Ship lean, stay green Preview · invite only

Cut your AI coding bill, not your output.

Your AI coding spend lands as one opaque number a month. Frugl reads every session, flags the tokens that bought you nothing, and shows each engineer how to get the same work done for less.

Start free See a live report
Free first report No card Redacted on-device
Reads across Claude Code Codex Gemini Cursor
Costs $0
Auto-playing
↑←↓→ or swipe — dodge the bills, ship lean
Tokens saved · 7d
1.2M
≈ $1,410 back in the budget
Flagged waste
$2,180
22% of spend · top: bloat
Top fix this week
−$720/wk
Trim bloated context · same output

The problem

You can’t cut what you can’t see.

Most of the burn is noise: a flaky test or a stack trace gets pasted back in, the assistant retries blind, and the same error loops for thousands of tokens before anything lands. Add bloated prompts and wandering tangents across four assistants, and it all arrives as one opaque number at month’s end — with no way to say which sessions paid off and which just spun on a noisy error.

Frugl reads the whole firehose and does two things: it flags the waste it finds — naming the smallest fix that claws the budget back — and it gives each engineer concrete feedback on how to waste less next time.

How it works

From every session to the receipt.

Raw AI coding sessions in · one ranked receipt out · nothing wasted in between.

  1. Your team’s sessions

    Every prompt, retry and tangent — across four AI assistants.

    8 engineers · 598 sessions
    Claude CodeCodexGeminiCursor

  2. One command to upload

    Redacts on-device, then uploads the anonymized sessions.

    $frugl upload
    ✓ 598 sessions · Claude Code, Codex, Gemini, Cursor
    ✓ redacted on-device · nothing sensitive left the machine
    → uploading 598 anonymized sessions…

  3. Frugl finds the waste

    Frugl ranks the spend, flags the waste, and writes each fix.

    Rank by spend Flag the waste Match to PRs Write each fix

  4. Your receipt & reports

    A shared team dashboard, and a private report for every engineer.

    The team site Where every dollar went — waste ranked, fixes named. See it ↓ Per-dev reports A private, quantified review for each engineer. See it ↓

01 · Find the waste

It finds the waste. Then it names the fix.

No vanity charts. Frugl ranks where the tokens went, flags what was wasted, and hands you the single change worth making this week — quantified, so cutting waste never slows the team down.

frugl.dev/acme-inc

Where Acme Inc's AI spend went

Last 30 days · 6 models · ranked by what's worth changing this week
7d14d30d90d
Total spend
$9,840
+18% vs. prev 30d
Flagged waste
$2,180
22% of total · top: bloat
Sessions
598
8 members · 4.1M tokens
Where the waste livesBy category · 30d
01
Root-context bloat$1,840
02
Skill & tool sprawl$1,320
03
Session looping$1,180
04
Wrong-sized models$860
Worth changing this weekRanked
01
Trim CLAUDE.md from 14.2k → 8k tokens
−$720/wk
02
Cut 9 skills nobody calls
−$410/wk
03
Cap retries on tool failures at 3
−$280/wk

Illustrative figures. Frugl is pre-launch — the dashboard reads live once your sessions are connected.

Stop guessing at your AI bill. Get your team’s first waste report — the same ranked receipt you just saw, with your numbers. Free, no card, redacted on-device.
Start free

02 · Level up every engineer

Your team gets better at AI, week over week.

Finding waste once is easy; using AI better is the part that sticks. Each engineer gets a private, plainspoken review of what cost the most in their own sessions — with the one change that fixes it. No blame, no public leaderboard. Just the receipt, and how to read it next time.

  • Private per-dev feedback, never a public ranking
  • Specific and quantified — “do this, save that”
  • Watch each fix’s savings add up, week over week
Jordan’s sessions this week 5 reviewed · 312k tokens
Reclaimable $540/wk
Context

4 MCP tools unused but loaded in root context — drop them from the config so they stop loading on every call.

−$180/wk
Loops

Six blind retries after the same failing test. Pin the test command in CLAUDE.md so it stops guessing.

−$140/wk
Model fit

Opus did featherweight edits in 18 sessions. Route quick edits to a smaller model.

−$220/wk
Across 8 weeks of feedback — team waste down 31%, same work shipped.

What Frugl counts

Everything counted, nothing wasted.

Frugl watches the parts of AI-assisted engineering that never make it onto an invoice — the token waste that hides between the lines of every session.

  • Root-context bloat

    CLAUDE.md and config files that have outgrown their usefulness — re-sent on every single call.

  • Loops & blind retries

    Runs that spun in circles, and the retries that fired again after the same error never lands.

  • Tool & MCP failures

    Which tools and MCP servers get called, how often they fail, and what those failures cost.

  • Skills nobody calls

    Installed skills and tools that sit in context burning tokens but never actually get used.

  • Wrong-sized models

    Where a heavyweight model did featherweight work, so you can right-size without losing quality.

  • Spend that shipped nothing

    Sessions tied to merged PRs vs. the spend that produced no code at all — told apart at a glance.

Privacy-first

Built to be trusted with the work that matters.

Anonymized on your machine Redaction runs locally before a single byte leaves. We never see your code or secrets.
Fail-closed by default If anonymization can't complete, the upload doesn't happen. No silent leaks, ever.
You own your data Only the anonymized data is stored — kept as an immutable record to re-derive your numbers, and deletable any time.
Preview · invite only

Get your team’s first report.

Frugl is in private preview. Request access and we’ll send an invite to app.frugl.dev — usually within two business days. Already in? Log in and pick up where you left off.

Request access Log in