Building in public · Waitlist open · Beta Q3 2026
Retken is the AI spend control layer we're building for the people whose Claude, ChatGPT, and API bills keep climbing without explanation. Join the waitlist to shape the product — the first 1,000 lock in Legacy Member pricing for life.
No spam. Early-access invites go out in waves as we onboard the first 1,000.
Free to join. See your savings before you pay a cent.
The Real Cost
You don't just lose money on bad outputs. You lose it on the wrong plans, retries, timeouts, and silent failures across every AI tool you use.
For professionals using 3+ AI tools daily, failed outputs, retries, and plan mismatch can quietly waste 20–35% of monthly spend.
Without spend visibility, users re-prompt the same failed request repeatedly — burning credits and wasting time every time.
OpenAI, Anthropic, and Groq expire prepaid API credits after 12 months — often with no warning. Documented user losses range from $60 to $28,000.
Supported Platforms
Retoken observes usage, spend, and errors across the platforms professionals use most.
Feature · Credit Expiry Timer
Most prepaid API credits on OpenAI, Anthropic, and Groq evaporate after 12 months — silently, with no email, no refund. Retoken tracks every credit balance you have, counts down to expiry, and tells you what to burn down before the timer hits zero.
Under the Hood
Our engine watches every AI interaction, spots waste, recommends cheaper paths, and generates dispute-ready evidence automatically.
How It Works
No workflow changes. You keep using your AI tools. Retoken handles logging, savings recommendations, and policy-aware recovery in the background.
Once you're off the waitlist, you'll link your AI tools via API key or OAuth in under five minutes. No code changes, no new habits — Retoken reads your usage, costs, and errors immediately.
Retoken captures spend patterns and failures, matches each against platform policy, and routes the result to save, recover, or monitor — instead of always auto-filing.
Depending on the platform, Retoken either files the claim for you, prepares it for your approval, or exports a ready-to-send package. You see the full audit trail in your dashboard.
Why Retoken
See what your AI bill actually looks like — and where Retoken saves, optimises, and recovers on your behalf.
Methodology
No black box. Every figure on your dashboard is computed from public platform data and your own usage — and we'll show you the formula.
current_plan_cost − best_fit_plan_cost
For each LLM, we model your last 90 days of usage against every published tier (free, Plus, Pro, Team, API pay-as-you-go). The delta against your current plan is the monthly savings figure.
Source: published platform pricing pagesΣ (premium_tokens × Δprice)
When a cheaper model could have answered a prompt at equivalent quality, we count the price difference as savings. Quality equivalence is determined by classification rules tuned on public benchmarks (MMLU, GSM8K, HumanEval).
Source: token pricing + benchmarksretry_tokens + over-spec_tokens
Tokens spent on duplicate retries, oversized context windows, and known-failure prompt patterns. We count the spend that would have happened if Retoken had not intervened.
Source: your own usage logsbalance × (1 − burn_rate × days_left)
For each prepaid balance, we project burn against historical usage. Anything not burnt by the platform's expiry window is flagged as at-risk — not as recovered savings, just as exposure.
Source: provider Ts & Cs (12-month expiry)Σ successful_claims (post-platform approval)
Only counted after the platform has approved the credit or refund. Pending claims appear separately. We never count "likely" or "eligible" credits in the recovered total.
Source: platform credit API receiptssavings + optimised + recovered
The headline figure on your dashboard. Prevention and optimisation typically do the heavy lifting; recovery is the smaller, harder-won piece. We show all three separately so you always know which lever moved.
Source: sum of the abovePricing
First 1,000 users become Legacy Members: 30-day free trial, Pro locked at $19/mo for life, founder Slack access, and priority on new features.
* Legacy Member pricing is locked for life. Capped at the first 1,000 users.
Why we exist
Early Feedback
"I had no idea how much I was losing to failed prompts. Retoken flagged $84 in claimable credits in the first week alone."
"The JSON error classification is genuinely impressive. It caught a Midjourney generation failure I'd completely written off."
"Finally — someone built the tool I've been manually doing in a spreadsheet for 8 months. Should've existed years ago."
FAQ
When the beta opens, you'll connect via API key or OAuth — the same way any developer tool integrates. We'll observe your usage and minimal prompt/response metadata to power savings insights and recovery. We never act outside what each platform's policy allows, and we never resell your data.
Yes. We only store metadata about failed outputs — timestamps, error codes, credit amounts. Prompt content and output text are hashed and never stored in plaintext. All data is encrypted at rest and in transit.
Most platforms (OpenAI ChatGPT, Anthropic Claude Pro/Max, Google Gemini, Perplexity, Cursor, GitHub Copilot) treat used credits as non-refundable. We don't pretend otherwise. Where recovery does work — API server errors, expiring unused credits, EU/UK/Turkey 14-day cancellation rights, published failed-job policies on Midjourney and Runway, and self-serve refund flows on DeepSeek and OpenRouter — Retoken automates it. Everywhere else, the bigger savings come from prevention and plan-fit, which is where most of your money actually leaks.
Yes, and it's a particular strength. Australian Consumer Law gives you statutory rights to remedies for AI services that fail to perform as advertised — rights no platform's terms of service can override. Retoken is built in Melbourne and flags ACL-eligible claims explicitly, preparing them in the format Australian users actually need.
No. Retoken is an AI spend control layer. It prevents waste with prompt linting, smart routing, and budgets, optimises plans and models per LLM, and then recovers credits where genuine failures still happen. Refunds are the last step, not the whole product.
Every platform is different, so Retoken offers three modes per integration: auto-file where the platform clearly allows it, approve-first where you click to send, and manual export where automation is risky or unsupported. We never bot support channels blindly.
No. Retoken works in the background. You use your tools exactly as you do today. The only difference is you start seeing what's wasted, what's recoverable, and which plan or model actually fits your usage.
At launch: OpenAI (ChatGPT + API), Anthropic (Claude), Google (Gemini), Midjourney, and Replit. We're adding Perplexity, Grok, Meta AI, DeepSeek, and Hugging Face in the months following launch. Founding users shape the priority order.
Limited Founding Access
The first 1,000 Legacy Members get 30 days free, founder pricing locked forever, founder Slack access, and priority on every new feature. We'll prove your savings before you pay anything — methodology open to inspection.
