Claudette’s source

This is the ‘literate’ source code for Claudette. You can view the fully rendered version of the notebook here, or you can clone the git repo and run the interactive notebook in Jupyter. The notebook is converted the Python module claudette/core.py using nbdev. The goal of this source code is to both create the Python module, and also to teach the reader how it is created, without assuming much existing knowledge about Claude’s API.

Most of the time you’ll see that we write some source code first, and then a description or discussion of it afterwards.

Setup

import os
# os.environ['ANTHROPIC_LOG'] = 'debug'

To print every HTTP request and response in full, uncomment the above line. This functionality is provided by Anthropic’s SDK.

from anthropic.types import Model
from claudette.text_editor import *
from typing import get_args
from datetime import datetime
from pprint import pprint
from IPython.display import Image
from cachy import enable_cachy

import warnings

enable_cachy()

warnings.filterwarnings("ignore", message="Pydantic serializer warnings")

Tip

If you’re reading the rendered version of this notebook, you’ll see an “Exported source” collapsible widget below. If you’re reading the source notebook directly, you’ll see #| exports at the top of the cell. These show that this piece of code will be exported into the python module that this notebook creates. No other code will be included – any other code in this notebook is just for demonstration, documentation, and testing.

You can toggle expanding/collapsing the source code of all exported sections by using the </> Code menu in the top right of the rendered notebook page.

Exported source

model_types = {
    # Anthropic
    'claude-opus-4-1-20250805': 'opus',
    'claude-sonnet-4-5': 'sonnet',
    'claude-haiku-4-5': 'haiku',
    'claude-opus-4-20250514': 'opus-4',
    'claude-3-opus-20240229': 'opus-3',
    'claude-sonnet-4-20250514': 'sonnet-4',
    'claude-3-7-sonnet-20250219': 'sonnet-3-7',
    'claude-3-5-sonnet-20241022': 'sonnet-3-5',
    'claude-3-haiku-20240307': 'haiku-3',
    'claude-3-5-haiku-20241022': 'haiku-3-5',
    # AWS
    'anthropic.claude-opus-4-1-20250805-v1:0': 'opus',
    'anthropic.claude-3-5-sonnet-20241022-v2:0': 'sonnet',
    'anthropic.claude-3-opus-20240229-v1:0': 'opus-3',
    'anthropic.claude-3-sonnet-20240229-v1:0': 'sonnet',
    'anthropic.claude-3-haiku-20240307-v1:0': 'haiku',
    # Google
    'claude-opus-4-1@20250805': 'opus',
    'claude-3-5-sonnet-v2@20241022': 'sonnet',
    'claude-3-opus@20240229': 'opus-3',
    'claude-3-sonnet@20240229': 'sonnet',
    'claude-3-haiku@20240307': 'haiku',
}

all_models = list(model_types)

models

['claude-opus-4-1-20250805',
 'claude-sonnet-4-5',
 'claude-haiku-4-5',
 'claude-opus-4-20250514',
 'claude-3-opus-20240229',
 'claude-sonnet-4-20250514',
 'claude-3-7-sonnet-20250219',
 'claude-3-5-sonnet-20241022']

Exported source

text_only_models = ('claude-3-5-haiku-20241022',)

Exported source

has_streaming_models = set(all_models)
has_system_prompt_models = set(all_models)
has_temperature_models = set(all_models)
has_extended_thinking_models = {
    'claude-opus-4-1-20250805', 'claude-opus-4-20250514', 'claude-sonnet-4-20250514', 'claude-3-7-sonnet-20250219',
    'sonnet-4-5', 'haiku-4-5'
}

has_extended_thinking_models

{'claude-3-7-sonnet-20250219',
 'claude-opus-4-1-20250805',
 'claude-opus-4-20250514',
 'claude-sonnet-4-20250514',
 'haiku-4-5',
 'sonnet-4-5'}

source

can_use_extended_thinking

 can_use_extended_thinking (m)

Exported source

def can_stream(m): return m in has_streaming_models
def can_set_system_prompt(m): return m in has_system_prompt_models
def can_set_temperature(m): return m in has_temperature_models
def can_use_extended_thinking(m): return m in has_extended_thinking_models

source

can_set_temperature

 can_set_temperature (m)

source

can_set_system_prompt

 can_set_system_prompt (m)

source

can_stream

 can_stream (m)

We include these functions to provide a uniform library interface with cosette since openai models such as o1 do not have many of these capabilities.

assert can_stream('claude-3-5-sonnet-20241022') and can_set_system_prompt('claude-3-5-sonnet-20241022') and can_set_temperature('claude-3-5-sonnet-20241022')

These are the current versions and prices of Anthropic’s models at the time of writing.

model = models[1]; model

'claude-sonnet-4-5'

For examples, we’ll use the latest Sonnet, since it’s awesome.

Antropic SDK

cli = Anthropic()

This is what Anthropic’s SDK provides for interacting with Python. To use it, pass it a list of messages, with content and a role. The roles should alternate between user and assistant.

Tip

After the code below you’ll see an indented section with an orange vertical line on the left. This is used to show the result of running the code above. Because the code is running in a Jupyter Notebook, we don’t have to use print to display results, we can just type the expression directly, as we do with r here.

m = {'role': 'user', 'content': "I'm Jeremy"}
r = cli.messages.create(messages=[m], model=model, max_tokens=100)
r

Hello Jeremy! Nice to meet you. How can I help you today?

id: msg_01M2o3SENPk6JXLarGYZffo4
content: [{'citations': None, 'text': 'Hello Jeremy! Nice to meet you. How can I help you today?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 10, 'output_tokens': 18, 'server_tool_use': None, 'service_tier': 'standard'}

Formatting output

That output is pretty long and hard to read, so let’s clean it up. We’ll start by pulling out the Content part of the message. To do that, we’re going to write our first function which will be included to the claudette/core.py module.

Tip

This is the first exported public function or class we’re creating (the previous export was of a variable). In the rendered version of the notebook for these you’ll see 4 things, in this order (unless the symbol starts with a single _, which indicates it’s private):

The signature (with the symbol name as a heading, with a horizontal rule above)
A table of paramater docs (if provided)
The doc string (in italics).
The source code (in a collapsible “Exported source” block)

After that, we generally provide a bit more detail on what we’ve created, and why, along with a sample usage.

source

find_block

 find_block (r:collections.abc.Mapping, blk_type:type|str=<class
             'anthropic.types.text_block.TextBlock'>)

Find the first block of type blk_type in r.content.

	Type	Default	Details
r	Mapping		The message to look in
blk_type	type \| str	TextBlock	The type of block to find

Exported source

def _type(x):
    try: return x.type
    except AttributeError: return x.get('type')

def find_block(r:abc.Mapping, # The message to look in
               blk_type:type|str=TextBlock  # The type of block to find
              ):
    "Find the first block of type `blk_type` in `r.content`."
    f = (lambda x:_type(x)==blk_type) if isinstance(blk_type,str) else (lambda x:isinstance(x,blk_type))
    return first(o for o in r.content if f(o))

This makes it easier to grab the needed parts of Claude’s responses, which can include multiple pieces of content. By default, we look for the first text block. That will generally have the content we want to display.

find_block(r)

TextBlock(citations=None, text='Hello Jeremy! Nice to meet you. How can I help you today?', type='text')

def contents(r):
    "Helper to get the contents from Claude response `r`."
    blk = find_block(r)
    if not blk and r.content: blk = r.content[0]
    if hasattr(blk,'text'): return blk.text.strip()
    elif hasattr(blk,'content'): return blk.content.strip()
    elif hasattr(blk,'source'): return f'*Media Type - {blk.type}*'
    return str(blk)

For display purposes, we often just want to show the text itself.

contents(r)

'Hello Jeremy! Nice to meet you. How can I help you today?'

Exported source

@patch
def _repr_markdown_(self:(Message)):
    det = '\n- '.join(f'{k}: `{v}`' for k,v in self.model_dump().items())
    cts = re.sub(r'\$', '&#36;', contents(self))  # escape `$` for jupyter latex
    return f"""{cts}

<details>

- {det}

</details>"""

Jupyter looks for a _repr_markdown_ method in displayed objects; we add this in order to display just the content text, and collapse full details into a hideable section. Note that patch is from fastcore, and is used to add (or replace) functionality in an existing class. We pass the class(es) that we want to patch as type annotations to self. In this case, _repr_markdown_ is being added to Anthropic’s Message class, so when we display the message now we just see the contents, and the details are hidden away in a collapsible details block.

Hello Jeremy! Nice to meet you. How can I help you today?

id: msg_01M2o3SENPk6JXLarGYZffo4
content: [{'citations': None, 'text': 'Hello Jeremy! Nice to meet you. How can I help you today?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 10, 'output_tokens': 18, 'server_tool_use': None, 'service_tier': 'standard'}

One key part of the response is the usage key, which tells us how many tokens we used by returning a Usage object.

We’ll add some helpers to make things a bit cleaner for creating and formatting these objects.

r.usage

In: 10; Out: 18; Cache create: 0; Cache read: 0; Total Tokens: 28; Search: 0

source

server_tool_usage

 server_tool_usage (web_search_requests=0)

Little helper to create a server tool usage object

Exported source

def server_tool_usage(web_search_requests=0):
    'Little helper to create a server tool usage object'
    return ServerToolUsage(web_search_requests=web_search_requests)

source

usage

 usage (inp=0, out=0, cache_create=0, cache_read=0,
        server_tool_use=ServerToolUsage(web_search_requests=0))

Slightly more concise version of Usage.

	Type	Default	Details
inp	int	0	input tokens
out	int	0	Output tokens
cache_create	int	0	Cache creation tokens
cache_read	int	0	Cache read tokens
server_tool_use	ServerToolUsage	ServerToolUsage(web_search_requests=0)	server tool use

Exported source

def usage(inp=0, # input tokens
          out=0,  # Output tokens
          cache_create=0, # Cache creation tokens
          cache_read=0, # Cache read tokens
          server_tool_use=server_tool_usage() # server tool use
         ):
    'Slightly more concise version of `Usage`.'
    return Usage(input_tokens=inp, output_tokens=out, cache_creation_input_tokens=cache_create,
                 cache_read_input_tokens=cache_read, server_tool_use=server_tool_use)

The constructor provided by Anthropic is rather verbose, so we clean it up a bit, using a lowercase version of the name.

usage(5)

In: 5; Out: 0; Cache create: 0; Cache read: 0; Total Tokens: 5; Search: 0

source

Usage.total

 Usage.total ()

Exported source

def _dgetattr(o,s,d): 
    "Like getattr, but returns the default if the result is None"
    return getattr(o,s,d) or d

@patch(as_prop=True)
def total(self:Usage): return self.input_tokens+self.output_tokens+_dgetattr(self, "cache_creation_input_tokens",0)+_dgetattr(self, "cache_read_input_tokens",0)

Adding a total property to Usage makes it easier to see how many tokens we’ve used up altogether.

usage(5,1).total

source

Usage.repr

 Usage.__repr__ ()

Return repr(self).

Exported source

@patch
def __repr__(self:Usage):
    io_toks = f'In: {self.input_tokens}; Out: {self.output_tokens}'
    cache_toks = f'Cache create: {_dgetattr(self, "cache_creation_input_tokens",0)}; Cache read: {_dgetattr(self, "cache_read_input_tokens",0)}'
    server_tool_use = _dgetattr(self, "server_tool_use",server_tool_usage())
    server_tool_use_str = f'Search: {server_tool_use.web_search_requests}'
    total_tok = f'Total Tokens: {self.total}'
    return f'{io_toks}; {cache_toks}; {total_tok}; {server_tool_use_str}'

In python, patching __repr__ lets us change how an object is displayed. (More generally, methods starting and ending in __ in Python are called dunder methods, and have some magic behavior – such as, in this case, changing how an object is displayed.) We won’t be directly displaying ServerToolUsage’s, so we can handle its display behavior in the same Usage __repr__

usage(5)

In: 5; Out: 0; Cache create: 0; Cache read: 0; Total Tokens: 5; Search: 0

source

ServerToolUsage.add

 ServerToolUsage.__add__ (b)

Add together each of the server tool use counts

Exported source

@patch
def __add__(self:ServerToolUsage, b):
    "Add together each of the server tool use counts"
    return ServerToolUsage(web_search_requests=self.web_search_requests+b.web_search_requests)

And, patching __add__ lets + work on a ServerToolUsage as well as a Usage object.

server_tool_usage(1) + server_tool_usage(2)

ServerToolUsage(web_search_requests=3)

source

Usage.add

 Usage.__add__ (b)

Add together each of input_tokens and output_tokens

Exported source

@patch
def __add__(self:Usage, b):
    "Add together each of `input_tokens` and `output_tokens`"
    return usage(self.input_tokens+b.input_tokens, self.output_tokens+b.output_tokens,
                 _dgetattr(self,'cache_creation_input_tokens',0)+_dgetattr(b,'cache_creation_input_tokens',0),
                 _dgetattr(self,'cache_read_input_tokens',0)+_dgetattr(b,'cache_read_input_tokens',0),
                 _dgetattr(self,'server_tool_use',server_tool_usage())+_dgetattr(b,'server_tool_use',server_tool_usage()))

r.usage+r.usage + usage(server_tool_use=server_tool_usage(1))

In: 20; Out: 36; Cache create: 0; Cache read: 0; Total Tokens: 56; Search: 1

Creating messages

Creating correctly formatted dicts from scratch every time isn’t very handy, so we’ll import a couple of helper functions from the msglm library.

Let’s use mk_msg to recreate our msg {'role': 'user', 'content': "I'm Jeremy"} from earlier.

prompt = "I'm Jeremy"
m = mk_msg(prompt)
r = cli.messages.create(messages=[m], model=model, max_tokens=100)
r

Hello Jeremy! Nice to meet you. How can I help you today?

id: msg_01M2o3SENPk6JXLarGYZffo4
content: [{'citations': None, 'text': 'Hello Jeremy! Nice to meet you. How can I help you today?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 10, 'output_tokens': 18, 'server_tool_use': None, 'service_tier': 'standard'}

We can pass more than just text messages to Claude. As we’ll see later we can also pass images, SDK objects, etc. To handle these different data types we need to pass the type along with our content to Claude.

Here’s an example of a multimodal message containing text and images.

{
    'role': 'user', 
    'content': [
        {'type':'text', 'text':'What is in the image?'},
        {
            'type':'image', 
            'source': {
                'type':'base64', 'media_type':'media_type', 'data': 'data'
            }
        }
    ]
}

mk_msg infers the type automatically and creates the appropriate data structure.

LLMs, don’t actually have state, but instead dialogs are created by passing back all previous prompts and responses every time. With Claude, they always alternate user and assistant. We’ll use mk_msgs from msglm to make it easier to build up these dialog lists.

msgs = mk_msgs([prompt, r, "I forgot my name. Can you remind me please?"]) 
msgs

[{'role': 'user', 'content': "I'm Jeremy"},
 {'role': 'assistant',
  'content': [TextBlock(citations=None, text='Hello Jeremy! Nice to meet you. How can I help you today?', type='text')]},
 {'role': 'user', 'content': 'I forgot my name. Can you remind me please?'}]

cli.messages.create(messages=msgs, model=model, max_tokens=200)

Yes, your name is Jeremy! You told me that at the start of our conversation.

id: msg_01W1nFmwinpP3YwaB9zLhP2Y
content: [{'citations': None, 'text': 'Yes, your name is Jeremy! You told me that at the start of our conversation.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 42, 'output_tokens': 21, 'server_tool_use': None, 'service_tier': 'standard'}

Client

source

Client

 Client (model, cli=None, log=False, cache=False)

Basic Anthropic messages client.

Exported source

class Client:
    def __init__(self, model, cli=None, log=False, cache=False):
        "Basic Anthropic messages client."
        self.model,self.use = model,usage()
        self.text_only = model in text_only_models
        self.log = [] if log else None
        self.c = (cli or Anthropic(default_headers={'anthropic-beta': 'prompt-caching-2024-07-31'}))
        self.cache = cache

We’ll create a simple Client for Anthropic which tracks usage stores the model to use. We don’t add any methods right away – instead we’ll use patch for that so we can add and document them incrementally.

c = Client(model)
c.use

In: 0; Out: 0; Cache create: 0; Cache read: 0; Total Tokens: 0; Search: 0

Exported source

@patch
def _r(self:Client, r:Message, prefill=''):
    "Store the result of the message and accrue total usage."
    if prefill:
        blk = find_block(r)
        if blk: blk.text = prefill + (blk.text or '')
    self.result = r
    self.use += r.usage
    self.stop_reason = r.stop_reason
    self.stop_sequence = r.stop_sequence
    return r

We use a _ prefix on private methods, but we document them here in the interests of literate source code.

_r will be used each time we get a new result, to track usage and also to keep the result available for later.

c._r(r)
c.use

In: 10; Out: 18; Cache create: 0; Cache read: 0; Total Tokens: 28; Search: 0

Whereas OpenAI’s models use a stream parameter for streaming, Anthropic’s use a separate method. We implement Anthropic’s approach in a private method, and then use a stream parameter in __call__ for consistency:

Exported source

@patch
def _log(self:Client, final, prefill, msgs, **kwargs):
    self._r(final, prefill)
    if self.log is not None: self.log.append({
        "msgs": msgs, **kwargs,
        "result": self.result, "use": self.use, "stop_reason": self.stop_reason, "stop_sequence": self.stop_sequence
    })
    return self.result

Once streaming is complete, we need to store the final message and call any completion callback that’s needed.

source

get_types

 get_types (msgs)

Exported source

@save_iter
def _stream(o, cm, prefill, cb):
    with cm as s:
        yield prefill
        yield from s.text_stream
        o.value = s.get_final_message()
        cb(o.value)

get_types(msgs)

['text', 'text', 'text']

source

mk_tool_choice

 mk_tool_choice (choose:Union[str,bool,NoneType])

Create a tool_choice dict that’s ‘auto’ if choose is None, ‘any’ if it is True, or ‘tool’ otherwise

print(mk_tool_choice('sums'))
print(mk_tool_choice(True))
print(mk_tool_choice(None))

{'type': 'tool', 'name': 'sums'}
{'type': 'any'}
{'type': 'auto'}

Claude can be forced to use a particular tool, or select from a specific list of tools, or decide for itself when to use a tool. If you want to force a tool (or force choosing from a list), include a tool_choice param with a dict from mk_tool_choice.

Claude supports adding an extra assistant message at the end, which contains the prefill – i.e. the text we want Claude to assume the response starts with. However Claude doesn’t actually repeat that in the response, so for convenience we add it.

source

Client.call

 Client.__call__ (msgs:list, sp='', temp=0, maxtok=4096, maxthinktok=0,
                  prefill='', stream:bool=False, stop=None,
                  tools:Optional[list]=None,
                  tool_choice:Optional[dict]=None, cb=None,
                  metadata:MetadataParam|Omit=<anthropic.Omit object at
                  0x7f469390a1d0>, service_tier:"Literal['auto','standard_
                  only']|Omit"=<anthropic.Omit object at 0x7f469390a1d0>,
                  stop_sequences:SequenceNotStr[str]|Omit=<anthropic.Omit
                  object at 0x7f469390a1d0>, system:Union[str,Iterable[Tex
                  tBlockParam]]|Omit=<anthropic.Omit object at
                  0x7f469390a1d0>, temperature:float|Omit=<anthropic.Omit
                  object at 0x7f469390a1d0>,
                  thinking:ThinkingConfigParam|Omit=<anthropic.Omit object
                  at 0x7f469390a1d0>, top_k:int|Omit=<anthropic.Omit
                  object at 0x7f469390a1d0>,
                  top_p:float|Omit=<anthropic.Omit object at
                  0x7f469390a1d0>, extra_headers:Headers|None=None,
                  extra_query:Query|None=None, extra_body:Body|None=None,
                  timeout:float|httpx.Timeout|None|NotGiven=NOT_GIVEN)

Make a call to Claude.

	Type	Default	Details
msgs	list		List of messages in the dialog
sp	str		The system prompt
temp	int	0	Temperature
maxtok	int	4096	Maximum tokens
maxthinktok	int	0	Maximum thinking tokens
prefill	str		Optional prefill to pass to Claude as start of its response
stream	bool	False	Stream response?
stop	NoneType	None	Stop sequence
tools	Optional	None	List of tools to make available to Claude
tool_choice	Optional	None	Optionally force use of some tool
cb	NoneType	None	Callback to pass result to when complete
metadata	MetadataParam \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
service_tier	Literal[‘auto’, ‘standard_only’] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
stop_sequences	SequenceNotStr[str] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
system	Union[str, Iterable[TextBlockParam]] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
temperature	float \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
thinking	ThinkingConfigParam \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
top_k	int \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
top_p	float \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
extra_headers	Optional	None	Use the following arguments if you need to pass additional parameters to the API that aren’t available via kwargs. The extra values given here take precedence over values defined on the client or passed to this method.
extra_query	Query \| None	None
extra_body	Body \| None	None
timeout	float \| httpx.Timeout \| None \| NotGiven	NOT_GIVEN

Exported source

@patch
def _precall(self:Client, msgs, prefill, sp, temp, maxtok, maxthinktok, stream,
             stop, tools, tool_choice, kwargs):
    if tools: kwargs['tools'] = [get_schema(o) if callable(o) else o for o in listify(tools)]
    if tool_choice: kwargs['tool_choice'] = mk_tool_choice(tool_choice)
    if maxthinktok: 
        kwargs['thinking'] = {'type':'enabled', 'budget_tokens':maxthinktok} 
        temp,prefill = 1,''
    pref = [prefill.strip()] if prefill else []
    if not isinstance(msgs,list): msgs = [msgs]
    if stop is not None:
        if not isinstance(stop, (list)): stop = [stop]
        kwargs["stop_sequences"] = stop
    msgs = mk_msgs(msgs+pref, cache=self.cache, cache_last_ckpt_only=self.cache)
    assert not ('image' in get_types(msgs) and self.text_only), f"Images not supported by: {self.model}"
    kwargs |= dict(max_tokens=maxtok, system=sp, temperature=temp)
    return msgs, kwargs

Exported source

@patch
@delegates(messages.Messages.create)
def __call__(self:Client,
             msgs:list, # List of messages in the dialog
             sp='', # The system prompt
             temp=0, # Temperature
             maxtok=4096, # Maximum tokens
             maxthinktok=0, # Maximum thinking tokens
             prefill='', # Optional prefill to pass to Claude as start of its response
             stream:bool=False, # Stream response?
             stop=None, # Stop sequence
             tools:Optional[list]=None, # List of tools to make available to Claude
             tool_choice:Optional[dict]=None, # Optionally force use of some tool
             cb=None, # Callback to pass result to when complete
             **kwargs):
    "Make a call to Claude."
    msgs,kwargs = self._precall(msgs, prefill, sp, temp, maxtok, maxthinktok, stream,
                                stop, tools, tool_choice, kwargs)
    m = self.c.messages
    f = m.stream if stream else m.create
    res = f(model=self.model, messages=msgs, **kwargs)
    def _cb(v):
        self._log(v, prefill=prefill, msgs=msgs, **kwargs)
        if cb: cb(v)
    if stream: return _stream(res, prefill, _cb)
    try: return res
    finally: _cb(res)

Defining __call__ let’s us use an object like a function (i.e it’s callable). We use it as a small wrapper over messages.create.

c = Client(model, log=True)
c.use

In: 0; Out: 0; Cache create: 0; Cache read: 0; Total Tokens: 0; Search: 0

c('Hi')

Hello! How can I help you today?

id: msg_01XR7UCdLWF8MUeJZR2hhy5j
content: [{'citations': None, 'text': 'Hello! How can I help you today?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 8, 'output_tokens': 12, 'server_tool_use': None, 'service_tier': 'standard'}

Usage details are automatically updated after each call:

c.use

In: 8; Out: 12; Cache create: 0; Cache read: 0; Total Tokens: 20; Search: 0

A log of all messages is kept if log=True is passed:

pprint(c.log)

[{'max_tokens': 4096,
  'msgs': [{'content': 'Hi', 'role': 'user'}],
  'result': Message(id='msg_01XR7UCdLWF8MUeJZR2hhy5j', content=[TextBlock(citations=None, text='Hello! How can I help you today?', type='text')], model='claude-sonnet-4-5-20250929', role='assistant', stop_reason='end_turn', stop_sequence=None, type='message', usage=In: 8; Out: 12; Cache create: 0; Cache read: 0; Total Tokens: 20; Search: 0),
  'stop_reason': 'end_turn',
  'stop_sequence': None,
  'system': '',
  'temperature': 0,
  'use': In: 8; Out: 12; Cache create: 0; Cache read: 0; Total Tokens: 20; Search: 0}]

Let’s try out prefill:

q = "Very concisely, what is the meaning of life?"
pref = 'According to Douglas Adams, '

c(q, prefill=pref)

According to Douglas Adams, it’s 42.

More seriously: Create meaning through connections, growth, and contribution to something beyond yourself.

id: msg_013f6m5WD8Km3CzRRbY2wqAa
content: [{'citations': None, 'text': "According to Douglas Adams, it's 42.\n\nMore seriously: Create meaning through connections, growth, and contribution to something beyond yourself.", 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 24, 'output_tokens': 27, 'server_tool_use': None, 'service_tier': 'standard'}

c.use

In: 32; Out: 39; Cache create: 0; Cache read: 0; Total Tokens: 71; Search: 0

We can pass stream=True to stream the response back incrementally:

r = c('Hi', stream=True)
for o in r: print(o, end='')

Hello! How can I help you today?

c.use

In: 40; Out: 51; Cache create: 0; Cache read: 0; Total Tokens: 91; Search: 0

The full final message after completion of streaming is in the value attr of the response:

r.value

Hello! How can I help you today?

id: msg_01SDCfi2p2xtS7EbtwEMWXW5
content: [{'citations': None, 'text': 'Hello! How can I help you today?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 8, 'output_tokens': 12, 'server_tool_use': None, 'service_tier': 'standard'}

for o in c(q, prefill=pref, stream=True): print(o, end='')

According to Douglas Adams,   it's 42.

More seriously: Create meaning through connections, growth, and contribution to something beyond yourself.

c.use

In: 64; Out: 78; Cache create: 0; Cache read: 0; Total Tokens: 142; Search: 0

Pass a stop sequence if you want claude to stop generating text when it encounters it.

c("Count from 1 to 10", stop="5")

1, 2, 3, 4,

id: msg_011Ew5LZ8U9toQLNgyzZbrD9
content: [{'citations': None, 'text': '1, 2, 3, 4, ', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: stop_sequence
stop_sequence: 5
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 15, 'output_tokens': 14, 'server_tool_use': None, 'service_tier': 'standard'}

This also works with streaming, and you can pass more than one stop sequence:

for o in c("Count from 1 to 10", stop=["3", "yellow"], stream=True): print(o, end='')
print()
print(c.stop_reason, c.stop_sequence)

1, 2, 
stop_sequence 3

We’ve shown the token usage but we really care about is pricing. Let’s extract the latest pricing from Anthropic into a pricing dict.

source

get_pricing

 get_pricing (m, u)

Exported source

def get_pricing(m, u):
    return pricing[m][:3] if u.prompt_token_count < 128_000 else pricing[m][3:]

Similarly, let’s get the pricing for the latest server tools:

We’ll patch Usage to enable it compute the cost given pricing.

source

Usage.cost

 Usage.cost (costs:tuple)

Exported source

@patch
def cost(self:Usage, costs:tuple) -> float:
    cache_w, cache_r = _dgetattr(self, "cache_creation_input_tokens",0), _dgetattr(self, "cache_read_input_tokens",0)
    tok_cost = sum([self.input_tokens * costs[0] +  self.output_tokens * costs[1] +  cache_w * costs[2] + cache_r * costs[3]]) / 1e6
    server_tool_use = _dgetattr(self, "server_tool_use",server_tool_usage())
    server_tool_cost = server_tool_use.web_search_requests * server_tool_pricing['web_search_requests'] / 1e3
    return tok_cost + server_tool_cost

source

Client.cost

 Client.cost ()

Exported source

@patch(as_prop=True)
def cost(self: Client) -> float: return self.use.cost(pricing[model_types[self.model]])

source

get_costs

 get_costs (c)

Exported source

def get_costs(c):
    costs = pricing[model_types[c.model]]
    
    inp_cost = c.use.input_tokens * costs[0] / 1e6
    out_cost = c.use.output_tokens * costs[1] / 1e6

    cache_w = c.use.cache_creation_input_tokens   
    cache_r = c.use.cache_read_input_tokens
    cache_cost = (cache_w * costs[2] + cache_r * costs[3]) / 1e6

    server_tool_use = c.use.server_tool_use
    server_tool_cost = server_tool_use.web_search_requests * server_tool_pricing['web_search_requests'] / 1e3
    return inp_cost, out_cost, cache_cost, cache_w + cache_r, server_tool_cost

The markdown repr of the client itself will show the latest result, along with the usage so far.

Exported source

@patch
def _repr_markdown_(self:Client):
    if not hasattr(self,'result'): return 'No results yet'
    msg = contents(self.result)
    inp_cost, out_cost, cache_cost, cached_toks, server_tool_cost = get_costs(self)
    return f"""{msg}

| Metric | Count | Cost (USD) |
|--------|------:|-----:|
| Input tokens | {self.use.input_tokens:,} | {inp_cost:.6f} |
| Output tokens | {self.use.output_tokens:,} | {out_cost:.6f} |
| Cache tokens | {cached_toks:,} | {cache_cost:.6f} |
| Server tool use | {self.use.server_tool_use.web_search_requests:,} | {server_tool_cost:.6f} |
| **Total** | **{self.use.total:,}** | **${self.cost:.6f}** |"""

1, 2,

Metric	Count	Cost (USD)
Input tokens	94	0.000282
Output tokens	100	0.001500
Cache tokens	0	0.000000
Server tool use	0	0.000000
Total	194	$0.001782

Pass a list of alternating user/assistant messages to give Claude a “dialog”.

c(["My name is Jeremy", "Hi Jeremy!", "Can you remind me what my name is?"])

Yes, your name is Jeremy!

id: msg_016EVejFbBZrcbXkjEYBAC12
content: [{'citations': None, 'text': 'Yes, your name is Jeremy!', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 29, 'output_tokens': 10, 'server_tool_use': None, 'service_tier': 'standard'}

Tool use

Let’s now look more at tool use (aka function calling).

For testing, we need a function that Claude can call; we’ll write a simple function that adds numbers together, and will tell us when it’s being called:

@dataclass
class MySum: val:int

def sums(
    a:int,  # First thing to sum
    b:int=1 # Second thing to sum
) -> int: # The sum of the inputs
    "Adds a + b."
    print(f"Finding the sum of {a} and {b}")
    return MySum(a + b)

a,b = 604542,6458932
pr = f"What is {a}+{b}?"
sp = "Always use tools when calculations are required."

Claudette can autogenerate a schema thanks to the toolslm library. We’ll force the use of the tool using the function we created earlier.

tools=[get_schema(sums)]
choice = mk_tool_choice('sums')

We’ll start a dialog with Claude now. We’ll store the messages of our dialog in msgs. The first message will be our prompt pr, and we’ll pass our tools schema.

msgs = mk_msgs(pr)
r = c(msgs, sp=sp, tools=tools, tool_choice=choice)
r

ToolUseBlock(id=‘toolu_01A7hoYwW5hnMpaKfFxVXAzB’, input={‘a’: 604542, ‘b’: 6458932}, name=‘sums’, type=‘tool_use’)

id: msg_019ybw6cSukU8zdHB2HaPZzD
content: [{'id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB', 'input': {'a': 604542, 'b': 6458932}, 'name': 'sums', 'type': 'tool_use'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: tool_use
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 713, 'output_tokens': 57, 'server_tool_use': None, 'service_tier': 'standard'}

When Claude decides that it should use a tool, it passes back a ToolUseBlock with the name of the tool to call, and the params to use.

We don’t want to allow it to call just any possible function (that would be a security disaster!) so we create a namespace – that is, a dictionary of allowable function names to call.

ns = mk_ns(sums)
ns

{'sums': <function __main__.sums(a: int, b: int = 1) -> int>}

ToolResult is used for two special cases:

When tool calls are RPCs with claudette running on an application server and code execution happening elsewhere, wrapping with a result_type field is used as a type descriptor for the claudette client.
Different types are handled in message history with specific format, so mk_funcres branches the Anthropic representation (see depending on the result_type.

Currently images are the only supported tool result type - see https://docs.anthropic.com/en/docs/agents-and-tools/tool-use/implement-tool-use#example-of-tool-result-with-images for the format implemented in mk_funcres.

source

ToolResult

 ToolResult (result_type:str, data)

Base class for objects needing a basic __repr__

source

mk_funcres

 mk_funcres (fc, ns)

Given tool use block ‘fc’, get tool result, and create a tool_result response.

We can now use the function requested by Claude. We look it up in ns, and pass in the provided parameters.

fcs = [o for o in r.content if isinstance(o,ToolUseBlock)]
fcs

[ToolUseBlock(id='toolu_01A7hoYwW5hnMpaKfFxVXAzB', input={'a': 604542, 'b': 6458932}, name='sums', type='tool_use')]

res = [mk_funcres(fc, ns=ns) for fc in fcs]
res

Finding the sum of 604542 and 6458932

[{'type': 'tool_result',
  'tool_use_id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB',
  'content': 'MySum(val=7063474)'}]

source

mk_toolres

 mk_toolres (r:collections.abc.Mapping,
             ns:Optional[collections.abc.Mapping]=None)

Create a tool_result message from response r.

	Type	Default	Details
r	Mapping		Tool use request response from Claude
ns	Optional	None	Namespace to search for tools

Exported source

def mk_toolres(
    r:abc.Mapping, # Tool use request response from Claude
    ns:Optional[abc.Mapping]=None # Namespace to search for tools
    ):
    "Create a `tool_result` message from response `r`."
    cts = getattr(r, 'content', [])
    res = [mk_msg(r.model_dump(), role='assistant')]
    if ns is None: ns=globals()
    tcs = [mk_funcres(o, ns) for o in cts if isinstance(o,ToolUseBlock)]
    if tcs: res.append(mk_msg(tcs))
    return res

foo = []
foo.append({})
foo.append({})
foo

[{}, {}]

In order to tell Claude the result of the tool call, we pass back the tool use assistant request and the tool_result response.

tr = mk_toolres(r, ns=ns)
tr

Finding the sum of 604542 and 6458932

[{'role': 'assistant',
  'content': [{'id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB',
    'input': {'a': 604542, 'b': 6458932},
    'name': 'sums',
    'type': 'tool_use'}]},
 {'role': 'user',
  'content': [{'type': 'tool_result',
    'tool_use_id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB',
    'content': 'MySum(val=7063474)'}]}]

msgs

[{'role': 'user', 'content': 'What is 604542+6458932?'}]

We add this to our dialog, and now Claude has all the information it needs to answer our question.

msgs += tr
contents(c(msgs, sp=sp, tools=tools))

'The sum of 604542 + 6458932 = **7,063,474**'

contents(msgs[-1])

'MySum(val=7063474)'

msgs

[{'role': 'user', 'content': 'What is 604542+6458932?'},
 {'role': 'assistant',
  'content': [{'id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB',
    'input': {'a': 604542, 'b': 6458932},
    'name': 'sums',
    'type': 'tool_use'}]},
 {'role': 'user',
  'content': [{'type': 'tool_result',
    'tool_use_id': 'toolu_01A7hoYwW5hnMpaKfFxVXAzB',
    'content': 'MySum(val=7063474)'}]}]

Text editing

Anthropic also has a special tool type specific to text editing.

tools = [text_editor_conf['sonnet']]
tools

[{'type': 'text_editor_20250728', 'name': 'str_replace_based_edit_tool'}]

pr = 'Could you please explain my _quarto.yml file?'
msgs = [mk_msg(pr)]
r = c(msgs, sp=sp, tools=tools)
find_block(r, ToolUseBlock)

ToolUseBlock(id='toolu_01LFKfdhgBTckSrecDjYisDY', input={'command': 'view', 'path': '.'}, name='str_replace_based_edit_tool', type='tool_use')

We’ve gone ahead and create a reference implementation that you can directly use from our text_editor module. Or use as reference for creating your own.

ns = mk_ns(str_replace_based_edit_tool)
tr = mk_toolres(r, ns=ns)
msgs += tr
print(contents(c(msgs, sp=sp, tools=tools))[:128])

ToolUseBlock(id='toolu_01BtxanjZJXxMTztgQwvci7k', input={'command': 'view', 'path': '/_quarto.yml'}, name='str_replace_based_edi

Structured data

a,b = 604542,6458932
pr = f"What is {a}+{b}?"
sp = "Always use your tools for calculations."

for tools in [sums, [get_schema(sums)]]:
    r = c(pr, tools=tools, tool_choice='sums')
    print(r)

Message(id='msg_017V8xDmc28RJQChvHYRLAtM', content=[ToolUseBlock(id='toolu_01WzqADg5E5n2AKczQDppugU', input={'a': 604542, 'b': 6458932}, name='sums', type='tool_use')], model='claude-sonnet-4-5-20250929', role='assistant', stop_reason='tool_use', stop_sequence=None, type='message', usage=In: 708; Out: 53; Cache create: 0; Cache read: 0; Total Tokens: 761; Search: 0)
Message(id='msg_017V8xDmc28RJQChvHYRLAtM', content=[ToolUseBlock(id='toolu_01WzqADg5E5n2AKczQDppugU', input={'a': 604542, 'b': 6458932}, name='sums', type='tool_use')], model='claude-sonnet-4-5-20250929', role='assistant', stop_reason='tool_use', stop_sequence=None, type='message', usage=In: 708; Out: 53; Cache create: 0; Cache read: 0; Total Tokens: 761; Search: 0)

ns = mk_ns(sums)
tr = mk_toolres(r, ns=ns)

Finding the sum of 604542 and 6458932

source

Client.structured

 Client.structured (msgs:list, tools:Optional[list]=None,
                    ns:Optional[collections.abc.Mapping]=None, sp='',
                    temp=0, maxtok=4096, maxthinktok=0, prefill='',
                    stream:bool=False, stop=None,
                    tool_choice:Optional[dict]=None, cb=None,
                    metadata:MetadataParam|Omit=<anthropic.Omit object at
                    0x7f469390a1d0>, service_tier:"Literal['auto','standar
                    d_only']|Omit"=<anthropic.Omit object at
                    0x7f469390a1d0>, stop_sequences:SequenceNotStr[str]|Om
                    it=<anthropic.Omit object at 0x7f469390a1d0>, system:U
                    nion[str,Iterable[TextBlockParam]]|Omit=<anthropic.Omi
                    t object at 0x7f469390a1d0>,
                    temperature:float|Omit=<anthropic.Omit object at
                    0x7f469390a1d0>,
                    thinking:ThinkingConfigParam|Omit=<anthropic.Omit
                    object at 0x7f469390a1d0>,
                    top_k:int|Omit=<anthropic.Omit object at
                    0x7f469390a1d0>, top_p:float|Omit=<anthropic.Omit
                    object at 0x7f469390a1d0>,
                    extra_headers:Headers|None=None,
                    extra_query:Query|None=None,
                    extra_body:Body|None=None,
                    timeout:float|httpx.Timeout|None|NotGiven=NOT_GIVEN)

Return the value of all tool calls (generally used for structured outputs)

	Type	Default	Details
msgs	list		List of messages in the dialog
tools	Optional	None	List of tools to make available to Claude
ns	Optional	None	Namespace to search for tools
sp	str		The system prompt
temp	int	0	Temperature
maxtok	int	4096	Maximum tokens
maxthinktok	int	0	Maximum thinking tokens
prefill	str		Optional prefill to pass to Claude as start of its response
stream	bool	False	Stream response?
stop	NoneType	None	Stop sequence
tool_choice	Optional	None	Optionally force use of some tool
cb	NoneType	None	Callback to pass result to when complete
metadata	MetadataParam \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
service_tier	Literal[‘auto’, ‘standard_only’] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
stop_sequences	SequenceNotStr[str] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
system	Union[str, Iterable[TextBlockParam]] \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
temperature	float \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
thinking	ThinkingConfigParam \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
top_k	int \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
top_p	float \| Omit	<anthropic.Omit object at 0x7f469390a1d0>
extra_headers	Optional	None	Use the following arguments if you need to pass additional parameters to the API that aren’t available via kwargs. The extra values given here take precedence over values defined on the client or passed to this method.
extra_query	Query \| None	None
extra_body	Body \| None	None
timeout	float \| httpx.Timeout \| None \| NotGiven	NOT_GIVEN

Exported source

@patch
@delegates(Client.__call__)
def structured(self:Client,
               msgs:list, # List of messages in the dialog
               tools:Optional[list]=None, # List of tools to make available to Claude
               ns:Optional[abc.Mapping]=None, # Namespace to search for tools
               **kwargs):
    "Return the value of all tool calls (generally used for structured outputs)"
    tools = listify(tools)
    res = self(msgs, tools=tools, tool_choice=tools, **kwargs)
    if ns is None: ns=mk_ns(*tools)
    cts = getattr(res, 'content', [])
    tcs = [call_func(o.name, o.input, ns=ns) for o in cts if isinstance(o,ToolUseBlock)]
    return tcs

Anthropic’s API does not support response formats directly, so instead we provide a structured method to use tool calling to achieve the same result. The result of the tool is not passed back to Claude in this case, but instead is returned directly to the user.

c.structured(pr, tools=[sums])

Finding the sum of 604542 and 6458932

[MySum(val=7063474)]

ToolUseBlock(id=‘toolu_01JuugWVrXHqxi2DCwMo9ZUy’, input={‘a’: 604542, ‘b’: 6458932}, name=‘sums’, type=‘tool_use’)

Metric	Count	Cost (USD)
Input tokens	8,677	0.026031
Output tokens	538	0.008070
Cache tokens	0	0.000000
Server tool use	0	0.000000
Total	9,215	$0.034101

Custom Types with Tools Use

We need to add tool support for custom types too. Let’s test out custom types using a minimal example.

class Book(BasicRepr):
    def __init__(self, title: str, pages: int): store_attr()
    def __repr__(self):
        return f"Book Title : {self.title}\nNumber of Pages : {self.pages}"

Book("War and Peace", 950)

Book Title : War and Peace
Number of Pages : 950

def find_page(book: Book, # The book to find the halfway point of
              percent: int, # Percent of a book to read to, e.g. halfway == 50, 
) -> int:
    "The page number corresponding to `percent` completion of a book"
    return round(book.pages * (percent / 100.0))

get_schema(find_page)

{'name': 'find_page',
 'description': 'The page number corresponding to `percent` completion of a book\n\nReturns:\n- type: integer',
 'input_schema': {'type': 'object',
  'properties': {'book': {'type': 'object',
    'description': 'The book to find the halfway point of',
    '$ref': '#/$defs/Book'},
   'percent': {'type': 'integer',
    'description': 'Percent of a book to read to, e.g. halfway == 50,'}},
  'required': ['book', 'percent'],
  '$defs': {'Book': {'type': 'object',
    'properties': {'title': {'type': 'string', 'description': ''},
     'pages': {'type': 'integer', 'description': ''}},
    'title': 'Book',
    'required': ['title', 'pages']}}}}

choice = mk_tool_choice('find_page')
choice

{'type': 'tool', 'name': 'find_page'}

Claudette will pack objects as dict, so we’ll transform tool functions with user-defined types into tool functions that accept a dict in lieu of the user-defined type.

First let’s convert a single argument:

_is_builtin decides whether to pass an argument through as-is. Let’s check the argument conversion:

(_is_builtin(int), _is_builtin(Book), _is_builtin(List))

(True, False, True)

(_convert(555, int),
 _convert({"title": "War and Peace", "pages": 923}, Book),
 _convert([1, 2, 3, 4], List))

(555,
 Book Title : War and Peace
 Number of Pages : 923,
 [1, 2, 3, 4])

To apply tool() to a function is to return a new function where the user-defined types are replaced with dictionary inputs.

source

tool

 tool (func)

A function is transformed into a function with dict arguments substituted for user-defined types. Built-in types such as percent here are left untouched.

find_page(book=Book("War and Peace", 950), percent=50)

tool(find_page)({"title": "War and Peace", "pages": 950}, percent=50)

By passing tools wrapped by tool(), user-defined types now work completes without failing in tool calls.

pr = "How many pages do I have to read to get halfway through my 950 page copy of War and Peace"
tools = tool(find_page)
tools

<function __main__.find_page(book: __main__.Book, percent: int) -> int>

r = c(pr, tools=[tools])
find_block(r, ToolUseBlock)

ToolUseBlock(id='toolu_01C4PLJtgj6rktLBoabzu2vG', input={'book': {'title': 'War and Peace', 'pages': 950}, 'percent': 50}, name='find_page', type='tool_use')

tr = mk_toolres(r, ns=[tools])
tr

[{'role': 'assistant',
  'content': [{'id': 'toolu_01C4PLJtgj6rktLBoabzu2vG',
    'input': {'book': {'title': 'War and Peace', 'pages': 950}, 'percent': 50},
    'name': 'find_page',
    'type': 'tool_use'}]},
 {'role': 'user',
  'content': [{'type': 'tool_result',
    'tool_use_id': 'toolu_01C4PLJtgj6rktLBoabzu2vG',
    'content': '475'}]}]

msgs = [pr]+tr
contents(c(msgs, sp=sp, tools=[tools]))

'To get halfway through your 950-page copy of War and Peace, you need to read **475 pages**.'

Chat

Rather than manually adding the responses to a dialog, we’ll create a simple Chat class to do that for us, each time we make a request. We’ll also store the system prompt and tools here, to avoid passing them every time.

source

Chat

 Chat (model:Optional[str]=None, cli:Optional[__main__.Client]=None,
       sp='', tools:Optional[list]=None, temp=0,
       cont_pr:Optional[str]=None, cache:bool=False, hist:list=None,
       ns:Optional[collections.abc.Mapping]=None)

Anthropic chat client.

	Type	Default	Details
model	Optional	None	Model to use (leave empty if passing `cli`)
cli	Optional	None	Client to use (leave empty if passing `model`)
sp	str		Optional system prompt
tools	Optional	None	List of tools to make available to Claude
temp	int	0	Temperature
cont_pr	Optional	None	User prompt to continue an assistant response
cache	bool	False	Use Claude cache?
hist	list	None	Initialize history
ns	Optional	None	Namespace to search for tools

The class stores the Client that will provide the responses in c, and a history of messages in h.

sp = "Never mention what tools you use."
chat = Chat(model, sp=sp)
chat.c.use, chat.h

(In: 0; Out: 0; Cache create: 0; Cache read: 0; Total Tokens: 0; Search: 0, [])

chat.c.use.cost(pricing[model_types[chat.c.model]])

0.0

This is clunky. Let’s add cost as a property for the Chat class. It will pass in the appropriate prices for the current model to the usage cost calculator.

source

Chat.cost

 Chat.cost ()

Exported source

@patch(as_prop=True)
def cost(self: Chat) -> float: return self.c.cost

chat.cost

0.0

source

Chat.call

 Chat.__call__ (pr=None, temp=None, maxtok=4096, maxthinktok=0,
                stream=False, prefill='', tool_choice:Optional[dict]=None,
                **kw)

Call self as a function.

	Type	Default	Details
pr	NoneType	None	Prompt / message
temp	NoneType	None	Temperature
maxtok	int	4096	Maximum tokens
maxthinktok	int	0	Maximum thinking tokens
stream	bool	False	Stream response?
prefill	str		Optional prefill to pass to Claude as start of its response
tool_choice	Optional	None	Optionally force use of some tool
kw	VAR_KEYWORD

Exported source

@patch
def _post_pr(self:Chat, pr, prev_role):
    if pr is None and prev_role == 'assistant':
        if self.cont_pr is None:
            raise ValueError("Prompt must be given after completion, or use `self.cont_pr`.")
        pr = self.cont_pr # No user prompt, keep the chain
    if pr: self.h.append(mk_msg(pr, cache=self.cache))

Exported source

@patch
def _append_pr(self:Chat, pr=None):
    prev_role = nested_idx(self.h, -1, 'role') if self.h else 'assistant' # First message should be 'user'
    if pr and prev_role == 'user': self() # already user request pending
    self._post_pr(pr, prev_role)

Exported source

@patch
def __call__(self:Chat,
             pr=None,  # Prompt / message
             temp=None, # Temperature
             maxtok=4096, # Maximum tokens
             maxthinktok=0, # Maximum thinking tokens
             stream=False, # Stream response?
             prefill='', # Optional prefill to pass to Claude as start of its response
             tool_choice:Optional[dict]=None, # Optionally force use of some tool
             **kw):
    if temp is None: temp=self.temp
    self._append_pr(pr)
    def _cb(v):
        self.last = mk_toolres(v, ns=self.ns)
        self.h += self.last
    return self.c(self.h, stream=stream, prefill=prefill, sp=self.sp, temp=temp, maxtok=maxtok, maxthinktok=maxthinktok,
                 tools=self.tools, tool_choice=tool_choice, cb=_cb, **kw)

The __call__ method just passes the request along to the Client, but rather than just passing in this one prompt, it appends it to the history and passes it all along. As a result, we now have state!

chat = Chat(model, sp=sp)

chat("I'm Jeremy")
chat("What's my name?")

Your name is Jeremy! You told me that when we started our conversation.

id: msg_01SbfLJTbZTUqaGhPfisiDfW
content: [{'citations': None, 'text': 'Your name is Jeremy! You told me that when we started our conversation.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 51, 'output_tokens': 18, 'server_tool_use': None, 'service_tier': 'standard'}

chat.use, chat.cost

(In: 68; Out: 44; Cache create: 0; Cache read: 0; Total Tokens: 112; Search: 0,
 0.000864)

Let’s try out prefill too:

q = "Very concisely, what is the meaning of life?"
pref = 'According to Douglas Adams,'

chat(q, prefill=pref)

According to Douglas Adams, it’s 42.

More seriously: to find purpose through connection, growth, and contributing something meaningful to others.

id: msg_01WXQKBxnxtqniY5LZmwdndW
content: [{'citations': None, 'text': "According to Douglas Adams, it's 42. \n\nMore seriously: to find purpose through connection, growth, and contributing something meaningful to others.", 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 89, 'output_tokens': 29, 'server_tool_use': None, 'service_tier': 'standard'}

By default messages must be in user, assistant, user format. If this isn’t followed (aka calling chat() without a user message) it will error out:

try: chat()
except ValueError as e: print("Error:", e)

Error: Prompt must be given after completion, or use `self.cont_pr`.

Setting cont_pr allows a “default prompt” to be specified when a prompt isn’t specified. Usually used to prompt the model to continue.

chat.cont_pr = "Tell me a little more..."
chat()

The meaning of life is deeply personal, but common threads emerge:

Connection - We find meaning through relationships, love, and belonging. Humans are social creatures who thrive when bonded to others.

Growth - Learning, overcoming challenges, and becoming more than we were yesterday gives life direction and satisfaction.

Contribution - Creating something, helping others, leaving things better than we found them. Meaning often comes from looking beyond ourselves.

Experience - Simply being present - savoring beauty, joy, even struggle. The richness of conscious existence itself.

Purpose - Having something to strive toward, whether that’s raising a family, mastering a craft, pursuing justice, or exploring ideas.

Many philosophers suggest we create meaning rather than discover it - that life is what we make of it through our choices and values. The “meaning” isn’t handed to us; it emerges from how we engage with the world and what we decide matters.

What resonates with you, Jeremy?

id: msg_01UWMgALUQCscQir2jtb87fs
content: [{'citations': None, 'text': 'The meaning of life is deeply personal, but common threads emerge:\n\n**Connection** - We find meaning through relationships, love, and belonging. Humans are social creatures who thrive when bonded to others.\n\n**Growth** - Learning, overcoming challenges, and becoming more than we were yesterday gives life direction and satisfaction.\n\n**Contribution** - Creating something, helping others, leaving things better than we found them. Meaning often comes from looking beyond ourselves.\n\n**Experience** - Simply being present - savoring beauty, joy, even struggle. The richness of conscious existence itself.\n\n**Purpose** - Having something to strive toward, whether that\'s raising a family, mastering a craft, pursuing justice, or exploring ideas.\n\nMany philosophers suggest we *create* meaning rather than discover it - that life is what we make of it through our choices and values. The "meaning" isn\'t handed to us; it emerges from how we engage with the world and what we decide matters.\n\nWhat resonates with you, Jeremy?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 127, 'output_tokens': 219, 'server_tool_use': None, 'service_tier': 'standard'}

We can also use streaming:

chat = Chat(model, sp=sp)
for o in chat("I'm Jeremy", stream=True): print(o, end='')

Hello Jeremy! Nice to meet you. How are you doing today? Is there anything I can help you with?

r = chat(q, prefill=pref, stream=True)
for o in r: print(o, end='')
r.value

According to Douglas Adams,  it's 42.

More seriously: to find meaning through connection, growth, and contributing something beyond yourself.

According to Douglas Adams, it’s 42.

More seriously: to find meaning through connection, growth, and contributing something beyond yourself.

id: msg_01EVuKVkDzMM3eB6QbQmgstY
content: [{'citations': None, 'text': "According to Douglas Adams, it's 42.\n\nMore seriously: to find meaning through connection, growth, and contributing something beyond yourself.", 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 63, 'output_tokens': 27, 'server_tool_use': None, 'service_tier': 'standard'}

You can provide a history of messages to initialise Chat with:

chat = Chat(model, sp=sp, hist=["Can you guess my name?", "Hmmm I really don't know. Is it 'Merlin G. Penfolds'?"])
chat('Wow how did you know?')

I actually didn’t know - I was just making a playful guess! “Merlin G. Penfolds” was a completely random name I made up.

Is that really your name, or are you playing along with me? If it actually is your name, that would be an absolutely wild coincidence!

id: msg_015GNjsYKrEcMnE2s3wqxECB
content: [{'citations': None, 'text': 'I actually didn\'t know - I was just making a playful guess! "Merlin G. Penfolds" was a completely random name I made up. \n\nIs that really your name, or are you playing along with me? If it actually is your name, that would be an absolutely wild coincidence!', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 58, 'output_tokens': 70, 'server_tool_use': None, 'service_tier': 'standard'}

Chat tool use

We automagically get streamlined tool use as well:

pr = f"What is {a}+{b}?"
pr

'What is 604542+6458932?'

chat = Chat(model, sp=sp, tools=[sums])
r = chat(pr)
r

Finding the sum of 604542 and 6458932

ToolUseBlock(id=‘toolu_0156PbBN1a7y1VFD4YuBPp9x’, input={‘a’: 604542, ‘b’: 6458932}, name=‘sums’, type=‘tool_use’)

id: msg_017i1HX51CG9a7hfKvFT5JDV
content: [{'id': 'toolu_0156PbBN1a7y1VFD4YuBPp9x', 'input': {'a': 604542, 'b': 6458932}, 'name': 'sums', 'type': 'tool_use'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: tool_use
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 620, 'output_tokens': 72, 'server_tool_use': None, 'service_tier': 'standard'}

Now we need to send this result to Claude—calling the object with no parameters tells it to return the tool result to Claude:

chat()

The sum of 604542 + 6458932 is 7,063,474.

id: msg_016aVUB1HeyqED7vBQziprpB
content: [{'citations': None, 'text': 'The sum of 604542 + 6458932 is **7,063,474**.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 713, 'output_tokens': 24, 'server_tool_use': None, 'service_tier': 'standard'}

It should be correct, because it actually used our Python function to do the addition. Let’s check:

a+b

Let’s try the same thing with streaming:

chat = Chat(model, sp=sp, tools=[sums])
r = chat(pr, stream=True)
for o in r: print(o, end='')

Finding the sum of 604542 and 6458932

The full message, including tool call details, are in value:

r.value

ToolUseBlock(id=‘toolu_01VRFyr4Rn6HHjVSXBLmdoZe’, input={‘a’: 604542, ‘b’: 6458932}, name=‘sums’, type=‘tool_use’)

id: msg_01E8hssNQ2r4wvZrYmQFS6UE
content: [{'id': 'toolu_01VRFyr4Rn6HHjVSXBLmdoZe', 'input': {'a': 604542, 'b': 6458932}, 'name': 'sums', 'type': 'tool_use'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: tool_use
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 620, 'output_tokens': 72, 'server_tool_use': None, 'service_tier': 'standard'}

r = chat(stream=True)
for o in r: print(o, end='')

The sum of 604542 + 6458932 is **7,063,474**.

r.value

The sum of 604542 + 6458932 is 7,063,474.

id: msg_01H6B8tR27YcTr2Yobk9e3Y7
content: [{'citations': None, 'text': 'The sum of 604542 + 6458932 is **7,063,474**.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 713, 'output_tokens': 24, 'server_tool_use': None, 'service_tier': 'standard'}

The history shows both the tool_use and tool_result messages:

chat.h

[{'role': 'user', 'content': 'What is 604542+6458932?'},
 {'role': 'assistant',
  'content': [{'id': 'toolu_01VRFyr4Rn6HHjVSXBLmdoZe',
    'input': {'a': 604542, 'b': 6458932},
    'name': 'sums',
    'type': 'tool_use'}]},
 {'role': 'user',
  'content': [{'type': 'tool_result',
    'tool_use_id': 'toolu_01VRFyr4Rn6HHjVSXBLmdoZe',
    'content': 'MySum(val=7063474)'}]},
 {'role': 'assistant',
  'content': [{'citations': None,
    'text': 'The sum of 604542 + 6458932 is **7,063,474**.',
    'type': 'text'}]}]

Let’s test a function with user defined types.

chat = Chat(model, sp=sp, tools=[find_page])
r = chat("How many pages is three quarters of the way through my 80 page edition of Tao Te Ching?")
r

ToolUseBlock(id=‘toolu_019Cc2pacZoZ4voqnA3JK4c5’, input={‘book’: {‘title’: ‘Tao Te Ching’, ‘pages’: 80}, ‘percent’: 75}, name=‘find_page’, type=‘tool_use’)

id: msg_012ZLiQV1ozrhizZxLY92mNd
content: [{'id': 'toolu_019Cc2pacZoZ4voqnA3JK4c5', 'input': {'book': {'title': 'Tao Te Ching', 'pages': 80}, 'percent': 75}, 'name': 'find_page', 'type': 'tool_use'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: tool_use
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 730, 'output_tokens': 86, 'server_tool_use': None, 'service_tier': 'standard'}

chat()

Three quarters of the way through your 80-page edition of Tao Te Ching is page 60.

id: msg_018SAcnBrSuuiWa8xT1Lppnz
content: [{'citations': None, 'text': 'Three quarters of the way through your 80-page edition of Tao Te Ching is page 60.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 830, 'output_tokens': 28, 'server_tool_use': None, 'service_tier': 'standard'}

Exported source

@patch
def _repr_markdown_(self:Chat):
    if not hasattr(self.c, 'result'): return 'No results yet'
    last_msg = contents(self.c.result)
    
    def fmt_msg(m):
        t = contents(m)
        if isinstance(t, dict): return t['content']
        return t
        
    history = '\n\n'.join(f"**{m['role']}**: {fmt_msg(m)}" 
                         for m in self.h)
    det = self.c._repr_markdown_().split('\n\n')[-1]
    if history: history = f"""
<details>
<summary>► History</summary>

{history}

</details>
"""

    return f"""{last_msg}
{history}
{det}"""

# TODO: fix history format

chat

Three quarters of the way through your 80-page edition of Tao Te Ching is page 60.

► History

user: H

assistant: {‘id’: ‘toolu_019Cc2pacZoZ4voqnA3JK4c5’, ‘input’: {‘book’: {‘title’: ‘Tao Te Ching’, ‘pages’: 80}, ‘percent’: 75}, ‘name’: ‘find_page’, ‘type’: ‘tool_use’}

user: 60

assistant: Three quarters of the way through your 80-page edition of Tao Te Ching is page 60.

Metric	Count	Cost (USD)
Input tokens	1,560	0.004680
Output tokens	114	0.001710
Cache tokens	0	0.000000
Server tool use	0	0.000000
Total	1,674	$0.006390

chat = Chat(model, tools=[text_editor_conf['sonnet']], ns=mk_ns(str_replace_based_edit_tool))

When not providing tools directly as Python functions (like sum), you must create and pass a namespace dictionary (mapping the tool name string to the function object) using the ns parameter to methods like mk_toolres or toolloop. toolslm cannot automatically generate the namespace in this case. For schema-based tools (i.e., Python functions), claudette handles namespace creation automatically.

r = chat('Please explain very concisely what my _quarto.yml does. It is in the current path. Use your tools')
find_block(r, ToolUseBlock)

ToolUseBlock(id='toolu_016nQmu59mG21Ck5u4wjPZL3', input={'command': 'view', 'path': '.'}, name='str_replace_based_edit_tool', type='tool_use')

chat()

ToolUseBlock(id=‘toolu_012VDLH46pmrYuD6bPtBTp2s’, input={‘command’: ‘view’, ‘path’: ’/_quarto.yml’}, name=‘str_replace_based_edit_tool’, type=‘tool_use’)

id: msg_01Vu3rW5fXgQYnoYE1g7pJ3b
content: [{'id': 'toolu_012VDLH46pmrYuD6bPtBTp2s', 'input': {'command': 'view', 'path': '/_quarto.yml'}, 'name': 'str_replace_based_edit_tool', 'type': 'tool_use'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: tool_use
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 3714, 'output_tokens': 80, 'server_tool_use': None, 'service_tier': 'standard'}

Images

Claude can handle image data as well. As everyone knows, when testing image APIs you have to use a cute puppy.

# Image is Cute_dog.jpg from Wikimedia
fn = Path('samples/puppy.jpg')
Image(filename=fn, width=200)

img = fn.read_bytes()

Claude expects an image message to have the following structure

{
    'role': 'user', 
    'content': [
        {'type':'text', 'text':'What is in the image?'},
        {
            'type':'image', 
            'source': {
                'type':'base64', 'media_type':'media_type', 'data': 'data'
            }
        }
    ]
}

msglm automatically detects if a message is an image, encodes it, and generates the data structure above. All we need to do is a create a list containing our image and a query and then pass it to mk_msg.

Let’s try it out…

q = "In brief, what color flowers are in this image?"
msg = mk_msg([img, q])

c([msg])

The flowers in this image are purple.

id: msg_011fqyUyucLCtDf9mke8Fa2K
content: [{'citations': None, 'text': 'The flowers in this image are **purple**.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 110, 'output_tokens': 12, 'server_tool_use': None, 'service_tier': 'standard'}

You don’t need to call mk_msg on each individual message before passing them to the Chat class. Instead you can pass your messages in a list and the Chat class will automatically call mk_msgs in the background.

c(["How are you?", r])

For messages that contain multiple content types (like an image with a question), you’ll need to enclose the message contents in a list as shown below:

c(["How are you?", r, [img, q]])

c = Chat(model)
c([img, q])

The flowers in this image are purple.

id: msg_011fqyUyucLCtDf9mke8Fa2K
content: [{'citations': None, 'text': 'The flowers in this image are **purple**.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 110, 'output_tokens': 12, 'server_tool_use': None, 'service_tier': 'standard'}

contents(c.h[0])

'*Media Type - image*'

The flowers in this image are purple.

► History

user: Media Type - image

assistant: The flowers in this image are purple.

Metric	Count	Cost (USD)
Input tokens	110	0.000330
Output tokens	12	0.000180
Cache tokens	0	0.000000
Server tool use	0	0.000000
Total	122	$0.000510

Note

Unfortunately, not all Claude models support images 😞. This table summarizes the capabilities of each Claude model and the different modalities they support.

Caching

Claude supports context caching by adding a cache_control header to the message content.

{
    "role": "user",
    "content": [
        {
            "type": "text", 
            "text": "Please cache my message", 
            "cache_control": {"type": "ephemeral"}
        }
    ]
}

To cache a message, we simply set cache=True when calling mk_msg.

mk_msg(['hi', 'there'], cache=True)

{ 'content': [ {'text': 'hi', 'type': 'text'},
               { 'cache_control': {'type': 'ephemeral'},
                 'text': 'there',
                 'type': 'text'}],
  'role': 'user'}

Claude also now supports smart cache look-ups, so it’s very simple to keep an entire conversation in cache by constantly telling it to update the cache with the latest message. To do this, we just need to set cache=True when creating a Chat.

chat = Chat(model, sp=sp, cache=True)

Caching has a minimum token limit of 1024 tokens for Sonnet and Opus, and 2048 for Haiku. If your conversation is below this limit, it will not be cached.

chat("Hi, I'm Jeremy.")

Hello Jeremy! Nice to meet you. How are you doing today? Is there anything I can help you with?

id: msg_01YZZxrch65BWrDiKQs4eSam
content: [{'citations': None, 'text': 'Hello Jeremy! Nice to meet you. How are you doing today? Is there anything I can help you with?', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 20, 'output_tokens': 26, 'server_tool_use': None, 'service_tier': 'standard'}

chat.use

In: 20; Out: 26; Cache create: 0; Cache read: 0; Total Tokens: 46; Search: 0

Note the usage: no cache is created, nor used. Now, let’s send a long enough message to trigger caching.

chat("""Lorem ipsum dolor sit amet""" * 150)

I see you’ve sent me a lot of repeated “Lorem ipsum dolor sit amet” text - that’s the classic placeholder text often used in design and publishing!

Was there something specific you wanted to discuss or ask about, Jeremy? Or were you perhaps testing something? I’m here to help with whatever you need.

id: msg_01UdXDvuTDHc2dY2YsGrtnFm
content: [{'citations': None, 'text': 'I see you\'ve sent me a lot of repeated "Lorem ipsum dolor sit amet" text - that\'s the classic placeholder text often used in design and publishing!\n\nWas there something specific you wanted to discuss or ask about, Jeremy? Or were you perhaps testing something? I\'m here to help with whatever you need.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 1094}, 'cache_creation_input_tokens': 1094, 'cache_read_input_tokens': 0, 'input_tokens': 3, 'output_tokens': 67, 'server_tool_use': None, 'service_tier': 'standard'}

chat.use

In: 23; Out: 93; Cache create: 1094; Cache read: 0; Total Tokens: 1210; Search: 0

The context is now long enough for cache to be used. All the conversation history has now been written to the temporary cache. Any subsequent message will read from it rather than re-processing the entire conversation history.

chat("Oh thank you! Sorry, my lorem ipsum generator got out of control!")

No worries at all, Jeremy! Those things can definitely get a mind of their own sometimes. 😄 It happens!

So, now that we’ve got that sorted - what can I actually help you with today? Whether it’s answering questions, brainstorming ideas, or just having a chat, I’m all ears!

id: msg_01GzfzcdUfYMbLwiv88zxX6r
content: [{'citations': None, 'text': "No worries at all, Jeremy! Those things can definitely get a mind of their own sometimes. 😄 It happens!\n\nSo, now that we've got that sorted - what can I actually help you with today? Whether it's answering questions, brainstorming ideas, or just having a chat, I'm all ears!", 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 85}, 'cache_creation_input_tokens': 85, 'cache_read_input_tokens': 1094, 'input_tokens': 3, 'output_tokens': 74, 'server_tool_use': None, 'service_tier': 'standard'}

chat.use

In: 26; Out: 167; Cache create: 1179; Cache read: 1094; Total Tokens: 2466; Search: 0

Extended Thinking

Claude >=3.7 Sonnet & Opus have enhanced reasoning capabilities for complex tasks. See docs for more info.

We can enable extended thinking by passing a thinking param with the following structure.

thinking={ "type": "enabled", "budget_tokens": 16000 }

When extended thinking is enabled a thinking block is included in the response as shown below.

{
  "content": [
    {
      "type": "thinking",
      "thinking": "To approach this, let's think about...",
      "signature": "Imtakcjsu38219c0.eyJoYXNoIjoiYWJjM0NTY3fQ...."
    },
    {
      "type": "text",
      "text": "Yes, there are infinitely many prime numbers such that..."
    }
  ]
}

Note: When thinking is enabled prefill must be empty and the temp must be 1.

source

think_md

 think_md (txt, thk)

def contents(r, show_thk=True):
    "Helper to get the contents from Claude response `r`."
    blk = find_block(r)
    if show_thk:
        tk_blk = find_block(r, blk_type=ThinkingBlock)
        if tk_blk: return think_md(blk.text.strip(), tk_blk.thinking.strip())
    if not blk and r.content: blk = r.content[0]
    if hasattr(blk,'text'): return blk.text.strip()
    elif hasattr(blk,'content'): return blk.content.strip()
    elif hasattr(blk,'source'): return f'*Media Type - {blk.type}*'
    return str(blk)

Let’s call the model without extended thinking enabled.

chat = Chat(model)

chat("Write a sentence about Python!")

Python is a versatile, high-level programming language known for its clean syntax and readability, making it popular for everything from web development to data science and machine learning.

id: msg_013yMES6TZEhsGDXBFGC9xf8
content: [{'citations': None, 'text': 'Python is a versatile, high-level programming language known for its clean syntax and readability, making it popular for everything from web development to data science and machine learning.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 13, 'output_tokens': 38, 'server_tool_use': None, 'service_tier': 'standard'}

Now, let’s call the model with extended thinking enabled.

chat("Write a sentence about Python!", maxthinktok=1024)

Python’s extensive standard library and vast ecosystem of third-party packages make it an excellent choice for beginners and experienced developers alike.

Thinking

The user is asking me to write another sentence about Python. I should write a different sentence than before to provide variety.

id: msg_013wXiJbSVz9Nc5vTD7R5v7Z
content: [{'signature': 'EqoCCkYICRgCKkBQB7PJH8fZPgTfBGbWKFq9xRPuQI4TpQ5IHN5KDZmYu5bF7CU0jq3gAkpKXdznIM0MggxsuKZ/KV31KHh+HJTBEgw45GDws13hgo3DKXoaDPTCerdY+hLN1fMCeyIwntAHJ/ILI9uDvbX9mwka6JLYpuu2XtSWx41YAWykvq2g/97GPnAUWFLCIL/Qo2efKpEBuRNRgKAU7LpGCEjZt/J1+nKIEI6phhKbovsBSPj7zg0e72Ts2gyxXtJnaLdWvNjnz+ujpRmVW3DDb62OptFFrslxKrneMqVaVw9RywRm8EdcktWMvJhnR412k94ueawbskXlLQ5qwZo6gYpDvZj6g+LTT6O4Vrgf1eVvR4Q8ANDV9JPTRhlPsSLbSmikCQnh/BgB', 'thinking': 'The user is asking me to write another sentence about Python. I should write a different sentence than before to provide variety.', 'type': 'thinking'}, {'citations': None, 'text': "Python's extensive standard library and vast ecosystem of third-party packages make it an excellent choice for beginners and experienced developers alike.", 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 88, 'output_tokens': 62, 'server_tool_use': None, 'service_tier': 'standard'}

Server Tools and Web Search

The str_replace special tool type is a client side tool, i.e., one where we provide the implementation. However, Anthropic also supports server side tools. The current one available is their search tool, which you can find the documentation for here. When provided as a tool to claude, claude can decide to search the web in order to answer or solve the task at hand.

source

search_conf

 search_conf (max_uses:int=None, allowed_domains:list=None,
              blocked_domains:list=None, user_location:dict=None)

Little helper to create a search tool config

Similar to client side tools, you provide to the tools argument in the anthropic api a non-schema dictionary with the tool’s name, type, and any additional metadata specific to that tool. Here’s a function to make that process easier for the web search tool.

search_conf()

{'type': 'web_search_20250305', 'name': 'web_search'}

The web search tool returns a list of TextBlocks comprised of response text from the model, ServerToolUseBlock and server tool results block such as WebSearchToolResultBlock. Some of these TextBlocks will contain citations with references to the results of the web search tool. Here is what all this looks like:

{
  "content": [
    {
      "type": "text",
      "text": "I'll check the current weather in...",
    },
    {
      "type": "server_tool_use",
      "name": "web_search",
      "input": {"query": "San Diego weather forecast today May 12 2025"},
      "id":"srvtoolu_014t7fS449voTHRCVzi5jQGC"
    },
    {
      "type": "web_search_tool_result",
      "tool_use_id": "srvtoolu_014t7fS449voTHRCVzi5jQGC",
      "content": [
        "type": "web_search_result",
        "title": "Heat Advisory issued May 9...",
        "url": "https://kesq.com/weather/...",
        ...
      ]
    }
    {
      "type": "text",
      "citations": [
        {
            "cited_text": 'The average temperature during this month...',
            "title": "Weather San Diego in May 2025:...",
            "url": "https://en.climate-data.org/...",
            "encrypted_index": "EpMBCioIAxgCIiQ4ODk4YTF..."
        }
      ],
      "text": "The average temperature in San Diego during May is..."
    },
    ...
  ]
}

Let’s update our contents function to handle these cases. For handling citations, we will use the excellent reference syntax in markdown to make clickable citation links.

source

find_blocks

 find_blocks (r, blk_type=<class 'anthropic.types.text_block.TextBlock'>,
              type='text')

Helper to find all blocks of type blk_type in response r.

source

blks2cited_txt

 blks2cited_txt (txt_blks)

Helper to get the contents from a list of TextBlocks, with citations.

source

 contents (r, show_thk=True)

Helper to get the contents from Claude response r.

chat = Chat(model, sp='Be concise in your responses.', tools=[search_conf()], cache=True)
pr = 'What is the weather in San Diego?'
r = chat(pr)
r

The weather in San Diego today (Tuesday, November 4) is cloudy giving way to sun with a high of 69°F ¹ and a low of 59°F tonight ². There are areas of dense morning fog ³, and conditions are expected to be partly cloudy with a high of 69°F ⁴. Winds are from the west at 5 to 10 mph ⁵.

id: msg_01BXNzz3CRitC4iVkFftFyV1
content: [{'id': 'srvtoolu_01NktSQtP5NmFjjAA8w7iLLq', 'input': {'query': 'San Diego weather today'}, 'name': 'web_search', 'type': 'server_tool_use'}, {'content': [{'encrypted_content': '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', 'page_age': '2 days ago', 'title': 'San Diego, CA Weather Forecast | AccuWeather', 'type': 'web_search_result', 'url': 'https://www.accuweather.com/en/us/san-diego/92101/weather-forecast/347628'}, {'encrypted_content': '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', 'page_age': None, 'title': 'San Diego weather forecast – NBC 7 San Diego', 'type': 'web_search_result', 'url': 'https://www.nbcsandiego.com/weather/'}, {'encrypted_content': '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', 'page_age': None, 'title': 'San Diego, CA Hourly Weather Forecast | Weather Underground', 'type': 'web_search_result', 'url': 'https://www.wunderground.com/hourly/us/ca/san-diego'}, {'encrypted_content': 'EpkCCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDK0widPjf1RM8sbCtRoM7V134UKitP2+u+yZIjB8KzC75xkQ1eC+zQbPK+kb2Yl/SVXic6NX/I2J1h46I6dqwhtQBcMO5ENuN1mwIAsqnAHEJSk2ON/UaiFU8RxYRBTRG0CpYosE6TC58SKhXAGFpDBpUpdb2/khDbD37TYf3wR2Fq31vZVUMippWuVSeb0eVh62NnFzclNisF9YivjOQav4LBUdnZKu5lOLu1SHR8tTPEgvYf5U60IPkf8m9AMr6HVlBKBQK8PK+tLpbEz5wYS5ncbgbP5cDQfA6q7OCDs4gmtHBpDlWs8H46YYAw==', 'page_age': None, 'title': 'San Diego, CA Weather Forecast | KGTV | kgtv.com', 'type': 'web_search_result', 'url': 'https://www.10news.com/weather'}, {'encrypted_content': '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', 'page_age': 'August 30, 2021', 'title': 'San Diego 7-Day Weather Forecast | FOX 5 San Diego', 'type': 'web_search_result', 'url': 'https://fox5sandiego.com/weather/forecast/'}, {'encrypted_content': 'EsoCCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDDT3ddADKqKVfEa8QBoMxjdY0AuEp2dTv1fAIjD6BJaw0nJ9/3rB02WqHstq1ObR5Kuu+pJGQ7zYKS+Axm3dfQpdbaXBZyq79RRzlc8qzQGmOlv+v/a1hI70CCgz8m9dtnQ0GCiw0GiKlVlHGBeI1O8mhEPfL65nvmBnOYKtCGwzRA1yNpGGsbbCtt37lvoppMhJGxaNWR6nwVblIEhKgzkoKd8ev7ouCUGegxqC4QqIuaVGKS4M4oLONqUbN4XOF4KO40YNZiCFEP/81GokLVxiz0jnJYKVetF0TJxtv9P9aUP6svyxR1FBdd7FspBXISg+89voDI/A44jltsYUuSWLumv0NiEMUtplvQTa7UKbXUhKXe87WzWEqJu2GAM=', 'page_age': '2 days ago', 'title': 'San Diego, CA', 'type': 'web_search_result', 'url': 'https://www.weather.gov/sgx/'}, {'encrypted_content': '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', 'page_age': None, 'title': 'San Diego, CA 10-Day Weather Forecast | Weather Underground', 'type': 'web_search_result', 'url': 'https://www.wunderground.com/forecast/us/ca/san-diego'}, {'encrypted_content': 'EssBCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDA6zDzPxmyp5sG0I/BoMDDD7LxEWXz092xEEIjBGLvE1eJCdbnxN9AS41CXPvyuItyRMv1ruSlqC+C67/xXFpWiK5ioZb2QGqVDHTkEqT1p0bttFthiYyFQeP0EkQwYU/lLcGxU3zGGlGwWP35u2k/XnX+00qTqdvAwzi/I4WxAj8vEH/tJTXWaXgsV36dH/sDQVBmjm9STduGo+0h8YAw==', 'page_age': 'September 12, 2025', 'title': 'San Diego, CA Weather Forecast, Conditions, and Maps – Yahoo Weather', 'type': 'web_search_result', 'url': 'https://weather.yahoo.com/us/ca/san-diego'}, {'encrypted_content': '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', 'page_age': None, 'title': 'San Diego, CA Weather Conditions | Weather Underground', 'type': 'web_search_result', 'url': 'https://www.wunderground.com/weather/us/ca/san-diego'}, {'encrypted_content': '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', 'page_age': None, 'title': 'San Diego, CA Hourly Weather | AccuWeather', 'type': 'web_search_result', 'url': 'https://www.accuweather.com/en/us/san-diego/92101/hourly-weather-forecast/347628'}], 'tool_use_id': 'srvtoolu_01NktSQtP5NmFjjAA8w7iLLq', 'type': 'web_search_tool_result'}, {'citations': None, 'text': 'The weather in San Diego today (Tuesday, November 4) is ', 'type': 'text'}, {'citations': [{'cited_text': 'San Diego, CA Weather Today WinterCast ... · Tue, Nov 4 · Clouds giving way to sun Hi: 69° · Tonight: Mostly clear this evening followed by low clouds...', 'encrypted_index': 'Eo8BCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDCI8RZHER5H4wy1UQxoM99Iu58rn2qcOIJ/RIjBF4Mk1hivspsvTmN8Wfdjc5cn9VLuEAOHUrwVSTb8nlITqE8D8+ca4oVFdKLHDZyMqE+maM1hVg/jzuO9k1Yddb4hhobYYBA==', 'title': 'San Diego, CA Weather Forecast | AccuWeather', 'type': 'web_search_result_location', 'url': 'https://www.accuweather.com/en/us/san-diego/92101/weather-forecast/347628'}], 'text': 'cloudy giving way to sun with a high of 69°F', 'type': 'text'}, {'citations': None, 'text': ' and ', 'type': 'text'}, {'citations': [{'cited_text': 'San Diego, CA Weather Today WinterCast ... · Tue, Nov 4 · Clouds giving way to sun Hi: 69° · Tonight: Mostly clear this evening followed by low clouds...', 'encrypted_index': 'Eo8BCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDGjWzpY3qwEIVY3izxoMzFzug5M+598r05G9IjBxnPARQnUfGTyRGu08qCyNc/fEhlCSUSr2V62BaKICPZsD4znO+YD7/7gsvE85P8UqE9+ZR/3ZK7Fjll1qnoa2n+DMEYAYBA==', 'title': 'San Diego, CA Weather Forecast | AccuWeather', 'type': 'web_search_result_location', 'url': 'https://www.accuweather.com/en/us/san-diego/92101/weather-forecast/347628'}], 'text': 'a low of 59°F tonight', 'type': 'text'}, {'citations': None, 'text': '. ', 'type': 'text'}, {'citations': [{'cited_text': 'zoom out · Showing Stations · Hourly Forecast for Today, Tuesday 11/04Hourly for Today, Tue 11/04 · Today 11/04 · 6% / 0 in · Areas of dense morning f...', 'encrypted_index': 'Eo8BCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDCUenOMJHS91dxgcZxoMNTwCctyl0i8iVTiCIjBLBuhDaOXCnyjiXzZzk0Jxh5p0m95RC8xnoFqZyFzG9gggm1UzuNeIr7Uvlgz0OCYqE4Gnk4d+UnVMFB9nCurVCsccLL8YBA==', 'title': 'San Diego, CA Hourly Weather Forecast | Weather Underground', 'type': 'web_search_result_location', 'url': 'https://www.wunderground.com/hourly/us/ca/san-diego'}], 'text': 'There are areas of dense morning fog', 'type': 'text'}, {'citations': None, 'text': ', and conditions are expected to be ', 'type': 'text'}, {'citations': [{'cited_text': 'Partly cloudy. High 69F.', 'encrypted_index': 'EpEBCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDDl7RZDzBvFObljPlxoMuuO58/cSxdUFQIDFIjCIK+q0nmRBx7jZCgw/fPPnNBsMJOGyJlP/YsoIHwDPGD+WevKHxkcUUDDQe/p+5V8qFR4+s+sAqNIvC0Io8QK85dnlkBEntBgE', 'title': 'San Diego, CA Hourly Weather Forecast | Weather Underground', 'type': 'web_search_result_location', 'url': 'https://www.wunderground.com/hourly/us/ca/san-diego'}], 'text': 'partly cloudy with a high of 69°F', 'type': 'text'}, {'citations': None, 'text': '. ', 'type': 'text'}, {'citations': [{'cited_text': 'Winds W at 5 to 10 mph. ', 'encrypted_index': 'EpABCioICRgCIiQ4ODk4YTFkYy0yMTNkLTRhNmYtOTljYi03ZTBlNTUzZDc0NWISDFnjnamfAYtQ4mn2FBoMaLpFtaTxV2joNHsZIjBf7LY2Bj3+0xdlqJ0kEZHSS1DUbofo0iI2sLQSxRExEMl551preQlEw4gJWvLj7WkqFOnWWkxYvgkUrIawSec7xz/qrDW1GAQ=', 'title': 'San Diego, CA Hourly Weather Forecast | Weather Underground', 'type': 'web_search_result_location', 'url': 'https://www.wunderground.com/hourly/us/ca/san-diego'}], 'text': 'Winds are from the west at 5 to 10 mph', 'type': 'text'}, {'citations': None, 'text': '.', 'type': 'text'}]
model: claude-sonnet-4-5-20250929
role: assistant
stop_reason: end_turn
stop_sequence: None
type: message
usage: {'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 2226}, 'cache_creation_input_tokens': 7606, 'cache_read_input_tokens': 2226, 'input_tokens': 10, 'output_tokens': 226, 'server_tool_use': {'web_search_requests': 1}, 'service_tier': 'standard'}

Third party providers

NB: The 3rd party model list is currently out of date–PRs to fix that would be welcome!

Amazon Bedrock

These are Amazon’s current Claude models:

models_aws

['anthropic.claude-opus-4-1-20250805-v1:0',
 'anthropic.claude-sonnet-4-20250514-v1:0',
 'claude-3-5-haiku-20241022',
 'claude-3-7-sonnet-20250219',
 'anthropic.claude-3-opus-20240229-v1:0',
 'anthropic.claude-3-5-sonnet-20241022-v2:0']

Provided boto3 is installed, we otherwise don’t need any extra code to support Amazon Bedrock – we just have to set up the approach client:

ab = AnthropicBedrock(
    aws_access_key=os.environ['AWS_ACCESS_KEY'],
    aws_secret_key=os.environ['AWS_SECRET_KEY'],
)
client = Client(models_aws[0], ab)

chat = Chat(cli=client)

chat("I'm Jeremy")

Google Vertex

models_goog

['claude-opus-4-1@20250805',
 'anthropic.claude-3-sonnet-20240229-v1:0',
 'anthropic.claude-3-haiku-20240307-v1:0',
 'claude-3-opus@20240229',
 'claude-3-5-sonnet-v2@20241022',
 'claude-3-sonnet@20240229',
 'claude-3-haiku@20240307']

from anthropic import AnthropicVertex
import google.auth

project_id = google.auth.default()[1]
region = "us-east5"
gv = AnthropicVertex(project_id=project_id, region=region)
client = Client(models_goog[-1], gv)

chat = Chat(cli=client)

chat("I'm Jeremy")

Footnotes

https://www.accuweather.com/en/us/san-diego/92101/weather-forecast/347628 “San Diego, CA Weather Today WinterCast … · Tue, Nov 4 · Clouds giving way to sun Hi: 69° · Tonight: Mostly clear this evening followed by low clouds…”↩︎
https://www.accuweather.com/en/us/san-diego/92101/weather-forecast/347628 “San Diego, CA Weather Today WinterCast … · Tue, Nov 4 · Clouds giving way to sun Hi: 69° · Tonight: Mostly clear this evening followed by low clouds…”↩︎
https://www.wunderground.com/hourly/us/ca/san-diego “zoom out · Showing Stations · Hourly Forecast for Today, Tuesday 11/04Hourly for Today, Tue 11/04 · Today 11/04 · 6% / 0 in · Areas of dense morning f…”↩︎
https://www.wunderground.com/hourly/us/ca/san-diego “Partly cloudy. High 69F.”↩︎
https://www.wunderground.com/hourly/us/ca/san-diego “Winds W at 5 to 10 mph.”↩︎