LOCAL AI SPEND INTELLIGENCE

Your 2026 AI coding budget.

Agentic tools burn far more per task, and the bill is now a defended line. Get the view across Claude Code, Codex and Pi before the next review.

Learn more
Itemised to branch, fork & PR Know before you build One view, every agent
Works with
Claude CodeCodexPi
Trendl is not affiliated with or endorsed by Anthropic or OpenAI.
History and foresight

A record you own, and a cost you can see coming.

Trendl scrapes usage in real time and keeps it, so you can look back over weeks, and know what to expect before you build.

Kept, not snapshotted

A per-event history stored on your machine, back as far as you have run it.

Weeks at a glance

Cumulative spend across every session and model over time.

Know before you build

A likely cost for the next piece of work, from the history you already have.

Spot what runs hot

See which kinds of work cost the most, so you can plan around them.

05B10B15B20B08 Jul19 Jul30 Jul10 Aug21 Aug
Cumulative usage kept across weeks. Illustrative.
Across every tool

Every agent, one view.

Claude Code, Codex and Pi read into a single dashboard, so the whole team's spend sits in one place.

One dashboard, every provider

Claude Code, Codex and Pi totalled together, per model.

Per session, per model

Every session priced, tokens and estimated cost on each row.

No setup overhead

Reads the sessions your tools already write. Nothing to wire up.

Estimated, not guessed

Costs from your logs and current price tables. Your invoice is the final word.

Claude · opusClaude · sonnetCodex · codexCodex · gpt-5-codexPi · qwenPi · kimi00%100M25%200M50%300M75%400M100%17 Aug18 Aug19 Aug20 Aug21 AugTurnsIdleEvents
Aggregated usage across Claude Code, Codex and Pi. Illustrative.
Attribution

Down to the branch and pull request.

The bill stops being one number. Every token is tied to the exact work that spent it.

Project, branch, fork, PR

Spend traced to the branch and pull request it came from.

Estimated cost per path

A dollar figure on every branch, not just a monthly total.

Which work it paid for

See the tickets and PRs the spend actually produced.

Ready for finance

A breakdown you can hand over and stand behind.

ProviderModelPathTokensEst. cost
Piqwen-3-8bcheckout-api-pr-174-discovery42.4M$30.27
Claudeclaude-sonnet-5payments-gateway-pr-333-discovery93.2M$82.50
Claudeclaude-sonnet-5notify-svc-pr-613-discovery100.0M$76.01
Claudeclaude-opus-4-8web-dashboard-pr-260-discovery98.6M$72.21
Claudeclaude-sonnet-5billing-worker-pr-115-discovery32.6M$28.08
Codexcodex-auto-reviewdata-pipeline-pr-205-discovery52.5M$34.37
Piqwen-3-8bauth-service-pr-190-discovery128.3M$101.08
Spend attributed per project and pull request, with estimated cost. Illustrative, sample data.
Rates and limits

See the burn as it happens.

Token rate, cost per hour, concurrency and prompt rate, as moving averages.

Live rates

Tokens per second and dollars per hour, smoothed.

Concurrency

How many turns run at once, and when it peaks.

Limits in view

Session and weekly windows, with time left on each.

Peaks named

The busiest hours, so the spend is never a mystery.

Token usage rate · tokens/s · peak 84.7k100k0Estimated cost rate · $/hr · peak $215$250$0Concurrent turns · turns · peak 330Prompt rate · prompts/hr · peak 176200021 Aug 08:0721 Aug 09:3021 Aug 10:50
Activity rates as moving averages. Illustrative.
The numbers

Estimates you can stand behind.

Trendl works cost out from what your tools actually did, so the figure you show finance holds up.

Priced from real usage

Costs are worked out from the sessions your tools already recorded, not sampled or guessed.

Current price tables

Every model priced at its current rate, updated as prices change.

Your invoice wins

Figures are estimates. Your provider invoice stays the source of truth.

Read-only

Trendl reads your sessions. It does not change how your tools run.

FAQ

Questions, answered.

Where do my Claude Code tokens actually go?
Trendl breaks spend down by project, branch and pull request across Claude Code, Codex and Pi, so you can see which work consumed the tokens.
How is the cost worked out?
Costs are estimates from your local logs and current price tables. Your provider invoice is the source of truth.
Which tools does it work with?
Claude Code, Codex and Pi. Trendl is not affiliated with or endorsed by Anthropic or OpenAI.
Can it tell me what work will cost?
Yes. From the history it keeps, Trendl shows the likely cost of the next piece of work, so you can decide before you build. These are estimates, not guarantees.
Is it for individuals or teams?
It is built for teams. Trendl rolls up spend across the whole team, not just one machine.

Your AI spend, itemised to the pull request.