TokenJam Lens

The local dashboard for every analyzer's findings.

TokenJam Lens is the dashboard you get when you run tj serve. It runs entirely on your machine (no signup, no cloud) and brings the output of every analyzer, every alert, and every cost trend into one screen you can open in your browser.

Get started GitHub

Overview, your triage front door

Three bands answer "where am I, and what should I look at next?" in one glance. Cost hero with a sparkline and an honest run-rate caption. Recoverable waste tiles, one per analyzer, showing dollar values or "✓ at 100%" when there's nothing to recover. A one-line health summary covering alerts, drift, and budgets. Every tile drills into the detail screen with filters pre-applied via URL.

127.0.0.1:7391/#/overview

Overview Last 7d

Spend · last 7d $284.71

At this pace, ~$1,219 by end of June

(linear run-rate, not a forecast)

Recoverable waste Estimated

Downsize

$42 / mo

Cache

✓ at 100%

Script

no findings

Trim

not run

Reuse

$54 / mo

Cache-rec.

enable cap.

3 unread alerts · 0 agents drifting · 1 budget near 80%

Every analyzer's findings in one place

Lens reads from the same tj optimize analyzers the CLI does, plus a dedicated Optimize tab where you see each analyzer's findings side by side, filter by agent, and compare against a previous window. New analyzers (like Reuse) appear automatically: the band is registry-driven.

127.0.0.1:7391/#/optimize

Window: Last 30d Agent: all Compare: previous

① Downsize $42 / mo

8 of 47 sessions match a smaller-model candidate shape. Estimated recoverable: $42 / mo. Review before applying.

② Cache ✓ at 100%

anthropic/claude-opus-4-7 · efficacy 100% (2.79 B tokens). Already optimized, nothing to recover.

⑤ Reuse $54 / mo

1 cluster found: 31 sessions share the "patch-release" skeleton (read changelog → bump version → run tests → tag → push). Estimated recoverable: $54 / mo (cache-reuse) or $61 / mo (script replacement).

Cost transparency, with the math

The Cost screen shows your spend as a real chart with hover tooltips, and the cost table now includes CACHE R and CACHE W columns so a cache-heavy $1.44 span is no longer a mystery. Run-rate projections are explicitly captioned "linear run-rate, not a forecast": Lens never claims to predict.

127.0.0.1:7391/#/cost

Window: 30d Group by: model

Spend over time $1,247.83

MODEL	TOKENS IN	CACHE R	CACHE W	COST
`claude-opus-4-8`	294 K	698.8 M	0	$468.53
`claude-opus-4-7`	38.8 K	1.53 B	1.48 M	$1,115.44
TOTAL	369 K	2.79 B	2.95 M	$1,247.83

Get started

Three commands. Local. No signup. Open the dashboard at 127.0.0.1:7391 in your browser.

install

$ pipx install tokenjam
$ tj serve
  → open http://127.0.0.1:7391/

Source on GitHub. Lens shipped in v0.4.0. Reference docs in the docs site.