Module: LlmCostTracker::Integrations::RubyLlm

Extended by:
Base
Defined in:
lib/llm_cost_tracker/integrations/ruby_llm.rb

Defined Under Namespace

Modules: ProviderPatch

Class Method Summary collapse

Methods included from Base

active?, enforce_budget!, gem_version, install, integration_name, minimum_version, patch_target, patch_targets, provider, record_safely, request_params, status, stream_collector, stream_pricing_mode, track_stream, wrap_blocking, wrap_stream

Class Method Details

.blocking_seam(resource, record_method, **extras) ⇒ Object 



184
185
186
187
188
189
190
191
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 184

def blocking_seam(resource, record_method, **extras)
  {
    provider: resource.slug.to_s,
    record: lambda do |response, request, latency_ms|
      public_send(record_method, resource, response, request: request, latency_ms: latency_ms, **extras)
    end
  }
end

.cache_creation_split(provider, response) ⇒ Object 



145
146
147
148
149
150
151
152
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 145

def cache_creation_split(provider, response)
  return [response.try(:cache_creation_tokens).to_i, 0] unless provider == "anthropic"

  cache = raw_body(response).dig("usage", "cache_creation")
  return [response.try(:cache_creation_tokens).to_i, 0] unless cache.is_a?(Hash)

  [cache["ephemeral_5m_input_tokens"].to_i, cache["ephemeral_1h_input_tokens"].to_i]
end

.image_token_detail(usage, direction) ⇒ Object 



77
78
79
80
81
82
83
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 77

def image_token_detail(usage, direction)
  container_key = direction == :input ? :input_tokens_details : :output_tokens_details
  details = usage[container_key]
  return 0 unless details.is_a?(Hash)

  details.with_indifferent_access[:image_tokens].to_i
end

.model_id_from_request(value) ⇒ Object 



154
155
156
157
158
159
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 154

def model_id_from_request(value)
  return nil if value.nil?
  return value.to_s if value.is_a?(String) || value.is_a?(Symbol)

  (value.try(:id) || value.try(:model_id) || value.try(:model))&.to_s
end

.patch_targets ⇒ Object 



13
14
15
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 13

def patch_targets
  [patch_target("RubyLLM::Provider", with: ProviderPatch)]
end

.pricing_mode_for(provider:, response:) ⇒ Object 



175
176
177
178
179
180
181
182
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 175

def pricing_mode_for(provider:, response:)
  body = raw_body(response)
  case provider
  when "anthropic" then body.dig("usage", "service_tier")
  when "gemini" then body.dig("usageMetadata", "serviceTier")
  else body["service_tier"]
  end
end

.provider_response_id_for(response) ⇒ Object 



161
162
163
164
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 161

def provider_response_id_for(response)
  body = raw_body(response)
  body["id"] || body["responseId"]
end

.raw_body(response) ⇒ Object 



166
167
168
169
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 166

def raw_body(response)
  body = response.try(:raw)&.body
  body.is_a?(Hash) ? body : {}
end

.record_completion(provider, response, request:, latency_ms:, has_block:) ⇒ Object 



17
18
19
20
21
22
23
24
25
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 17

def record_completion(provider, response, request:, latency_ms:, has_block:)
  record_usage(
    provider: provider.slug.to_s,
    model: response_model_id(response) || model_id_from_request(request[:model]),
    response: response,
    latency_ms: latency_ms,
    stream: has_block || request[:stream] == true
  )
end

.record_embedding(provider, response, request:, latency_ms:) ⇒ Object 



27
28
29
30
31
32
33
34
35
36
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 27

def record_embedding(provider, response, request:, latency_ms:)
  record_usage(
    provider: provider.slug.to_s,
    model: response_model_id(response) || model_id_from_request(request[:model]),
    response: response,
    latency_ms: latency_ms,
    stream: false,
    output_tokens: 0
  )
end

.record_image(provider, response, request:, latency_ms:) ⇒ Object 



48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 48

def record_image(provider, response, request:, latency_ms:)
  usage = response.usage.with_indifferent_access
  raw_input = usage[:input_tokens].to_i
  raw_output = usage[:output_tokens].to_i
  image_input = image_token_detail(usage, :input)
  image_output = image_token_detail(usage, :output)
  record_passthrough(
    provider: provider.slug.to_s,
    model: response_model_id(response) || model_id_from_request(request[:model]),
    response: response,
    latency_ms: latency_ms,
    input_tokens: [raw_input - image_input, 0].max,
    image_input_tokens: image_input,
    output_tokens: [raw_output - image_output, 0].max,
    image_output_tokens: image_output
  )
end

.record_moderation(provider, response, request:, latency_ms:) ⇒ Object 



66
67
68
69
70
71
72
73
74
75
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 66

def record_moderation(provider, response, request:, latency_ms:)
  record_passthrough(
    provider: provider.slug.to_s,
    model: response_model_id(response) || model_id_from_request(request[:model]),
    response: response,
    latency_ms: latency_ms,
    input_tokens: 0,
    output_tokens: 0
  )
end

.record_passthrough(provider:, model:, response:, latency_ms:, input_tokens:, output_tokens:, image_input_tokens: 0, image_output_tokens: 0) ⇒ Object 



85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 85

def record_passthrough(provider:,
                       model:,
                       response:,
                       latency_ms:,
                       input_tokens:,
                       output_tokens:,
                       image_input_tokens: 0,
                       image_output_tokens: 0)
  return unless active?

  record_safely do
    LlmCostTracker::Tracker.record(
      event: Event.build(
        provider: provider,
        model: model,
        token_usage: Usage::TokenUsage.build(
          input_tokens: input_tokens,
          output_tokens: output_tokens,
          image_input_tokens: image_input_tokens,
          image_output_tokens: image_output_tokens
        ),
        usage_source: LlmCostTracker::Usage::Source::SDK_RESPONSE,
        provider_response_id: provider_response_id_for(response)
      ),
      latency_ms: latency_ms
    )
  end
end

.record_transcription(provider, response, request:, latency_ms:) ⇒ Object 



38
39
40
41
42
43
44
45
46
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 38

def record_transcription(provider, response, request:, latency_ms:)
  record_usage(
    provider: provider.slug.to_s,
    model: response_model_id(response) || model_id_from_request(request[:model]),
    response: response,
    latency_ms: latency_ms,
    stream: false
  )
end

.record_usage(provider:, model:, response:, latency_ms:, stream:, output_tokens: nil) ⇒ Object 



114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 114

def record_usage(provider:, model:, response:, latency_ms:, stream:, output_tokens: nil)
  return unless active?

  record_safely do
    input_tokens = response.input_tokens
    output_tokens = response.output_tokens if output_tokens.nil?
    next if input_tokens.nil? && output_tokens.nil?

    cache_write_5m, cache_write_1h = cache_creation_split(provider, response)
    LlmCostTracker::Tracker.record(
      event: Event.build(
        provider: provider,
        model: model,
        pricing_mode: pricing_mode_for(provider: provider, response: response),
        token_usage: Usage::TokenUsage.build(
          input_tokens: input_tokens.to_i,
          output_tokens: output_tokens.to_i,
          cache_read_input_tokens: response.try(:cached_tokens).to_i,
          cache_write_input_tokens: cache_write_5m,
          cache_write_extended_input_tokens: cache_write_1h,
          hidden_output_tokens: response.try(:thinking_tokens).to_i
        ),
        stream: stream,
        usage_source: LlmCostTracker::Usage::Source::SDK_RESPONSE,
        provider_response_id: provider_response_id_for(response)
      ),
      latency_ms: latency_ms
    )
  end
end

.response_model_id(response) ⇒ Object 



171
172
173
 
# File 'lib/llm_cost_tracker/integrations/ruby_llm.rb', line 171

def response_model_id(response)
  (response.try(:model_id) || response.try(:model))&.to_s
end