Token Optimizer Kit v2.1 — TokOp App + Claude Code Routing Policy

Stop paying flagship prices for cheap work. Most people run every task through their coding agent the same way: full model, full context. So a routine log scan costs as much as a decision that actually needed your best model. This kit fixes that two ways: a routing policy you drop into Claude Code, and (new in v2) TokOp, an app that runs the policy for you.

THE PROBLEM (measured, not assumed): one week of my own Claude Code usage was 241M tokens, about 100% on flagship models and 0% on the cheap tier, even though much of it was routine work a cheap model would have finished for a fraction of the cost.

NEW IN V2, THE TOKOP APP: a point-and-click window, no commands to learn. Type a request and click Ask, Do it, or Save for off-peak. TokOp checks what you mean, asks a quick question if it's unclear, rewrites it for a one-shot answer, and sends it to the cheapest Claude model that can do the job (Haiku, then Sonnet, then Opus only when needed), with the answer right in the same window. NEW IN V2.1, PEAK HOURS: AI providers ration capacity in weekday business hours. During peak, TokOp spots work that isn't urgent (a report, a batch of drafts, a cleanup) and offers a one-click Save for off-peak, so it runs later at a cheaper time while you're away. Questions you're waiting on are never delayed. Also included: a real token/$ usage meter from your Claude Code logs, an always-on safety gate (it never auto-runs sends, deploys, spending or deletes), and an optional background runner that uses leftover plan quota while you're away, preferring off-peak hours. Needs Python 3.10+ and the Claude Code CLI; it runs on your existing Claude login with no API key. A 5-minute quick-start tour and a step-by-step setup guide get you running in about 10 minutes.

THE POLICY: a triage rule for every task (trivial, mechanical, research or high-stakes); bulk work routed to a cheaper model; research forked out of your main context so raw output doesn't bloat it; money, legal, security and irreversible actions never downgraded; and your instructions file split into load-every-time vs load-on-demand to cut fixed overhead.

WHAT'S INSIDE (125 files): the TokOp app with its quick-start and setup guides; a copy-paste CLAUDE.md template; a guide to wiring it in and measuring results over two weeks; an honest breakdown of when it saves real money (most on API-billed and team usage, less on a flat subscription seat); a worked example; an FAQ; drop-in Claude Code agent files; and verified drop-in configs for 31 other AI coding tools.

OPTIONAL, OFF BY DEFAULT: TokOp can also send prompts to your own logged-in ChatGPT, Perplexity or Venice. Automating those websites goes against their terms of service and can get an account flagged, so the setup guide explains that risk plainly before showing how to turn it on. Everything else works without it.

Instant download. 7-day no-questions-asked refund if it's not a fit.