optimAIzr

Spend fewer tokens on the same work.

optimAIzr is a local CLI that finds wasted tokens in Claude Code, Codex and your own API calls, shows what each fix would save, and checks it on your own traffic before you switch. No account, no upload, no telemetry. It reads token counts, never your prompts.

First presented on September 13, 2026, and open source since September 30, 2026.

npm i -g optimaizr
optimaizr profile

Or try it once without installing: npx optimaizr profile.

New in 0.8.0: inside Claude Code

Claude Code now runs mods, plugins that work inside it. The optimAIzr mod puts the numbers where you work, lets Y in optimaizr live switch the session you are in, and shows what each switch saved.

  ⏺ optimaizr: this turn $0.18 · 4 requests · 5h 55% → 56%

  optimAIzr  █████████░░░░░░░  56% of 5h · resets 15:00
  ❯

Install it from a Claude Code session (2.1.287 or later):

/plugin marketplace add blendbunjaku/optimaizr
/plugin install optimaizr@optimaizr

• The cost while Claude works. The spinner shows what the turn has cost so far and how full your 5-hour window is. The band above the prompt shows the window, how long it lasts at this pace and when it resets, from Claude Code's own limit meter. Each answer gets one line with its cost, its requests and how far the window moved.
• Switches in the session you are in. Press Y on a model swap in optimaizr live and the running session uses the cheaper model from its next request. No restart, no /model. Subagents switch at once; a long conversation waits until reloading it into the new model pays back. A reasoning-effort finding lowers effort the same way. With optimaizr live --auto, confident switches apply without asking.
• What a switch saved. A switched turn's line leads with the saving: the same tokens priced on the original model, less what they cost. The band keeps a running total.
• Back in one command. Harder task? /optimaizr off returns the session you are in to its own model and effort; /optimaizr on resumes.
• A guard against retry loops. When the same command fails twice in a row with nothing changed, the next identical attempt is held once and Claude is asked to change something first.
• /optimaizr prints the session's spend, what it saved, both plan windows and any switch. /optimaizr hud shows it all as gauges and a sparkline.

It reads usage figures and the commands Claude runs, never your prompts or file contents, and makes no network calls. The source is in mods/optimaizr.

What it does

See where your tokens go

optimaizr profile reads the sessions already on your machine and puts spend, waste and the biggest fix on one screen.

  Spend                     $476.70  $452.56/month at the last 30 days' rate
  Calls                       4,062
  Tokens                     743.4M  739.7M in / 3.7M out

  Flagged calls                 931  22.9% of calls
  Potential savings       $44.59/mo  $542.48/year

  ! Model mismatch

    17% of your requests use a model whose capabilities exceed the
    detected workload requirements.
    671 calls affected.

Find out why

optimaizr why breaks spend down by provider, model, project and kind of work, each level as a share of the one above.

  $132.47 total  across 1,807 requests

    anthropic                         $132.47  100% 1807 calls
      -> sonnet-5                         $99.47   75% 1640 calls
        -> kopshti-back                   $93.34   94% 1541 calls
          -> Mechanical                   $36.11   39% 746 calls
          -> Reasoning                    $27.52   29% 335 calls
          -> Generation                   $15.22   16% 141 calls

Catch waste while you work

optimaizr live runs beside your agent and raises a fix the moment a pattern shows up. Press Y and your app switches from its next request, and so does Claude Code with the optimAIzr mod. Without it, Claude Code switches from its next session.

  ⚡ optimAIzr

  This task looks suitable for a cheaper model.

  Current:       Sonnet 5
  Suggested:     Haiku 4.5
  Observed cost: $0.202 / 26 calls

  [Y] Apply optimization
  [N] Continue
  [D] Why?

Know what a fix is worth

Every finding comes with its own arithmetic: what the traffic costs now, what it would cost after, and how far to trust the figure. --why prints the full calculation and every assumption.

MEDIUM  441 mechanical calls ran on an over-specified model
        $5.76/mo est.  |  $70.12/yr  |  needs verification  |  model-selection

        now    $14.07  ->  after     $6.75  (441 calls, 12% of spend)
        confidence medium   quality impact medium   basis estimated

Prove it before you switch

optimaizr verify replays your own recorded requests on the cheaper option and scores the answers against your quality bar, so you never trade cost for worse output.

   PASS   40 samples replayed

  cost/call   $0.0121 -> $0.0034
  projected   $61.40/mo saved
  this check cost you $0.38

  Quality checks
    ok   no-refusal               candidate 40/40, baseline 40/40
    ok   tool-name-matches        candidate 39/40, baseline 39/40

Stay under your limits

On Claude Pro, Max or Team, --plan pro shows your 5-hour sessions and how much of each goes on waste. On ChatGPT plans it reads OpenAI's own meter from Codex. On a company budget, --budget 300 names the day the money runs out.

  ChatGPT Plus            $20.00/mo  what you pay
  API-equivalent         $117.72/mo  5.9x what you pay
  5-hour window                 82%  used · resets 01:49 · a full window ~$4.76
  Weekly window                 38%  used · resets Sep 25, 22:49
  Waste                          9%  fix it and you'd hit the limits ~10% later

Works with

13 detectors cover oversized models, cache misses, bloated prompts, repeated file reads, retry loops, runaway reasoning, cost spikes and more.

Commands

Run optimaizr --help for everything.

Privacy

Everything runs on your machine and stays in ~/.optimaizr/. The only network calls are verify (to your own provider, with your own key) and the optional Jev second opinion. SECURITY.md lists every file read and written.

Documentation

• Package README: the full usage guide
• CLI reference: every command, flag and figure explained
• Changelog
• optimaizr.com/docs

Development

npm install && npm run build && npm test

Requires Node 20.11+. See CONTRIBUTING.

optimAIzr Pro, the hosted product coming to optimaizr.com, is separate from this repository.

License

MIT