:

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/pregel/_retry.py:126, in run_with_retry(task, retry_policy, configurable)
    124     task.writes.clear()
    125     # run the task
--> 126     return task.proc.invoke(task.input, config)
    127 except ParentCommand as exc:
    128     ns: str = config[CONF][CONFIG_KEY_CHECKPOINT_NS]

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/_internal/_runnable.py:656, in RunnableSeq.invoke(self, input, config, **kwargs)
    654     # run in context
    655     with set_config_context(config, run) as context:
--> 656         input = context.run(step.invoke, input, config, **kwargs)
    657 else:
    658     input = step.invoke(input, config)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langgraph/_internal/_runnable.py:400, in RunnableCallable.invoke(self, input, config, **kwargs)
    398         run_manager.on_chain_end(ret)
    399 else:
--> 400     ret = self.func(*args, **kwargs)
    401 if self.recurse and isinstance(ret, Runnable):
    402     return ret.invoke(input, config)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain/agents/factory.py:1310, in create_agent.<locals>.model_node(state, runtime)
   1298 request = ModelRequest(
   1299     model=model,
   1300     tools=default_tools,
   (...)   1306     runtime=runtime,
   1307 )
   1309 if wrap_model_call_handler is None:
-> 1310     model_response = _execute_model_sync(request)
   1311     return _build_commands(model_response)
   1313 result = wrap_model_call_handler(request, _execute_model_sync)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain/agents/factory.py:1282, in create_agent.<locals>._execute_model_sync(request)
   1279 if request.system_message:
   1280     messages = [request.system_message, *messages]
-> 1282 output = model_.invoke(messages)
   1283 if name:
   1284     output.name = name

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_core/runnables/base.py:5695, in RunnableBindingBase.invoke(self, input, config, **kwargs)
   5688 @override
   5689 def invoke(
   5690     self,
   (...)   5693     **kwargs: Any | None,
   5694 ) -> Output:
-> 5695     return self.bound.invoke(
   5696         input,
   5697         self._merge_configs(config),
   5698         **{**self.kwargs, **kwargs},
   5699     )

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py:455, in BaseChatModel.invoke(self, input, config, stop, **kwargs)
    441 @override
    442 def invoke(
    443     self,
   (...)    448     **kwargs: Any,
    449 ) -> AIMessage:
    450     config = ensure_config(config)
    451     return cast(
    452         "AIMessage",
    453         cast(
    454             "ChatGeneration",
--> 455             self.generate_prompt(
    456                 [self._convert_input(input)],
    457                 stop=stop,
    458                 callbacks=config.get("callbacks"),
    459                 tags=config.get("tags"),
    460                 metadata=config.get("metadata"),
    461                 run_name=config.get("run_name"),
    462                 run_id=config.pop("run_id", None),
    463                 **kwargs,
    464             ).generations[0][0],
    465         ).message,
    466     )

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py:1198, in BaseChatModel.generate_prompt(self, prompts, stop, callbacks, **kwargs)
   1189 @override
   1190 def generate_prompt(
   1191     self,
   (...)   1195     **kwargs: Any,
   1196 ) -> LLMResult:
   1197     prompt_messages = [p.to_messages() for p in prompts]
-> 1198     return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py:1005, in BaseChatModel.generate(self, messages, stop, callbacks, tags, metadata, run_name, run_id, **kwargs)
   1002 for i, m in enumerate(input_messages):
   1003     try:
   1004         results.append(
-> 1005             self._generate_with_cache(
   1006                 m,
   1007                 stop=stop,
   1008                 run_manager=run_managers[i] if run_managers else None,
   1009                 **kwargs,
   1010             )
   1011         )
   1012     except BaseException as e:
   1013         if run_managers:

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py:1310, in BaseChatModel._generate_with_cache(self, messages, stop, run_manager, **kwargs)
   1308     result = generate_from_stream(iter(chunks))
   1309 elif inspect.signature(self._generate).parameters.get("run_manager"):
-> 1310     result = self._generate(
   1311         messages, stop=stop, run_manager=run_manager, **kwargs
   1312     )
   1313 else:
   1314     result = self._generate(messages, stop=stop, **kwargs)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_anthropic/chat_models.py:1566, in ChatAnthropic._generate(self, messages, stop, run_manager, **kwargs)
   1564     data = self._create(payload)
   1565 except anthropic.BadRequestError as e:
-> 1566     _handle_anthropic_bad_request(e)
   1567 return self._format_output(data, **kwargs)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_anthropic/chat_models.py:1564, in ChatAnthropic._generate(self, messages, stop, run_manager, **kwargs)
   1562 payload = self._get_request_payload(messages, stop=stop, **kwargs)
   1563 try:
-> 1564     data = self._create(payload)
   1565 except anthropic.BadRequestError as e:
   1566     _handle_anthropic_bad_request(e)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/langchain_anthropic/chat_models.py:1228, in ChatAnthropic._create(self, payload)
   1226 def _create(self, payload: dict) -> Any:
   1227     if "betas" in payload:
-> 1228         return self._client.beta.messages.create(**payload)
   1229     return self._client.messages.create(**payload)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/anthropic/_utils/_utils.py:283, in required_args.<locals>.inner.<locals>.wrapper(*args, **kwargs)
    281             msg = f"Missing required argument: {quote(missing[0])}"
    282     raise TypeError(msg)
--> 283 return func(*args, **kwargs)

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/anthropic/resources/beta/messages/messages.py:1100, in Messages.create(self, max_tokens, messages, model, cache_control, container, context_management, inference_geo, mcp_servers, metadata, output_config, output_format, service_tier, speed, stop_sequences, stream, system, temperature, thinking, tool_choice, tools, top_k, top_p, user_profile_id, betas, extra_headers, extra_query, extra_body, timeout)
   1093 merged_output_config = _merge_output_configs(output_config, output_format)
   1095 extra_headers = {
   1096     **strip_not_given({"anthropic-beta": ",".join(str(e) for e in betas) if is_given(betas) else not_given}),
   1097     **_stainless_helper_header(tools, messages),
   1098     **(extra_headers or {}),
   1099 }
-> 1100 return self._post(
   1101     "/v1/messages?beta=true",
   1102     body=maybe_transform(
   1103         {
   1104             "max_tokens": max_tokens,
   1105             "messages": messages,
   1106             "model": model,
   1107             "cache_control": cache_control,
   1108             "container": container,
   1109             "context_management": context_management,
   1110             "inference_geo": inference_geo,
   1111             "mcp_servers": mcp_servers,
   1112             "metadata": metadata,
   1113             "output_config": merged_output_config,
   1114             "output_format": omit,
   1115             "service_tier": service_tier,
   1116             "speed": speed,
   1117             "stop_sequences": stop_sequences,
   1118             "stream": stream,
   1119             "system": system,
   1120             "temperature": temperature,
   1121             "thinking": thinking,
   1122             "tool_choice": tool_choice,
   1123             "tools": tools,
   1124             "top_k": top_k,
   1125             "top_p": top_p,
   1126             "user_profile_id": user_profile_id,
   1127         },
   1128         message_create_params.MessageCreateParamsStreaming
   1129         if stream
   1130         else message_create_params.MessageCreateParamsNonStreaming,
   1131     ),
   1132     options=make_request_options(
   1133         extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
   1134     ),
   1135     cast_to=BetaMessage,
   1136     stream=stream or False,
   1137     stream_cls=Stream[BetaRawMessageStreamEvent],
   1138 )

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/anthropic/_base_client.py:1368, in SyncAPIClient.post(self, path, cast_to, body, content, options, files, stream, stream_cls)
   1359     warnings.warn(
   1360         "Passing raw bytes as `body` is deprecated and will be removed in a future version. "
   1361         "Please pass raw bytes via the `content` parameter instead.",
   1362         DeprecationWarning,
   1363         stacklevel=2,
   1364     )
   1365 opts = FinalRequestOptions.construct(
   1366     method="post", url=path, json_data=body, content=content, files=to_httpx_files(files), **options
   1367 )
-> 1368 return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))

File ~/ai24/dist/export/python/virtualenvs/langchain_compaction_test/lib/python3.12/site-packages/anthropic/_base_client.py:1141, in SyncAPIClient.request(self, cast_to, options, stream, stream_cls)
   1138             err.response.read()
   1140         log.debug("Re-raising status error")
-> 1141         raise self._make_status_error_from_response(err.response) from None
   1143     break
   1145 assert response is not None, "could not resolve response (should never happen)"

BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages.63.content.0.compaction.index: Extra inputs are not permitted'}, 'request_id': 'req_011CaMEooKTSiAAWiBiYo2yL'}
During task with name 'model' and id '62a7431b-3f32-e2e5-fbdf-1040c1bf2015'

                    elif block["type"] == "compaction":
                        content.append(
                            {
                                k: v
                                for k, v in block.items()
                                if k in ("type", "content", "cache_control")
                            },
                        )