T3 Code usage hierarchy

Provider token volume first, costs and token detail before behavior

Current window
30 days
Processed
19.9B
Input
2.2B
Output
7.2M
Turns
2,092
Token and activity totals use the current bb-1 snapshot. Cost values are marked estimated and illustrate the ccusage-style pricing layer still needed for machine-wide ingestion.
A

Provider Volume

The current direction, corrected. Daily token volume is the hero and provider identity is immediate.

Best default. Costs and token totals stay above the chart. Token accounting and provider totals follow. Skills move below the primary analytical layer.

T3Code
Usage
● 1 connected

Usage

Jul 8 to Aug 7 · bb-1 reporting

All environments⌄30 days⌄↻
$1,284.6196% priced · ccusage rates
19.9Bacross all provider sessions
14.4B$438.16 estimated · 72.4%
5.5B$846.45 estimated · 27.6%
7.2M4.3M visible · 2.9M reasoning

Daily processed tokens

TOKENSCOSTTURNS
CodexClaude Code
Peak day1.30B · Aug 2
Daily average663M
Input observed2.2B
Cache reads2.12B
Output observed7.2M

Token accounting

observed snapshots
Cached input
2.12B
Uncached input
76.7M
Output
4.3M
Reasoning
2.9M

Provider totals

cost · tokens · turns
ProviderCostTokensTurns
Codex$438.1614.4B470
Claude Code$846.455.5B1,622
Projectst3code · 2,047 turns
Skills and toolsfurther down the page ↓
B

Cost First

Financial answer first, provider token area chart second, detailed token and cost accounting immediately below.

Best if the page should answer “what did my agents cost?” before anything else. The provider chart remains dominant but shares the first viewport with pricing coverage.

T3Code
Usage
● 1 connected

Usage

Jul 8 to Aug 7 · estimated pricing coverage 96%

All environments⌄30 days⌄↻
$1,284.61+$196.38 vs previous 30 days · exact where reported
Claude Code$846.45
65.9% of spend · 5.5B tokens
Codex$438.16
34.1% of spend · 14.4B tokens

Daily processed tokens

TOKENSCOST
CodexClaude Code
19.9B663M daily average
2.12B96.5% of observed input
76.7M$0.38 per million average
4.3Mplus 2.9M reasoning
221h2,092 turns · 98.5% complete

Recent daily cost

estimated · click through to turns
DayCodexClaude CodeTotalTokens
Aug 7$18.21$42.89$61.101.01B
Aug 6$14.83$36.10$50.93921M
Aug 5$21.04$27.66$48.70804M
Aug 4$12.91$31.22$44.13738M

Cost quality

coverage
Provider reported18.2%
Model priced77.8%
Unpriced sessions4.0%
Cache savings$392.08
Projects and modelsnext section ↓
Skills and toolsbelow financial detail ↓
C

Provider Compare

Codex and Claude Code get equal analytical weight before a shared area chart and token matrix.

Best when the interesting question is “how are these providers different?” It exposes spend, token share, turns, and cost per turn without promoting skills into the first viewport.

T3Code
Usage
● 1 connected

Usage

Provider comparison · Jul 8 to Aug 7

bb-1⌄30 days⌄↻
$1,284.61$42.82 per active day
19.9B663M daily average
2.2B2.12B cached
4.3M2.9M reasoning
2,09298.5% completion
Codex 72.4% of tokens
Tokens14.4B
Est. cost$438.16
Cost / turn$0.93
Claude Code 27.6% of tokens
Tokens5.5B
Est. cost$846.45
Cost / turn$0.52

Daily processed tokens

ABSOLUTESHARECOST
CodexClaude Code

Provider token matrix

observed snapshots · cost-aware categories
Token type
Codex
Claude Code
Total
Cached input
1.54B
580M
2.12B
Uncached input
21.4M
55.3M
76.7M
Output
1.1M
3.2M
4.3M
Reasoning
2.4M
0.5M
2.9M
Models and projectsprovider-specific next ↓
Skills and toolsintentionally below usage data ↓
A

Control Room

Broad, visual, and immediately useful. The best default if Usage should answer “what is happening?” in ten seconds.

Global overview first, then environment, provider, agent, and skill breakdowns. This feels like a native top-level T3 Code page and is the most balanced direction.

T3Code
Usage
● 4 online⌘ K
Across every connected environment

Usage

Jul 9 to Aug 7 · updated 8s ago
All environments⌄
30 days⌄
UTC-7
Estimated spend
$1,284.61
↑ 18% vs previous 30d · estimated
Processed tokens
324.8M
91.4% input · 5.7% output · 2.9% reasoning
Turns completed
2,358
97.9% completion · 38 errors
Agent runtime
247.3h
6m 11s median · 18 concurrent peak

Spend and token volume

daily · all providers
CodexClaudeTokens
JUL 09JUL 23AUG 07

Environments

share of spend
Mac Studio
online · 212 threads
$873.22
MacBook Pro
online · 74 threads
$269.11
Build box
online · 32 threads
$102.87
Cloud sandbox
last seen 3h · 16 threads
$39.41
Coverage
3 exact · 1 estimated
76%

Skill preference

observed reads / calls
Codex
24 distinct
test-t3-app70
github67
babysit-pr40
browser40
html-communication38

Skill preference

explicit calls
Claude
18 distinct
babysit-pr33
file-pr32
html-planning29
html-communication27
codex-review22

Token mix

estimated billable
Cached input176.2M
Uncached input96.7M
Output35.8M
Reasoning16.1M
Cache savings$438.92
B

Usage Ledger

Precise, sortable, and financially legible. Best if Usage should feel trustworthy enough to reconcile.

A daily ledger is the primary object. Every aggregate can be traced to environment, project, provider, model, or thread. Less visual spectacle, more operational confidence.

T3Code
Usage / Ledger
● syncedExport CSV
Usage / Ledger

Everything accounted for.

4 environments · 2 providers · 12 models · 18 projects
Jul 9 to Aug 7
All environments⌄
All providers⌄
Monthly budget
$1,284.61 / $2,000.00
BY DAY   PROJECT   AGENT   MODEL
All activityExact cost onlyEstimatedAnomalies
⌕ filter rows
DaySpendTokensTurnsRuntime$/turn
Today, Aug 7$143.8231.4M21516.3h$0.67
Wed, Aug 6$119.0426.2M10515.8h$1.13
Tue, Aug 5$91.7219.7M7511.1h$1.22
Mon, Aug 4$128.3033.8M16310.9h$0.79
Sun, Aug 3$54.1613.3M565.0h$0.97
Sat, Aug 2$71.4917.9M798.7h$0.91
Fri, Aug 1$37.088.2M216.9h$1.77
Thu, Jul 31$82.1120.4M885.2h$0.93
Wed, Jul 30$106.5727.1M13519.6h$0.79
Tue, Jul 29$68.0217.2M875.5h$0.78
Mon, Jul 28$78.6620.1M1086.7h$0.73
Sun, Jul 27$120.2529.5M17011.6h$0.71
30d total$1,284.61
exact coverage76.2%
median turn6m 11s
error spend$21.44
C

Agent Atlas

Personal, comparative, and fun. Best if Usage should reveal how each agent actually works for you.

Agent behavior is the hero. Spend still exists, but the page is organized around skill signatures, tool habits, model choices, project affinity, and when your fleet comes alive.

T3Code
Usage / Agent Atlas
● 4 environmentsLast 30 days⌄
Your fleet, decoded

How your agents build.

Behavior derived from 2,408 turns and 50,855 tool starts across every connected environment. Preferences are observed actions, not guessed from message content.

$1,284.61estimated 30d spend · details →
Mac Studio68%
MacBook Pro21%
Build box8%
Cloud sandbox3%
CX
Codex
142 threads · 669 turns · 71.7h
27.9%of turns
Signature skills
test-t3-app70 reads
github67 reads
control-in-app-browser40 reads
babysit-pr40 reads
html-communication38 reads
Habits
Favorite toolshell · 20,712
Preferred effortxhigh · 43 threads
Visual checks39 image views
Web research253 searches
Shell first
20,712
Files changed
1,531
CL
Claude
185 threads · 1,733 turns · 175.6h
72.1%of turns
Signature skills
babysit-pr33 calls
file-pr32 calls
html-planning29 calls
html-communication27 calls
codex-review22 calls
Habits
Favorite toolshell · 18,149
Preferred modelfable-5 · 134
File changes5,413 starts
Subagents306 calls
Agent depth
306
File first
5,413

Active days

40 days in snapshot

Project affinity

by completed turns
t3code2,124
secondary workspace217
t3chat37
other30

When the fleet runs

hour of day · local
12a6a12p6p12a

Shared data model

● Environment-local facts

Each server summarizes turns, tokens, tools, skills, files, duration, provider, model, and exact or estimated cost into small daily buckets.

● Client-side merge

Desktop and web request the same typed usage summary from every connected environment, then merge by environment ID and time zone. Raw messages never need to cross environments.

● Data to add

Persist turn-completed usage and provider-reported cost. Version the summary contract so offline or older environments degrade to partial coverage instead of blocking the page.