Tokens Boutique

Plan, audit, and route your AI spend.

local-first cost engineering

Cost planner

From usage receipt to deployable routing policy.

Inspect real token usage, find price-compatible migrations, build a budget-aware model mix, export it, and see how conversation history compounds cost.

Shared inputs

Workload baseline

Changes flow through migration, routing, and policy export.

01 · Audit

Usage Receipt Inspector

Paste or upload OpenAI-style, Anthropic-style, JSONL, or CSV usage. Parsing and pricing happen entirely in your browser.

Recognized model names are matched automatically. Unknown names use the fallback above.

02 · Compare

Model Migration Planner

Find lower-cost candidates that can hold the same input and requested output. Capability equivalence is intentionally not assumed.

Current monthly estimate$195.00
CandidateContext / max outMonthlySavingsAction
qwen3.7-flashAlibaba1M / 64K$0.50+$194.50
solar-pro4Upstage · frontier512K / 128K$0.54+$194.46
nova-microAmazon128K / 5K$0.63+$194.37
nova-liteAmazon300K / 5K$1.08+$193.92
ministral-3b-latestMistral256K / 256K*$1.35+$193.65
muse-spark-1.2-contributorMeta1M / 128K$1.50+$193.50
ernie-4.5-turboBaidu128K* / 8K*$2.00+$193.01
ministral-8b-latestMistral256K / 256K*$2.02+$192.97

Compatibility checks cover published context and max-output limits only. Evaluate quality, latency, features, and compliance before migrating.

03 · Allocate

Budget Router Builder

Choose a baseline and premium model. The builder computes the maximum premium traffic your monthly budget can support.

Recommended traffic mix

100.0% premium

$195.00projected / month

qwen3.7-flash requests

0

claude-fable-5 requests

1K

04 · Ship

Pricing Policy Exporter

Export the workload and current router as portable JSON, YAML, or a small TypeScript routing helper.

{
  "version": 1,
  "budget": {
    "monthly_usd": 1000
  },
  "workload": {
    "input_tokens": 12000,
    "output_tokens": 1500,
    "cache_hit_percent": 0
  },
  "routing": {
    "strategy": "budget-weighted",
    "primary": "Alibaba/qwen3.7-flash",
    "premium": "Anthropic/claude-fable-5",
    "premium_traffic_percent": 100,
    "fallback": "Alibaba/qwen3.7-flash"
  }
}

This is vendor-neutral policy data. Map model IDs to the SDK or gateway names used in your own stack.

05 · Simulate

Context Growth Visualizer

See how replaying conversation history increases billed input on every turn, including the cost effect for a selected model.

Session cost

$1.58

Billed input

127.8K

Final / first turn

3.6×

Turn 1: 4,600 input tokens, $0.07
Turn 2: 5,700 input tokens, $0.08
Turn 3: 6,800 input tokens, $0.09
Turn 4: 7,900 input tokens, $0.10
Turn 5: 9,000 input tokens, $0.11
Turn 6: 10,100 input tokens, $0.13
Turn 7: 11,200 input tokens, $0.14
Turn 8: 12,300 input tokens, $0.15
Turn 9: 13,400 input tokens, $0.16
Turn 10: 14,500 input tokens, $0.17
Turn 11: 15,600 input tokens, $0.18
Turn 12: 16,700 input tokens, $0.19
Turn 1Turn 12