>_ Token observability for AI coding tools

See where every token goes — and get it back.

Tokup tracks, diagnoses, and reduces the token spend of Claude Code, Codex, Cursor, and every AI coding tool your team runs — down to the task, the model, and the dollar.

Start free See a team's usage

No credit cardRead-only by defaultSet up in minutes

SAVED THIS MONTH

$4,280

↓ 18% vs last month

SPEND BY TOOL

Claude Code

$2,190

Codex

$1,340

Cursor

$810

!Payments team re-sent 40K tokens of context 3× this week on refactor tasks — switch to a diff-based prompt to cut this by ~60%.

What engineers on teams are already struggling with

Real patterns we keep hearing from developers and engineering leads wrestling with AI coding tool spend — paraphrased and anonymized.

“In one month we paid $10,550 to a single AI routing bill — and I couldn't tell you which part of that was actually necessary.”

Tom, founder at an early-stage startup

“We shipped fewer PRs that month but the bill jumped almost 20x. Turned out 84% of it was cache reads and writes — not the tokens I'd been staring at.”

Sacha, engineer at an early-stage startup

“I ended up building my own pipeline just to find out what my own team was spending on AI tools. Nothing made it easy to see.”

Lea, engineering lead

“Is there a tool that just shows me which sessions are eating my usage? I genuinely can't find one.”

Noah, indie developer

“I'm furious there's no single view of spend across the three AI tools we use — and that basic visibility shouldn't require an enterprise plan.”

Marine, CTO at a startup

“Found out 61% of my weekly usage came from one session I'd simply forgotten to close.”

Hugo, developer

“In one month we paid $10,550 to a single AI routing bill — and I couldn't tell you which part of that was actually necessary.”

Tom, founder at an early-stage startup

“We shipped fewer PRs that month but the bill jumped almost 20x. Turned out 84% of it was cache reads and writes — not the tokens I'd been staring at.”

Sacha, engineer at an early-stage startup

“I ended up building my own pipeline just to find out what my own team was spending on AI tools. Nothing made it easy to see.”

Lea, engineering lead

“Is there a tool that just shows me which sessions are eating my usage? I genuinely can't find one.”

Noah, indie developer

“I'm furious there's no single view of spend across the three AI tools we use — and that basic visibility shouldn't require an enterprise plan.”

Marine, CTO at a startup

“Found out 61% of my weekly usage came from one session I'd simply forgotten to close.”

Hugo, developer

Works with the tools your team already runs

Claude CodeCodex CLICursorWindsurfGemini CLI+ MCP servers

Claude Code today · Codex, Cursor and more next

The product

Built on four things, not one.

Most tools stop at a usage graph. Tokup watches, explains, rolls up, and acts.

Observe

Attribute every token to the task, model, tool, and developer that spent it — not just a daily total.

Diagnose

Specific fixes, not vague advice: "this session re-sent 40K tokens of context 3×; switch to a diff."

Aggregate

Roll individual usage up into team and fleet-wide views a lead engineer or CFO can actually act on.

Intervene

Set budgets, get a forecast if nothing changes, and let Tokup react before a task or team goes over.

Fleet governance

One view across every tool, endpoint, and skill your team runs.

See adoption, usage, and which harnesses, models, and MCP servers need attention — then decide which ones your team can actually use.

Workstations

Developer machines — via built-in OpenTelemetry

Hosted agents

Codex Cloud, Claude Code Cloud — via vendor APIs

>_See which harnesses, models, and skills are adopted — and by whom.
>_Define which tools your team is allowed to run, in the context of the work.
>_Attribute spend per task and per type of work, not just per seat.

TOOLADOPTIONMONTHLY SPENDTREND

Claude Code82%$2,190↓ 12%

Codex CLI41%$1,340↑ 6%

Cursor29%$810↓ 3%

MCP: internal-docs17%$210needs review

Why Tokup

“Couldn't I just...”

The three things people reach for before they reach for Tokup — and where each one runs out.

Use my AI tool's own usage dashboard?

Claude Code, Cursor and Codex each show you their own usage — but your team almost certainly runs more than one. None of them show you a single, combined view across tools, so you're stitching together three tabs by hand every time you want the real number.

Track it through my spend-management tool?

General spend-management tools are built for invoices, not sessions. They typically can't attribute usage routed through something like OpenRouter, can't break spend down per developer, and often require an enterprise plan with each AI vendor just to connect — three real gaps engineering teams run into today.

Just script it myself?

Plenty of engineers do — a scheduled job pulling logs into a spreadsheet or a warehouse. It works until a vendor changes an API or reprices a model, and then someone has to notice and fix it. Tokup is built to keep itself current against exactly that, so nobody has to own a brittle internal script.

PAYMENTS TEAM — MONTHLY BUDGET

$3,000 cap

On track to hit cap in 9 days

$2,040 spentDay 21 of 30

FORECAST IF NOTHING CHANGES

Projected month-end: $3,320 — 11% over cap

Budgets & forecasting

Set a budget. See it coming, not just blowing past it.

A lead engineer sets a token or dollar budget per team, project, or task type. Tokup forecasts the month-end outcome under current behavior — and a second forecast for what happens if your recommendations get followed.

>_Per-task cost ceilings, not just a blanket monthly number.
>_Alerts at threshold — before the breach, not after the invoice.
>_Two forecasts side by side: "as-is" and "if the fixes are applied."

Ask Tokup

A chat that actually knows your usage.

Ask about your own spend or your whole team's, in plain language — Tokup answers from the same data behind the dashboard, not a generic model guessing.

Why did my Tuesday session cost $12 more than usual?

That session hit the model's context limit twice during a large refactor, triggering two full re-sends of ~35K tokens each. Splitting the task into smaller diffs would have kept it under your usual $8 average.

What's the team's biggest cost driver this week?

Code-review tasks on the payments repo — 34% of spend, mostly from re-reading the same large files. Caching those files would save roughly $310/week.

Estimate

What this could mean for your team

A directional estimate, not a guarantee — built from published benchmarks on AI coding tool spend. Your real numbers will differ.

Developers using AI coding tools20

Est. monthly AI-coding spend

$715

Est. monthly savings with Tokup

$179

Est. savings over 12 months

$2,145

Projected spend, as-is vs. with Tokup

Current path

With Tokup

Based on ~65% of developers actively using AI coding tools at ~$55/seat/month blended across Claude Code, Cursor and similar tools, and a 20–35% reduction from Tokup's diagnostics — conservative versus the 40–80% reductions reported in some vendor case studies. Connect your own tools after signup for your real numbers.

Pricing