1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
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
113
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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
//! Context use, token and cost totals, and session statistics.
use yapi_types::message::{ContentBlock, Message, StopReason};
use yapi_types::session::FileEntry;
use crate::compaction::{calculate_context_tokens, estimate_projected_context_tokens};
use crate::session::build_projection;
use super::AgentSession;
/// How full the context window is.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct ContextUsage {
/// Estimated context tokens, unknown right after a compaction.
pub tokens: Option<u64>,
/// The model's context window.
pub context_window: u64,
}
impl ContextUsage {
/// Percent of the window in use, when known.
pub fn percent(&self) -> Option<f64> {
self.tokens
.map(|tokens| tokens as f64 / self.context_window as f64 * 100.0)
}
}
/// Session-wide token and cost totals.
#[derive(Clone, Copy, Debug, Default, PartialEq)]
pub struct UsageTotals {
/// Input tokens.
pub input: u64,
/// Output tokens.
pub output: u64,
/// Cache read tokens.
pub cache_read: u64,
/// Cache write tokens.
pub cache_write: u64,
/// Cost in US dollars.
pub cost: f64,
/// Cache hit rate of the latest assistant message, in percent.
pub cache_hit_rate: Option<f64>,
}
/// Message counts and per-model cost, as `/session` shows them.
#[derive(Clone, Debug, Default, PartialEq)]
pub struct SessionStats {
/// Message entries.
pub total_messages: usize,
/// User messages.
pub user_messages: usize,
/// Assistant messages.
pub assistant_messages: usize,
/// Tool calls in assistant messages.
pub tool_calls: usize,
/// Tool results.
pub tool_results: usize,
/// Cost and tokens by `provider/model`, costliest first. Summaries and
/// tool usage are grouped as `Tools/summaries`.
pub breakdown: Vec<(String, f64, u64)>,
}
impl AgentSession {
/// Context use of the current branch: `tokens` is `None` after a compaction
/// until the model responds again. `None` without a model or window.
pub fn context_usage(&self) -> Option<ContextUsage> {
let model = self.model()?;
let context_window = model.context_window;
if context_window == 0 {
return None;
}
self.with_session(|session| {
let branch = session.branch_path(None);
let projection = build_projection(&branch);
let latest_compaction = branch
.iter()
.rposition(|entry| matches!(entry, FileEntry::Compaction(_)));
if let Some(compaction) = latest_compaction {
let has_usage = |entry: &&FileEntry| {
let Some(id) = entry.meta().map(|meta| meta.id.as_str()) else {
return false;
};
projection.entries.iter().any(|projected| {
projected.source.meta().is_some_and(|meta| meta.id == id)
&& projected.messages.iter().any(|message| {
matches!(message, Message::Assistant(assistant)
if !matches!(assistant.stop_reason, StopReason::Aborted | StopReason::Error)
&& calculate_context_tokens(&assistant.usage) > 0)
})
})
};
if !branch[compaction + 1..].iter().any(has_usage) {
return Some(ContextUsage {
tokens: None,
context_window,
});
}
}
let tokens = estimate_projected_context_tokens(&projection, &branch).tokens;
Some(ContextUsage {
tokens: Some(tokens),
context_window,
})
})
}
/// Token and cost totals over every entry of the session file, and the
/// cache hit rate of the latest assistant message.
pub fn usage_totals(&self) -> UsageTotals {
self.with_session(|session| {
let mut totals = UsageTotals::default();
for entry in session.entries() {
let usage = match entry {
FileEntry::Usage(entry) => Some(&entry.usage),
FileEntry::Message(entry) => match &entry.message {
Message::Assistant(assistant) => {
let usage = &assistant.usage;
let prompt = usage.input + usage.cache_read + usage.cache_write;
totals.cache_hit_rate = (prompt > 0)
.then(|| usage.cache_read as f64 / prompt as f64 * 100.0);
Some(usage)
}
Message::ToolResult(result) => result.usage.as_ref(),
_ => None,
},
FileEntry::Compaction(entry) => entry.usage.as_ref(),
FileEntry::BranchSummary(entry) => entry.usage.as_ref(),
_ => None,
};
if let Some(usage) = usage {
totals.input += usage.input;
totals.output += usage.output;
totals.cache_read += usage.cache_read;
totals.cache_write += usage.cache_write;
totals.cost += usage.cost.total;
}
}
totals
})
}
/// pi's `getSessionStats` counts and `getUsageCostBreakdown`, over every
/// entry of the session file.
pub fn session_stats(&self) -> SessionStats {
self.with_session(|session| {
let mut stats = SessionStats::default();
let mut breakdown: Vec<(String, f64, u64)> = Vec::new();
let mut add = |key: String, usage: &yapi_types::message::Usage| {
let tokens = usage.input + usage.output + usage.cache_read + usage.cache_write;
match breakdown
.iter_mut()
.find(|(existing, _, _)| *existing == key)
{
Some((_, cost, total)) => {
*cost += usage.cost.total;
*total += tokens;
}
None => breakdown.push((key, usage.cost.total, tokens)),
}
};
for entry in session.entries() {
match entry {
FileEntry::Usage(entry) => {
add(format!("{}/{}", entry.provider, entry.model), &entry.usage);
}
FileEntry::Compaction(entry) => {
if let Some(usage) = &entry.usage {
add("Tools/summaries".into(), usage);
}
}
FileEntry::BranchSummary(entry) => {
if let Some(usage) = &entry.usage {
add("Tools/summaries".into(), usage);
}
}
FileEntry::Message(entry) => {
stats.total_messages += 1;
match &entry.message {
Message::User(_) => stats.user_messages += 1,
Message::ToolResult(result) => {
stats.tool_results += 1;
if let Some(usage) = &result.usage {
add("Tools/summaries".into(), usage);
}
}
Message::Assistant(assistant) => {
stats.assistant_messages += 1;
stats.tool_calls += assistant
.content
.iter()
.filter(|block| matches!(block, ContentBlock::ToolCall(_)))
.count();
let model = assistant
.response_model
.as_deref()
.unwrap_or(&assistant.model);
add(format!("{}/{model}", assistant.provider), &assistant.usage);
}
_ => {}
}
}
_ => {}
}
}
breakdown.retain(|(_, cost, tokens)| *cost > 0.0 || *tokens > 0);
breakdown.sort_by(|a, b| b.1.total_cmp(&a.1));
stats.breakdown = breakdown;
stats
})
}
}