How hard are you TokenMaxxing?

Find out your tier in 3 commands. No signup. Runs locally.

find out your tier

$ pipx install tokenjam$ tj onboard --claude-code$ tj tokenmaxx

Reads your existing ~/.claude/projects/ logs · 30 days of history · MIT-licensed

Here's what you'll see: a real run for a Max-5x user.

╭─ TokenJam TokenMaxxing Report ─────────────────────────────────────────────╮
│                                                                              │
│  ⚖️ You're a SteadyState.                                                    │
│                                                                              │
│  Normal drift. A /compact habit keeps it lean.                              │
│                                                                              │
│  64% of your quota went to overhead (re-reading history, CLAUDE.md,        │
│  tool output). 36% did net-new work.                                        │
│  Measured over 33 sessions in the last 30d.                                  │
│                                                                              │
│  💡 Reclaim overhead with tj context, or run tj optimize for the full      │
│  savings report.                                                             │
│                                                                              │
╰──────────────────────────────────────────────────────────────────────────────╯
  Share your tier: screenshot the above and tag @tokenjamdev

The 5 tiers

Classification is based on how much of your quota goes to overhead: the share of your last-30-days tokens spent re-reading context (conversation history, CLAUDE.md, tool output) instead of doing net-new work. Lower overhead is leaner, so a lower share is a better tier.

TokenMinimizer ≤ 30% overhead

Surgical context. Teach the rest of us.
LeanOperator ≤ 50% overhead

Tight loop. Most of your quota does real work.
SteadyState ≤ 70% overhead

Normal drift. A `/compact` habit keeps it lean.
ContextHeavy ≤ 85% overhead

Overhead is winning. Run `tj context`.
QuotaSink > 85% overhead

Re-reading is eating your quota. Run `tj context`.

The overhead share is a measured fraction of your own tokens, not a guaranteed saving. Subscription plans see the headline as a token share; API plans get an "implied API value" line below it. The tier is a starting point. Run tj context to see where the overhead comes from.

What does it actually do with that?

The tier is the shock value. The point of the tool is what comes next: tj optimize runs five analyzers against your usage history and surfaces specific savings candidates: sessions where a cheaper model would have done the same job, prompt sections you could trim, prefixes you could cache, deterministic sequences you could replace with a script, and repeated planning you could serve from a template instead of regenerating.

See how Downsize works → Or the full observability stack →

Post your tier on X.

What does it actually do with that?