AI-Anthropic
view release on metacpan or search on metacpan
lib/AI/Anthropic.pm view on Meta::CPAN
package AI::Anthropic;
use strict;
use warnings;
use 5.010;
our $VERSION = '0.01';
use Carp qw(croak);
use JSON::PP;
use HTTP::Tiny;
use MIME::Base64 qw(encode_base64);
# Constants
use constant {
API_BASE => 'https://api.anthropic.com',
API_VERSION => '2023-06-01',
DEFAULT_MODEL => 'claude-sonnet-4-20250514',
};
=head1 NAME
AI::Anthropic - Perl interface to Anthropic's Claude API
=head1 SYNOPSIS
use AI::Anthropic;
my $claude = AI::Anthropic->new(
api_key => 'sk-ant-api03-your-key-here',
);
# Simple message
my $response = $claude->message("What is the capital of France?");
print $response; # prints response text
# Chat with history
my $response = $claude->chat(
messages => [
{ role => 'user', content => 'Hello!' },
{ role => 'assistant', content => 'Hello! How can I help you today?' },
{ role => 'user', content => 'What is 2+2?' },
],
);
# With system prompt
my $response = $claude->chat(
system => 'You are a helpful Perl programmer.',
messages => [
{ role => 'user', content => 'How do I read a file?' },
],
);
# Streaming
$claude->chat(
messages => [ { role => 'user', content => 'Tell me a story' } ],
stream => sub {
my ($chunk) = @_;
print $chunk;
},
);
=head1 DESCRIPTION
AI::Anthropic provides a Perl interface to Anthropic's Claude API.
It supports all Claude models including Claude 4 Opus, Claude 4 Sonnet,
and Claude Haiku.
=head1 METHODS
=head1 AUTHOR
Vugar Bakhshaliyev <d7951500@gmail.com>
=head1 LICENSE
This is free software; you can redistribute it and/or modify it under
the same terms as the Perl 5 programming language system itself.
=head2 new
my $claude = AI::Anthropic->new(
api_key => 'your-api-key', # required (or use ANTHROPIC_API_KEY env)
model => 'claude-sonnet-4-20250514', # optional
max_tokens => 4096, # optional
timeout => 120, # optional, seconds
);
=cut
sub new {
my ($class, %args) = @_;
my $api_key = $args{api_key} // $ENV{ANTHROPIC_API_KEY}
or croak "API key required. Set api_key parameter or ANTHROPIC_API_KEY environment variable";
my $self = {
api_key => $api_key,
model => $args{model} // DEFAULT_MODEL,
max_tokens => $args{max_tokens} // 4096,
timeout => $args{timeout} // 120,
api_base => $args{api_base} // API_BASE,
_http => HTTP::Tiny->new(
timeout => $args{timeout} // 120,
),
_json => JSON::PP->new->utf8->allow_nonref,
};
return bless $self, $class;
}
=head2 message
Simple interface for single message:
my $response = $claude->message("Your question here");
my $response = $claude->message("Your question", system => "You are helpful");
print $response->text;
=cut
sub message {
my ($self, $content, %opts) = @_;
croak "Message content required" unless defined $content;
return $self->chat(
messages => [ { role => 'user', content => $content } ],
%opts,
);
}
=head2 chat
Full chat interface:
my $response = $claude->chat(
messages => \@messages, # required
system => $system_prompt, # optional
model => $model, # optional, overrides default
max_tokens => $max_tokens, # optional
temperature => 0.7, # optional, 0.0-1.0
stream => \&callback, # optional, for streaming
tools => \@tools, # optional, for function calling
);
=cut
sub chat {
my ($self, %args) = @_;
my $messages = $args{messages}
or croak "messages parameter required";
# Build request body
my $body = {
model => $args{model} // $self->{model},
max_tokens => $args{max_tokens} // $self->{max_tokens},
messages => $self->_normalize_messages($messages),
};
# Optional parameters
$body->{system} = $args{system} if defined $args{system};
$body->{temperature} = $args{temperature} if defined $args{temperature};
$body->{tools} = $args{tools} if defined $args{tools};
lib/AI/Anthropic.pm view on Meta::CPAN
}
sub _image_from_file {
my ($self, $path) = @_;
open my $fh, '<:raw', $path
or croak "Cannot open image file '$path': $!";
local $/;
my $data = <$fh>;
close $fh;
# Detect media type
my $media_type = 'image/png';
if ($path =~ /\.jpe?g$/i) {
$media_type = 'image/jpeg';
} elsif ($path =~ /\.gif$/i) {
$media_type = 'image/gif';
} elsif ($path =~ /\.webp$/i) {
$media_type = 'image/webp';
}
return {
type => 'image',
source => {
type => 'base64',
media_type => $media_type,
data => encode_base64($data, ''),
},
};
}
sub _image_from_url {
my ($self, $url) = @_;
return {
type => 'image',
source => {
type => 'url',
url => $url,
},
};
}
sub _request {
my ($self, $body) = @_;
my $response = $self->{_http}->post(
$self->{api_base} . '/v1/messages',
{
headers => $self->_headers,
content => $self->{_json}->encode($body),
}
);
return $self->_handle_response($response);
}
sub _stream_request {
my ($self, $body, $callback) = @_;
$body->{stream} = \1; # JSON true
my $full_text = '';
my $response_data;
# HTTP::Tiny doesn't support streaming well, so we use a data callback
my $response = $self->{_http}->post(
$self->{api_base} . '/v1/messages',
{
headers => $self->_headers,
content => $self->{_json}->encode($body),
data_callback => sub {
my ($chunk, $res) = @_;
# Parse SSE events
for my $line (split /\n/, $chunk) {
next unless $line =~ /^data: (.+)/;
my $data = $1;
next if $data eq '[DONE]';
eval {
my $event = $self->{_json}->decode($data);
if ($event->{type} eq 'content_block_delta') {
my $text = $event->{delta}{text} // '';
$full_text .= $text;
$callback->($text) if $callback;
} elsif ($event->{type} eq 'message_stop') {
$response_data = $event;
}
};
}
},
}
);
unless ($response->{success}) {
return $self->_handle_response($response);
}
# Return a response object with the full text
return AI::Anthropic::Response->new(
text => $full_text,
raw_response => $response_data,
);
}
sub _headers {
my ($self) = @_;
return {
'Content-Type' => 'application/json',
'x-api-key' => $self->{api_key},
'anthropic-version' => API_VERSION,
};
}
sub _handle_response {
my ($self, $response) = @_;
my $data;
( run in 0.809 second using v1.01-cache-2.11-cpan-cdf2f3d4e48 )