llm_cost_tracker 0.3.0 → 0.3.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/CHANGELOG.md +23 -0
- data/CODE_OF_CONDUCT.md +23 -0
- data/README.md +86 -8
- data/SECURITY.md +36 -0
- data/app/assets/llm_cost_tracker/application.css +1 -4
- data/app/controllers/llm_cost_tracker/assets_controller.rb +0 -2
- data/app/controllers/llm_cost_tracker/calls_controller.rb +9 -13
- data/app/controllers/llm_cost_tracker/dashboard_controller.rb +8 -19
- data/app/controllers/llm_cost_tracker/data_quality_controller.rb +1 -2
- data/app/controllers/llm_cost_tracker/models_controller.rb +5 -2
- data/app/controllers/llm_cost_tracker/tags_controller.rb +2 -4
- data/app/helpers/llm_cost_tracker/dashboard_filter_options_helper.rb +1 -7
- data/app/helpers/llm_cost_tracker/dashboard_query_helper.rb +5 -9
- data/app/services/llm_cost_tracker/dashboard/data_quality.rb +10 -10
- data/app/services/llm_cost_tracker/dashboard/filter.rb +6 -26
- data/app/services/llm_cost_tracker/dashboard/provider_breakdown.rb +0 -3
- data/app/services/llm_cost_tracker/dashboard/tag_breakdown.rb +0 -2
- data/app/services/llm_cost_tracker/pagination.rb +1 -9
- data/app/views/layouts/llm_cost_tracker/application.html.erb +1 -16
- data/app/views/llm_cost_tracker/calls/index.html.erb +13 -13
- data/app/views/llm_cost_tracker/calls/show.html.erb +8 -3
- data/app/views/llm_cost_tracker/dashboard/index.html.erb +1 -1
- data/app/views/llm_cost_tracker/data_quality/index.html.erb +36 -14
- data/app/views/llm_cost_tracker/models/index.html.erb +10 -9
- data/app/views/llm_cost_tracker/shared/_spend_chart.html.erb +0 -1
- data/app/views/llm_cost_tracker/shared/_tag_chips.html.erb +0 -1
- data/app/views/llm_cost_tracker/tags/index.html.erb +1 -1
- data/app/views/llm_cost_tracker/tags/show.html.erb +1 -1
- data/lib/llm_cost_tracker/configuration.rb +0 -1
- data/lib/llm_cost_tracker/event.rb +1 -0
- data/lib/llm_cost_tracker/event_metadata.rb +1 -0
- data/lib/llm_cost_tracker/generators/llm_cost_tracker/add_provider_response_id_generator.rb +29 -0
- data/lib/llm_cost_tracker/generators/llm_cost_tracker/templates/add_provider_response_id_to_llm_api_calls.rb.erb +15 -0
- data/lib/llm_cost_tracker/generators/llm_cost_tracker/templates/create_llm_api_calls.rb.erb +2 -0
- data/lib/llm_cost_tracker/llm_api_call.rb +6 -2
- data/lib/llm_cost_tracker/middleware/faraday.rb +1 -0
- data/lib/llm_cost_tracker/parameter_hash.rb +33 -0
- data/lib/llm_cost_tracker/parsed_usage.rb +14 -3
- data/lib/llm_cost_tracker/parsers/anthropic.rb +47 -28
- data/lib/llm_cost_tracker/parsers/gemini.rb +28 -4
- data/lib/llm_cost_tracker/parsers/openai_compatible.rb +5 -6
- data/lib/llm_cost_tracker/parsers/openai_usage.rb +14 -0
- data/lib/llm_cost_tracker/price_registry.rb +22 -7
- data/lib/llm_cost_tracker/price_sync/refresh_plan_builder.rb +162 -0
- data/lib/llm_cost_tracker/price_sync/registry_loader.rb +55 -0
- data/lib/llm_cost_tracker/price_sync/registry_writer.rb +25 -0
- data/lib/llm_cost_tracker/price_sync.rb +16 -184
- data/lib/llm_cost_tracker/pricing.rb +0 -11
- data/lib/llm_cost_tracker/railtie.rb +2 -1
- data/lib/llm_cost_tracker/report.rb +0 -5
- data/lib/llm_cost_tracker/storage/active_record_store.rb +10 -11
- data/lib/llm_cost_tracker/stream_collector.rb +17 -13
- data/lib/llm_cost_tracker/tags_column.rb +4 -0
- data/lib/llm_cost_tracker/tracker.rb +10 -2
- data/lib/llm_cost_tracker/version.rb +1 -1
- data/lib/llm_cost_tracker.rb +6 -14
- data/llm_cost_tracker.gemspec +3 -1
- metadata +37 -1
|
@@ -4,9 +4,6 @@ module LlmCostTracker
|
|
|
4
4
|
module Dashboard
|
|
5
5
|
ProviderRow = Data.define(:provider, :calls, :total_cost, :share_percent)
|
|
6
6
|
|
|
7
|
-
# Aggregates cost and call counts per provider for a given scope.
|
|
8
|
-
# Sorted by total cost descending; providers with zero cost fall to the bottom
|
|
9
|
-
# but are still returned so users can see calls without pricing.
|
|
10
7
|
class ProviderBreakdown
|
|
11
8
|
def self.call(scope: LlmCostTracker::LlmApiCall.all)
|
|
12
9
|
new(scope: scope).rows
|
|
@@ -9,8 +9,6 @@ module LlmCostTracker
|
|
|
9
9
|
:average_cost_per_call
|
|
10
10
|
)
|
|
11
11
|
|
|
12
|
-
# Aggregates calls grouped by the distinct values of a single tag key.
|
|
13
|
-
# Invalid keys raise InvalidFilterError so controllers can return HTTP 400.
|
|
14
12
|
class TagBreakdown
|
|
15
13
|
class << self
|
|
16
14
|
def call(key:, scope: LlmCostTracker::LlmApiCall.all)
|
|
@@ -9,21 +9,13 @@ module LlmCostTracker
|
|
|
9
9
|
attr_reader :page, :per
|
|
10
10
|
|
|
11
11
|
def self.call(params)
|
|
12
|
-
params =
|
|
12
|
+
params = LlmCostTracker::ParameterHash.with_indifferent_access(params)
|
|
13
13
|
new(
|
|
14
14
|
page: integer_param(params, :page, default: MIN_PAGE, min: MIN_PAGE),
|
|
15
15
|
per: integer_param(params, :per, default: DEFAULT_PER, min: 1, max: MAX_PER)
|
|
16
16
|
)
|
|
17
17
|
end
|
|
18
18
|
|
|
19
|
-
def self.normalize_params(params)
|
|
20
|
-
return {}.with_indifferent_access if params.nil?
|
|
21
|
-
|
|
22
|
-
raw = params.respond_to?(:to_unsafe_h) ? params.to_unsafe_h : params.to_h
|
|
23
|
-
raw.with_indifferent_access
|
|
24
|
-
end
|
|
25
|
-
private_class_method :normalize_params
|
|
26
|
-
|
|
27
19
|
def self.integer_param(params, key, default:, min:, max: nil)
|
|
28
20
|
value = Integer(params[key], 10)
|
|
29
21
|
value = [value, min].max
|
|
@@ -6,7 +6,7 @@
|
|
|
6
6
|
<title>LLM Cost Tracker</title>
|
|
7
7
|
<%= stylesheet_link_tag stylesheet_path %>
|
|
8
8
|
</head>
|
|
9
|
-
<body>
|
|
9
|
+
<body class="lct-body">
|
|
10
10
|
<div class="lct-app">
|
|
11
11
|
<main class="lct-shell">
|
|
12
12
|
<header class="lct-header">
|
|
@@ -25,20 +25,5 @@
|
|
|
25
25
|
<%= yield %>
|
|
26
26
|
</main>
|
|
27
27
|
</div>
|
|
28
|
-
<script>
|
|
29
|
-
document.addEventListener("keydown", function(event) {
|
|
30
|
-
if (event.key !== "/" || event.defaultPrevented || event.metaKey || event.ctrlKey || event.altKey) return;
|
|
31
|
-
|
|
32
|
-
var target = event.target;
|
|
33
|
-
if (target && (target.isContentEditable || /^(INPUT|TEXTAREA|SELECT)$/.test(target.tagName))) return;
|
|
34
|
-
|
|
35
|
-
var input = document.querySelector("[data-lct-filter-input]");
|
|
36
|
-
if (!input) return;
|
|
37
|
-
|
|
38
|
-
event.preventDefault();
|
|
39
|
-
input.focus();
|
|
40
|
-
if (typeof input.select === "function") input.select();
|
|
41
|
-
});
|
|
42
|
-
</script>
|
|
43
28
|
</body>
|
|
44
29
|
</html>
|
|
@@ -10,7 +10,7 @@
|
|
|
10
10
|
<div class="lct-filter-row lct-filter-row-with-sort">
|
|
11
11
|
<div class="lct-field">
|
|
12
12
|
<label for="lct-from">From</label>
|
|
13
|
-
<input id="lct-from"
|
|
13
|
+
<input id="lct-from" type="date" name="from" value="<%= params[:from] %>">
|
|
14
14
|
</div>
|
|
15
15
|
|
|
16
16
|
<div class="lct-field">
|
|
@@ -46,16 +46,17 @@
|
|
|
46
46
|
|
|
47
47
|
<div class="lct-field">
|
|
48
48
|
<label for="lct-sort">Sort</label>
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
49
|
+
<%= select_tag :sort,
|
|
50
|
+
options_for_select(
|
|
51
|
+
[["Recent first", ""],
|
|
52
|
+
["Most expensive", "expensive"],
|
|
53
|
+
["Largest input", "input"],
|
|
54
|
+
["Largest output", "output"]] +
|
|
55
|
+
(@latency_available ? [["Slowest", "slow"]] : []) +
|
|
56
|
+
[["Unknown pricing only", "unknown_pricing"]],
|
|
57
|
+
@sort
|
|
58
|
+
),
|
|
59
|
+
id: "lct-sort" %>
|
|
59
60
|
</div>
|
|
60
61
|
|
|
61
62
|
<div class="lct-filter-actions">
|
|
@@ -112,7 +113,6 @@
|
|
|
112
113
|
</thead>
|
|
113
114
|
<tbody>
|
|
114
115
|
<% @calls.each do |call| %>
|
|
115
|
-
<% tags = call.parsed_tags %>
|
|
116
116
|
<tr>
|
|
117
117
|
<td class="lct-nowrap"><%= format_date(call.tracked_at) %></td>
|
|
118
118
|
<td><%= call.provider %></td>
|
|
@@ -124,7 +124,7 @@
|
|
|
124
124
|
<% if @latency_available %>
|
|
125
125
|
<td class="lct-num<%= ' lct-num-muted' if call.latency_ms.nil? %>"><%= call.latency_ms ? "#{number(call.latency_ms)}ms" : "n/a" %></td>
|
|
126
126
|
<% end %>
|
|
127
|
-
<td><%= render "llm_cost_tracker/shared/tag_chips", tags:
|
|
127
|
+
<td><%= render "llm_cost_tracker/shared/tag_chips", tags: call.parsed_tags %></td>
|
|
128
128
|
<td><%= link_to "Details", call_path(call), class: "lct-button lct-button-secondary lct-button-compact" %></td>
|
|
129
129
|
</tr>
|
|
130
130
|
<% end %>
|
|
@@ -73,6 +73,11 @@
|
|
|
73
73
|
<dt>Pricing Status</dt>
|
|
74
74
|
<dd><%= pricing_status(@call) %></dd>
|
|
75
75
|
|
|
76
|
+
<% if LlmCostTracker::LlmApiCall.provider_response_id_column? %>
|
|
77
|
+
<dt>Provider Response ID</dt>
|
|
78
|
+
<dd><%= @call.provider_response_id.presence || "n/a" %></dd>
|
|
79
|
+
<% end %>
|
|
80
|
+
|
|
76
81
|
<% if @call.has_attribute?("created_at") %>
|
|
77
82
|
<dt>Created At</dt>
|
|
78
83
|
<dd><%= format_date(@call.created_at) %></dd>
|
|
@@ -113,12 +118,12 @@
|
|
|
113
118
|
|
|
114
119
|
<section class="lct-panel">
|
|
115
120
|
<h2 class="lct-section-title">Tags</h2>
|
|
116
|
-
<pre class="lct-pre"><%= safe_json(@
|
|
121
|
+
<pre class="lct-pre"><%= safe_json(@call.parsed_tags) %></pre>
|
|
117
122
|
</section>
|
|
118
123
|
|
|
119
|
-
<% if @
|
|
124
|
+
<% if @call.has_attribute?("metadata") %>
|
|
120
125
|
<section class="lct-panel">
|
|
121
126
|
<h2 class="lct-section-title">Metadata</h2>
|
|
122
|
-
<pre class="lct-pre"><%= safe_json(@metadata) %></pre>
|
|
127
|
+
<pre class="lct-pre"><%= safe_json(@call.read_attribute("metadata")) %></pre>
|
|
123
128
|
</section>
|
|
124
129
|
<% end %>
|
|
@@ -5,7 +5,7 @@
|
|
|
5
5
|
<div class="lct-filter-row lct-filter-row-basic">
|
|
6
6
|
<div class="lct-field">
|
|
7
7
|
<label for="lct-overview-from">From</label>
|
|
8
|
-
<input id="lct-overview-from"
|
|
8
|
+
<input id="lct-overview-from" type="date" name="from" value="<%= params[:from] || @from_date.iso8601 %>">
|
|
9
9
|
</div>
|
|
10
10
|
|
|
11
11
|
<div class="lct-field">
|
|
@@ -1,10 +1,7 @@
|
|
|
1
1
|
<% total = @stats.total_calls %>
|
|
2
|
-
<% known_pricing_calls = total - @stats.unknown_pricing_count %>
|
|
3
|
-
<% tagged_calls = total - @stats.untagged_calls_count %>
|
|
4
|
-
<% latency_calls = @stats.latency_column_present ? total - @stats.missing_latency_count : nil %>
|
|
5
2
|
<% streaming_count = @stats.streaming_count %>
|
|
6
3
|
<% streaming_missing_usage = @stats.streaming_missing_usage_count %>
|
|
7
|
-
<%
|
|
4
|
+
<% calls_with_provider_response_id = @stats.provider_response_id_column_present ? total - @stats.missing_provider_response_id_count : nil %>
|
|
8
5
|
|
|
9
6
|
<section class="lct-panel lct-toolbar">
|
|
10
7
|
<div class="lct-toolbar-head">
|
|
@@ -15,7 +12,7 @@
|
|
|
15
12
|
<div class="lct-filter-row lct-filter-row-basic">
|
|
16
13
|
<div class="lct-field">
|
|
17
14
|
<label for="lct-quality-from">From</label>
|
|
18
|
-
<input id="lct-quality-from"
|
|
15
|
+
<input id="lct-quality-from" type="date" name="from" value="<%= params[:from] %>">
|
|
19
16
|
</div>
|
|
20
17
|
|
|
21
18
|
<div class="lct-field">
|
|
@@ -113,6 +110,14 @@
|
|
|
113
110
|
</article>
|
|
114
111
|
<% end %>
|
|
115
112
|
<% end %>
|
|
113
|
+
|
|
114
|
+
<% if @stats.provider_response_id_column_present %>
|
|
115
|
+
<article class="lct-stat">
|
|
116
|
+
<p class="lct-stat-label">Calls with provider response ID</p>
|
|
117
|
+
<p class="lct-stat-value"><%= number(calls_with_provider_response_id) %></p>
|
|
118
|
+
<p class="lct-stat-sub"><%= percent(coverage_percent(calls_with_provider_response_id, total)) %> of calls</p>
|
|
119
|
+
</article>
|
|
120
|
+
<% end %>
|
|
116
121
|
</div>
|
|
117
122
|
</div>
|
|
118
123
|
</section>
|
|
@@ -136,41 +141,51 @@
|
|
|
136
141
|
</tr>
|
|
137
142
|
</thead>
|
|
138
143
|
<tbody>
|
|
139
|
-
<% cost_coverage = coverage_percent(
|
|
144
|
+
<% cost_coverage = coverage_percent(total - @stats.unknown_pricing_count, total) %>
|
|
140
145
|
<tr>
|
|
141
146
|
<td>Cost (pricing known)</td>
|
|
142
147
|
<td class="lct-num"><%= percent(cost_coverage) %></td>
|
|
143
|
-
<td class="lct-num"><%= number(
|
|
148
|
+
<td class="lct-num"><%= number(total - @stats.unknown_pricing_count) %></td>
|
|
144
149
|
<td><%= render "llm_cost_tracker/shared/bar", value: cost_coverage, max: 100.0 %></td>
|
|
145
150
|
</tr>
|
|
146
151
|
|
|
147
|
-
<% tag_coverage = coverage_percent(
|
|
152
|
+
<% tag_coverage = coverage_percent(total - @stats.untagged_calls_count, total) %>
|
|
148
153
|
<tr>
|
|
149
154
|
<td>Tags (at least one tag)</td>
|
|
150
155
|
<td class="lct-num"><%= percent(tag_coverage) %></td>
|
|
151
|
-
<td class="lct-num"><%= number(
|
|
156
|
+
<td class="lct-num"><%= number(total - @stats.untagged_calls_count) %></td>
|
|
152
157
|
<td><%= render "llm_cost_tracker/shared/bar", value: tag_coverage, max: 100.0 %></td>
|
|
153
158
|
</tr>
|
|
154
159
|
|
|
155
160
|
<% if @stats.latency_column_present %>
|
|
156
|
-
<% latency_coverage = coverage_percent(
|
|
161
|
+
<% latency_coverage = coverage_percent(total - @stats.missing_latency_count, total) %>
|
|
157
162
|
<tr>
|
|
158
163
|
<td>Latency</td>
|
|
159
164
|
<td class="lct-num"><%= percent(latency_coverage) %></td>
|
|
160
|
-
<td class="lct-num"><%= number(
|
|
165
|
+
<td class="lct-num"><%= number(total - @stats.missing_latency_count) %></td>
|
|
161
166
|
<td><%= render "llm_cost_tracker/shared/bar", value: latency_coverage, max: 100.0 %></td>
|
|
162
167
|
</tr>
|
|
163
168
|
<% end %>
|
|
164
169
|
|
|
165
|
-
<% if @stats.stream_column_present &&
|
|
166
|
-
<% stream_coverage = coverage_percent(
|
|
170
|
+
<% if @stats.stream_column_present && streaming_count.to_i.positive? && streaming_missing_usage %>
|
|
171
|
+
<% stream_coverage = coverage_percent(streaming_count - streaming_missing_usage, streaming_count) %>
|
|
167
172
|
<tr>
|
|
168
173
|
<td>Streaming usage captured</td>
|
|
169
174
|
<td class="lct-num"><%= percent(stream_coverage) %></td>
|
|
170
|
-
<td class="lct-num"><%= number(
|
|
175
|
+
<td class="lct-num"><%= number(streaming_count - streaming_missing_usage) %> / <%= number(streaming_count) %></td>
|
|
171
176
|
<td><%= render "llm_cost_tracker/shared/bar", value: stream_coverage, max: 100.0 %></td>
|
|
172
177
|
</tr>
|
|
173
178
|
<% end %>
|
|
179
|
+
|
|
180
|
+
<% if @stats.provider_response_id_column_present %>
|
|
181
|
+
<% provider_response_id_coverage = coverage_percent(calls_with_provider_response_id, total) %>
|
|
182
|
+
<tr>
|
|
183
|
+
<td>Provider response ID</td>
|
|
184
|
+
<td class="lct-num"><%= percent(provider_response_id_coverage) %></td>
|
|
185
|
+
<td class="lct-num"><%= number(calls_with_provider_response_id) %></td>
|
|
186
|
+
<td><%= render "llm_cost_tracker/shared/bar", value: provider_response_id_coverage, max: 100.0 %></td>
|
|
187
|
+
</tr>
|
|
188
|
+
<% end %>
|
|
174
189
|
</tbody>
|
|
175
190
|
</table>
|
|
176
191
|
</section>
|
|
@@ -216,6 +231,13 @@
|
|
|
216
231
|
<td>Send OpenAI requests with <code class="lct-code">stream_options: { include_usage: true }</code>, or wrap custom clients with <code class="lct-code">LlmCostTracker.track_stream</code>.</td>
|
|
217
232
|
</tr>
|
|
218
233
|
<% end %>
|
|
234
|
+
<% if @stats.provider_response_id_column_present && @stats.missing_provider_response_id_count.to_i.positive? %>
|
|
235
|
+
<tr>
|
|
236
|
+
<td>Missing provider response IDs</td>
|
|
237
|
+
<td>Proof of provider-issued responses is weaker when calls cannot be tied back to provider objects.</td>
|
|
238
|
+
<td>Upgrade to the latest parser coverage and pass <code class="lct-code">provider_response_id:</code> for custom clients when the provider exposes one.</td>
|
|
239
|
+
</tr>
|
|
240
|
+
<% end %>
|
|
219
241
|
</tbody>
|
|
220
242
|
</table>
|
|
221
243
|
</section>
|
|
@@ -7,7 +7,7 @@
|
|
|
7
7
|
<div class="lct-filter-row lct-filter-row-with-sort">
|
|
8
8
|
<div class="lct-field">
|
|
9
9
|
<label for="lct-models-from">From</label>
|
|
10
|
-
<input id="lct-models-from"
|
|
10
|
+
<input id="lct-models-from" type="date" name="from" value="<%= params[:from] %>">
|
|
11
11
|
</div>
|
|
12
12
|
|
|
13
13
|
<div class="lct-field">
|
|
@@ -33,14 +33,15 @@
|
|
|
33
33
|
|
|
34
34
|
<div class="lct-field">
|
|
35
35
|
<label for="lct-models-sort">Sort</label>
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
|
|
36
|
+
<%= select_tag :sort,
|
|
37
|
+
options_for_select(
|
|
38
|
+
[["Total spend", "cost"],
|
|
39
|
+
["Call volume", "calls"],
|
|
40
|
+
["Avg cost / call", "avg_cost"]] +
|
|
41
|
+
(@latency_available ? [["Avg latency", "latency"]] : []),
|
|
42
|
+
@sort.presence || "cost"
|
|
43
|
+
),
|
|
44
|
+
id: "lct-models-sort" %>
|
|
44
45
|
</div>
|
|
45
46
|
|
|
46
47
|
<div class="lct-filter-actions">
|
|
@@ -7,7 +7,7 @@
|
|
|
7
7
|
<div class="lct-filter-row lct-filter-row-basic">
|
|
8
8
|
<div class="lct-field">
|
|
9
9
|
<label for="lct-tags-from">From</label>
|
|
10
|
-
<input id="lct-tags-from"
|
|
10
|
+
<input id="lct-tags-from" type="date" name="from" value="<%= params[:from] %>">
|
|
11
11
|
</div>
|
|
12
12
|
|
|
13
13
|
<div class="lct-field">
|
|
@@ -12,7 +12,7 @@
|
|
|
12
12
|
<div class="lct-filter-row lct-filter-row-basic">
|
|
13
13
|
<div class="lct-field">
|
|
14
14
|
<label for="lct-tag-show-from">From</label>
|
|
15
|
-
<input id="lct-tag-show-from"
|
|
15
|
+
<input id="lct-tag-show-from" type="date" name="from" value="<%= params[:from] %>">
|
|
16
16
|
</div>
|
|
17
17
|
|
|
18
18
|
<div class="lct-field">
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
# frozen_string_literal: true
|
|
2
|
+
|
|
3
|
+
require "rails/generators"
|
|
4
|
+
require "rails/generators/active_record"
|
|
5
|
+
|
|
6
|
+
module LlmCostTracker
|
|
7
|
+
module Generators
|
|
8
|
+
class AddProviderResponseIdGenerator < Rails::Generators::Base
|
|
9
|
+
include ActiveRecord::Generators::Migration
|
|
10
|
+
|
|
11
|
+
source_root File.expand_path("templates", __dir__)
|
|
12
|
+
|
|
13
|
+
desc "Creates a migration to add llm_api_calls.provider_response_id"
|
|
14
|
+
|
|
15
|
+
def create_migration_file
|
|
16
|
+
migration_template(
|
|
17
|
+
"add_provider_response_id_to_llm_api_calls.rb.erb",
|
|
18
|
+
"db/migrate/add_provider_response_id_to_llm_api_calls.rb"
|
|
19
|
+
)
|
|
20
|
+
end
|
|
21
|
+
|
|
22
|
+
private
|
|
23
|
+
|
|
24
|
+
def migration_version
|
|
25
|
+
"[#{ActiveRecord::VERSION::MAJOR}.#{ActiveRecord::VERSION::MINOR}]"
|
|
26
|
+
end
|
|
27
|
+
end
|
|
28
|
+
end
|
|
29
|
+
end
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
class AddProviderResponseIdToLlmApiCalls < ActiveRecord::Migration<%= migration_version %>
|
|
2
|
+
def up
|
|
3
|
+
return if column_exists?(:llm_api_calls, :provider_response_id)
|
|
4
|
+
|
|
5
|
+
add_column :llm_api_calls, :provider_response_id, :string
|
|
6
|
+
add_index :llm_api_calls, :provider_response_id
|
|
7
|
+
end
|
|
8
|
+
|
|
9
|
+
def down
|
|
10
|
+
return unless column_exists?(:llm_api_calls, :provider_response_id)
|
|
11
|
+
|
|
12
|
+
remove_index :llm_api_calls, :provider_response_id if index_exists?(:llm_api_calls, :provider_response_id)
|
|
13
|
+
remove_column :llm_api_calls, :provider_response_id
|
|
14
|
+
end
|
|
15
|
+
end
|
|
@@ -12,6 +12,7 @@ class CreateLlmApiCalls < ActiveRecord::Migration<%= migration_version %>
|
|
|
12
12
|
t.integer :latency_ms
|
|
13
13
|
t.boolean :stream, null: false, default: false
|
|
14
14
|
t.string :usage_source
|
|
15
|
+
t.string :provider_response_id
|
|
15
16
|
if postgresql?
|
|
16
17
|
t.jsonb :tags, null: false, default: {}
|
|
17
18
|
else
|
|
@@ -28,6 +29,7 @@ class CreateLlmApiCalls < ActiveRecord::Migration<%= migration_version %>
|
|
|
28
29
|
add_index :llm_api_calls, [:provider, :tracked_at]
|
|
29
30
|
add_index :llm_api_calls, :stream
|
|
30
31
|
add_index :llm_api_calls, :usage_source
|
|
32
|
+
add_index :llm_api_calls, :provider_response_id
|
|
31
33
|
add_index :llm_api_calls, :tags, using: :gin if postgresql?
|
|
32
34
|
end
|
|
33
35
|
|
|
@@ -16,7 +16,6 @@ module LlmCostTracker
|
|
|
16
16
|
|
|
17
17
|
self.table_name = "llm_api_calls"
|
|
18
18
|
|
|
19
|
-
# Scopes for querying
|
|
20
19
|
scope :with_cost, -> { where.not(total_cost: nil) }
|
|
21
20
|
scope :without_cost, -> { where(total_cost: nil) }
|
|
22
21
|
scope :unknown_pricing, -> { without_cost }
|
|
@@ -24,6 +23,12 @@ module LlmCostTracker
|
|
|
24
23
|
scope :streaming, -> { stream_column? ? where(stream: true) : none }
|
|
25
24
|
scope :non_streaming, -> { stream_column? ? where(stream: [false, nil]) : all }
|
|
26
25
|
scope :by_usage_source, ->(source) { usage_source_column? ? where(usage_source: source.to_s) : none }
|
|
26
|
+
scope :with_provider_response_id, lambda {
|
|
27
|
+
provider_response_id_column? ? where.not(provider_response_id: [nil, ""]) : none
|
|
28
|
+
}
|
|
29
|
+
scope :missing_provider_response_id, lambda {
|
|
30
|
+
provider_response_id_column? ? where(provider_response_id: [nil, ""]) : none
|
|
31
|
+
}
|
|
27
32
|
scope :streaming_missing_usage, lambda {
|
|
28
33
|
return none unless stream_column? && usage_source_column?
|
|
29
34
|
|
|
@@ -51,7 +56,6 @@ module LlmCostTracker
|
|
|
51
56
|
TagQuery.apply(self, tags)
|
|
52
57
|
end
|
|
53
58
|
|
|
54
|
-
# Aggregations
|
|
55
59
|
def self.total_cost
|
|
56
60
|
sum(:total_cost).to_f
|
|
57
61
|
end
|
|
@@ -61,6 +61,7 @@ module LlmCostTracker
|
|
|
61
61
|
latency_ms: latency_ms,
|
|
62
62
|
stream: parsed.stream,
|
|
63
63
|
usage_source: parsed.usage_source,
|
|
64
|
+
provider_response_id: parsed.provider_response_id,
|
|
64
65
|
metadata: resolved_tags(request_env).merge(parsed.metadata)
|
|
65
66
|
)
|
|
66
67
|
rescue LlmCostTracker::Error
|
|
@@ -0,0 +1,33 @@
|
|
|
1
|
+
# frozen_string_literal: true
|
|
2
|
+
|
|
3
|
+
module LlmCostTracker
|
|
4
|
+
module ParameterHash
|
|
5
|
+
class << self
|
|
6
|
+
def hash_like?(value)
|
|
7
|
+
value.is_a?(Hash) || action_controller_parameters?(value)
|
|
8
|
+
end
|
|
9
|
+
|
|
10
|
+
def to_hash(value)
|
|
11
|
+
return {} if value.nil?
|
|
12
|
+
return value.to_unsafe_h if action_controller_parameters?(value)
|
|
13
|
+
return value.to_h if value.is_a?(Hash)
|
|
14
|
+
return {} unless value.respond_to?(:to_h)
|
|
15
|
+
|
|
16
|
+
hash = value.to_h
|
|
17
|
+
hash.is_a?(Hash) ? hash : {}
|
|
18
|
+
rescue ArgumentError, TypeError
|
|
19
|
+
{}
|
|
20
|
+
end
|
|
21
|
+
|
|
22
|
+
def with_indifferent_access(value)
|
|
23
|
+
to_hash(value).with_indifferent_access
|
|
24
|
+
end
|
|
25
|
+
|
|
26
|
+
private
|
|
27
|
+
|
|
28
|
+
def action_controller_parameters?(value)
|
|
29
|
+
defined?(ActionController::Parameters) && value.is_a?(ActionController::Parameters)
|
|
30
|
+
end
|
|
31
|
+
end
|
|
32
|
+
end
|
|
33
|
+
end
|
|
@@ -12,11 +12,21 @@ module LlmCostTracker
|
|
|
12
12
|
:cache_creation_input_tokens,
|
|
13
13
|
:reasoning_tokens,
|
|
14
14
|
:stream,
|
|
15
|
-
:usage_source
|
|
15
|
+
:usage_source,
|
|
16
|
+
:provider_response_id
|
|
16
17
|
)
|
|
17
18
|
|
|
18
19
|
class ParsedUsage
|
|
19
|
-
TRACKING_KEYS = %i[
|
|
20
|
+
TRACKING_KEYS = %i[
|
|
21
|
+
provider
|
|
22
|
+
model
|
|
23
|
+
input_tokens
|
|
24
|
+
output_tokens
|
|
25
|
+
total_tokens
|
|
26
|
+
stream
|
|
27
|
+
usage_source
|
|
28
|
+
provider_response_id
|
|
29
|
+
].freeze
|
|
20
30
|
|
|
21
31
|
def self.build(**attributes)
|
|
22
32
|
new(
|
|
@@ -30,7 +40,8 @@ module LlmCostTracker
|
|
|
30
40
|
cache_creation_input_tokens: attributes[:cache_creation_input_tokens],
|
|
31
41
|
reasoning_tokens: attributes[:reasoning_tokens],
|
|
32
42
|
stream: attributes[:stream] || false,
|
|
33
|
-
usage_source: attributes[:usage_source]
|
|
43
|
+
usage_source: attributes[:usage_source],
|
|
44
|
+
provider_response_id: attributes[:provider_response_id]
|
|
34
45
|
)
|
|
35
46
|
end
|
|
36
47
|
|
|
@@ -31,6 +31,7 @@ module LlmCostTracker
|
|
|
31
31
|
|
|
32
32
|
ParsedUsage.build(
|
|
33
33
|
provider: "anthropic",
|
|
34
|
+
provider_response_id: response["id"],
|
|
34
35
|
model: response["model"] || request["model"],
|
|
35
36
|
input_tokens: usage["input_tokens"].to_i,
|
|
36
37
|
output_tokens: usage["output_tokens"].to_i,
|
|
@@ -48,35 +49,9 @@ module LlmCostTracker
|
|
|
48
49
|
request = safe_json_parse(request_body)
|
|
49
50
|
model = stream_model(events) || request["model"]
|
|
50
51
|
usage = stream_usage(events)
|
|
52
|
+
response_id = stream_response_id(events)
|
|
51
53
|
|
|
52
|
-
|
|
53
|
-
input = usage["input_tokens"].to_i
|
|
54
|
-
output = usage["output_tokens"].to_i
|
|
55
|
-
cache_read = usage["cache_read_input_tokens"].to_i
|
|
56
|
-
cache_creation = usage["cache_creation_input_tokens"].to_i
|
|
57
|
-
|
|
58
|
-
ParsedUsage.build(
|
|
59
|
-
provider: "anthropic",
|
|
60
|
-
model: model,
|
|
61
|
-
input_tokens: input,
|
|
62
|
-
output_tokens: output,
|
|
63
|
-
total_tokens: input + output + cache_read + cache_creation,
|
|
64
|
-
cache_read_input_tokens: usage["cache_read_input_tokens"],
|
|
65
|
-
cache_creation_input_tokens: usage["cache_creation_input_tokens"],
|
|
66
|
-
stream: true,
|
|
67
|
-
usage_source: :stream_final
|
|
68
|
-
)
|
|
69
|
-
else
|
|
70
|
-
ParsedUsage.build(
|
|
71
|
-
provider: "anthropic",
|
|
72
|
-
model: model,
|
|
73
|
-
input_tokens: 0,
|
|
74
|
-
output_tokens: 0,
|
|
75
|
-
total_tokens: 0,
|
|
76
|
-
stream: true,
|
|
77
|
-
usage_source: :unknown
|
|
78
|
-
)
|
|
79
|
-
end
|
|
54
|
+
usage ? build_stream_result(model, usage, response_id) : build_unknown_stream_result(model, response_id)
|
|
80
55
|
end
|
|
81
56
|
|
|
82
57
|
private
|
|
@@ -114,6 +89,50 @@ module LlmCostTracker
|
|
|
114
89
|
end
|
|
115
90
|
nil
|
|
116
91
|
end
|
|
92
|
+
|
|
93
|
+
def stream_response_id(events)
|
|
94
|
+
events.each do |event|
|
|
95
|
+
data = event[:data]
|
|
96
|
+
next unless data.is_a?(Hash)
|
|
97
|
+
|
|
98
|
+
id = data.dig("message", "id") || data["id"]
|
|
99
|
+
return id if id && !id.to_s.empty?
|
|
100
|
+
end
|
|
101
|
+
nil
|
|
102
|
+
end
|
|
103
|
+
|
|
104
|
+
def build_stream_result(model, usage, response_id)
|
|
105
|
+
input = usage["input_tokens"].to_i
|
|
106
|
+
output = usage["output_tokens"].to_i
|
|
107
|
+
cache_read = usage["cache_read_input_tokens"].to_i
|
|
108
|
+
cache_creation = usage["cache_creation_input_tokens"].to_i
|
|
109
|
+
|
|
110
|
+
ParsedUsage.build(
|
|
111
|
+
provider: "anthropic",
|
|
112
|
+
provider_response_id: response_id,
|
|
113
|
+
model: model,
|
|
114
|
+
input_tokens: input,
|
|
115
|
+
output_tokens: output,
|
|
116
|
+
total_tokens: input + output + cache_read + cache_creation,
|
|
117
|
+
cache_read_input_tokens: usage["cache_read_input_tokens"],
|
|
118
|
+
cache_creation_input_tokens: usage["cache_creation_input_tokens"],
|
|
119
|
+
stream: true,
|
|
120
|
+
usage_source: :stream_final
|
|
121
|
+
)
|
|
122
|
+
end
|
|
123
|
+
|
|
124
|
+
def build_unknown_stream_result(model, response_id)
|
|
125
|
+
ParsedUsage.build(
|
|
126
|
+
provider: "anthropic",
|
|
127
|
+
provider_response_id: response_id,
|
|
128
|
+
model: model,
|
|
129
|
+
input_tokens: 0,
|
|
130
|
+
output_tokens: 0,
|
|
131
|
+
total_tokens: 0,
|
|
132
|
+
stream: true,
|
|
133
|
+
usage_source: :unknown
|
|
134
|
+
)
|
|
135
|
+
end
|
|
117
136
|
end
|
|
118
137
|
end
|
|
119
138
|
end
|