TokenGauge v0.3.0 — Cost attribution workflow

TokenGauge v0.3.0 turns provider-level token totals into a practical application-attribution workflow.

Start free

• Audit an AI API bill locally in the browser.
• Attribute cost by project and workflow with the canonical CSV ledger.
• Read the LLM cost-per-customer and feature guide.

The guide separates the provider ledger from application context, recommends opaque customer labels instead of personal data, accounts for cache buckets and retries, compares cost per accepted answer, and keeps modeled token cost distinct from the provider invoice.

Fixed-scope implementation

The £75 attribution setup covers one authorized Node.js/TypeScript or Python codebase and one existing provider path: project/workflow labels, retry-aware usage fields, a TokenGauge-compatible CSV export, focused tests, handoff notes, and one revision after written scope.

No credentials or private source belong in the initial enquiry. No savings are guaranteed, and API credits are not included.

Existing workbench

• 52 dated pricing cards across nine provider families
• 120 evidence cards, including 12 open cards
• 1,316 atomic research candidates and 1,184 compound configurations—2,500 research rows, not 2,500 distinct proven methods
• verified-email and ChatGPT-linked accounts, TOTP 2FA, encrypted provider connections, Stripe Checkout, dashboard, and controlled A/B lab

Validation

The release passed atlas freshness verification, ESLint, strict TypeScript, 44 Vitest tests, a 31-route Next.js production build, responsive 390px and 1440px browser checks, live canonical/sitemap/schema checks, and credential-pattern scanning. The live mirror and maintained application are independent of any listed model provider.

TokenGauge v0.2.0 — Research Atlas

Research Atlas

• 2,500 mechanically compiled research rows: 1,316 atomic candidates and 1,184 compound configurations
• 12 representative rows remain open; every paid plan gets server-filtered search over the complete atlas
• candidates, configurations, curated evidence cards, and supported lab recipes remain explicitly distinct
• deterministic compiler and CI freshness check prevent generated-data drift
• provider adapters remain included in Pro; Pro+ and Ultimate expand experiment history/export depth

Verification

• 38 tests, ESLint, strict TypeScript, and production build pass
• anonymous API returns 12 rows and withholds the other 2,488
• live 390px browser check has no horizontal overflow, 16px inputs, 44px controls, and zero console messages

Live atlas: https://tokengauge.enby.fish/atlas

No savings are guaranteed. Research rows are test candidates and configurations, not 2,500 distinct methods or implemented adapters.

TokenGauge v0.1.0 — Launch 100

TokenGauge is an evidence-backed LLM cost-intelligence workbench for comparing official API rates, modelling realistic token-cost changes, and running controlled A/B experiments without storing prompts or outputs.

Included

• 52 dated pricing cards across OpenAI, Anthropic, Gemini, xAI, DeepSeek, Kimi, Qwen, Mistral, and Cohere
• 120 published evidence cards, including 12 open cards with guided measurement plans
• free scenario calculator and public rate directory
• verified-email accounts with password recovery, TOTP 2FA, recovery codes, lockout, and session controls
• optional primary sign-in through the open-source ChatGPT device-login integration
• encrypted BYOK provider connections and randomized paired experiments
• one-time Pro, Pro+, and Ultimate purchases through Stripe Checkout

Launch 100

The first 100 authenticated identities can purchase Pro for £5, Pro+ for £15, or Ultimate for £20. Normal one-time prices are £9, £19, and £39. API usage is billed separately by the connected provider.

Open TokenGauge · Read the methodology

TokenGauge is independent software and is not affiliated with or endorsed by OpenAI or any listed provider. Savings are not guaranteed; calculator results and paired-test deltas must be validated against the intended workload and quality bar.