AI usage and estimated spend
Review AI usage evidence and estimated spend.
Claude CodeCodexCursorOpenCode
Find out what your agents actually cost.
Claude Code and Codex telemetry, broken down by user, model, repo, tool, skill and session. You get a number you can act on, the play-by-play behind it, and you choose what is ever stored.
estimated spend · trend
~$2,000
Trending up as adoption grows
projectedAI spend sits in the same workspace as your deploys and your audit trail.
One connected loop, held on the stage this capability serves. The other stages stay as context so you can see what feeds in and what comes next.
BD
Discovery
Feedback
Product
Engineering
Release
Reliability
Evidence
Evidence
Prove what happened.
Available now
Audit logs
Review every workspace change and who made it.
How we get to a number
We price the tokens your agents reported against a dated public price catalog, split four ways: input, output, cache read, cache write. Read-time and backfill use identical logic, so the number is stable. It is a fast, dated estimate you can check, not your provider's invoice.
Where the money actually lands.
Read spend by team and model at a glance. Each cell fills by its share, so a heavy team or an expensive model stands out.
teamopussonnethaiku
platform$820$310$90
web$540$260$120
mobile$210$180$60
data$130$90$40
estimated spend · darker fill means more
Not a monthly number — the play-by-play.
Open any session and see the whole run: an activity strip across its span, one swimlane per tool, and every failed call flagged in place.
activity1,284 events
09:1409:52
- read42
- edit18
- bash23
- grep15
- write9
flagged call · a tool call that failed or timed out
How the work actually happened.
Per-tool event counts, cache hit rate, reasoning-effort mix, and time to first token — the fine-grained signals a per-seat spend report can never produce.
events by tool
- Read1,240
- Bash820
- Edit610
- Grep430
- Write210
71%
cache hit rate
p95 3.4s
time to first token
38%
high reasoning effort
Break it down, then drill in
Group spend and usage by user, model, client, repo, tool, slash-command skill, or session — then open a leaderboard and drill into a single user, repo, or session. Save the cuts you check often as segments, scoped to you or the whole org.
Ingest, price, break down.
From event to rollup.
Send OTLP or a custom JSON envelope. Each event carries tokens, duration, and git context, keyed so retries never double-count.
OTLP or JSONidempotency keyhourly and daily
- 01
Emit
OTLP or JSON envelope
- 02
Dedupe
by idempotency key
- 03
Cost
tokens x dated price
- 04
Rollup
hourly and daily
Estimated spend, summed as you watch.
Spend is grouped by the provider behind each model, computed from public token prices applied to reported tokens. The total counts up from the parts.
anthropicopenaigoogle
estimated · this month
$3,020
- Anthropic$2,000
- OpenAI$720
- Google$300
The same four-way split every time.
Cost comes from a dated price catalog, split four ways: input, output, cache read, and cache write. Read-time and backfill use identical logic.
dated cataloganomaly detectionoverride
- Output$0.62
- Input$0.28
- Cache read$0.14
- Cache write$0.09
Not just what it cost — how it went.
Every event carries its duration, time-to-first-token percentiles, and an exit status, so a slow or failing agent shows up beside its bill.
p50 / p95 / p99event durationexit status
- ok96.2%
- error2.1%
- timeout1.1%
- denied0.6%
One command, then it flows.
The published CLI wires Claude Code, Codex, Cursor and OpenCode to your workspace, backfills history from local transcripts, and tails events live.
@lubed/clibackfill historylive tail
- 01
Install
npm i -g @lubed/cli
- 02
Connect
lube ai connect
- 03
Emit
agents send telemetry
- 04
See it
spend in the dashboard
Privacy is a control
Four ranked levels decide what a payload keeps, from metadata_only structural telemetry up to opt-in full_content. Gitleaks and redaction tripwires reject any batch that leaks secrets, and every privacy toggle is written to the audit log.
See what else lands here.
See what your agents cost this month.
Start free
Four privacy levels. You pick what gets stored.