Frequently Asked Questions

Kontinuity is a Chrome extension that tracks your token usage and costs across AI platforms — currently Claude and ChatGPT. It shows you in real time how many tokens each message uses, what it costs, and how your spending adds up over sessions and projects.

Install directly from the Chrome Web Store — just click "Add to Chrome". No account or credit card needed, and every feature is available straight away. Once installed, open Claude or ChatGPT and the K icon appears in your toolbar. Click it to see live stats.

Yes. Kontinuity currently supports claude.ai and chatgpt.com. It auto-detects which platform you're using and applies the correct model pricing. More platforms (Gemini, Grok) are on the roadmap.

No. Kontinuity works with the regular web interfaces at claude.ai and chatgpt.com — no API key required. It monitors network calls to count tokens, so your existing subscriptions work as-is.

100% on your device, in your browser's IndexedDB storage. No data is ever sent to our servers. We don't have servers that receive your usage data. Your conversations, costs, and session history are completely private to you.

Kontinuity intercepts the network stream to count characters and tokens — but it doesn't store or transmit your conversation content. Only token counts, costs, timestamps, and model names are saved locally. Your actual messages are never logged.

Kontinuity requests access to claude.ai and chatgpt.com — the sites it tracks. It also uses a few background permissions to keep tracking reliable: alarms and tabs to periodically check that tracking is active, notifications to alert you if a tab needs refreshing, and scripting to restore tracking if it stops. None of these access other websites or your browsing history. All data stays on your device.

Yes. Use the CSV export buttons in the Dashboard to download your full session and message history. To delete your data, go to Settings in the extension and use the clear data option, or clear site data for the extension in Chrome's settings.

A token is roughly 4 characters of text, or about ¾ of a word. AI models don't process raw text — they break your input into tokens and price their responses by the token. A typical sentence is 15–25 tokens. A long coding prompt might be 500–2,000 tokens. The exact breakdown depends on the model's tokenizer.

Very accurate for OpenAI models — Kontinuity uses a precise tokenizer that matches OpenAI's own encoding for exact counts. For Claude, it uses a calibrated heuristic that closely approximates Anthropic's tokenizer. Small variations from the exact count are possible, but the cost estimates are reliable for budgeting purposes.

Every message you send includes more than just your text. The full API payload includes system prompts (Claude's instructions), the entire conversation history so far, and any tool definitions. This "overhead" is often 5–10x your actual message — which is why costs add up faster than expected. Kontinuity shows both so you understand the real cost.

It reads the model identifier from the API response stream. For Claude this is the claude-opus-4-6 style slug; for ChatGPT it's the model field in the response. The pricing table is updated with current rates for all major models.

This means Kontinuity is loaded and ready, but hasn't intercepted any messages yet on this tab. Simply send a message on Claude or ChatGPT and the popup will update immediately with your token count, cost, and context health. If you just refreshed the page, the extension resets and waits for new activity.

This usually means the tracking connection was interrupted. Try these steps in order: (1) Send a message on the page — the popup updates after your first exchange. (2) Refresh the tab (Ctrl+R or Cmd+R) and send another message. (3) If it still shows dashes, go to chrome://extensions, find Kontinuity, click the reload button, then refresh your AI tabs. This restores the tracking connection.

Kontinuity runs a background health check every 30 seconds. If it detects that tracking has stopped on a tab (usually after a Chrome update or extension reload), it sends a notification asking you to refresh. Just click back to that tab and press Ctrl+R — tracking will resume immediately.

No. The popup is just a display window. Tracking runs in the background via content scripts injected into your AI chat tabs. You can close the popup, switch tabs, or minimize Chrome — tracking continues as long as the tab is open. Your data is always there when you reopen the popup.

Check two things: (1) The popup's "This Session" cost only counts messages sent while Kontinuity was active on that tab. If you started a conversation before installing or before refreshing, earlier messages aren't counted. (2) If you're using a model like Claude Haiku or GPT-4o-mini, costs are genuinely very low — often fractions of a cent per message. Check the "Full API Breakdown" section in the popup to see the full token count including system overhead.

Context health shows how much of the AI's context window is still available. As conversations get longer, earlier messages start getting pushed out. When health drops to Declining (yellow), the AI may start forgetting earlier context. At Critical (red), consider starting a new conversation. For Claude, this uses exact data from the server. For ChatGPT, it's an estimate based on token counting.

Your data isn't lost — it's in the Dashboard. The popup tracks the current session on the active tab. Refreshing the page starts a new session. All previous sessions, costs, and token counts are preserved in the Dashboard (click "Dashboard" at the bottom of the popup) and in your CSV exports.

Yes. Every feature is available to everyone — both platforms, the full dashboard, project tagging, conversation search, CSV/JSON export, 18 currencies, and unlimited history. There is no trial clock, no locked panel, and no payment details to enter. If you hit something that looks gated, that's a bug — tell me and I'll fix it.

That's the intention. If a paid version ever exists it would be something built specifically for teams, and what you use today would stay as it is. Features already in your hands won't be taken away and put behind a payment.

Because I can't see your bill, and I won't pretend otherwise. Kontinuity counts tokens locally and multiplies by published API rates. Providers have also stopped reporting some underlying data on newer models, so treat the number as a sense of scale — useful for comparing one conversation against another, not for reconciling an invoice. That's why it appears with ≈ throughout.

It changes what you actually pay, not what Kontinuity shows. On a flat monthly plan you pay the same regardless of usage, so the estimate is best read as "what this conversation would have cost at API rates" — a measure of how much work it took, not a charge you'll see.

A live reading of how full the conversation's context window is. As a chat grows, earlier messages get squeezed out and answers drift from what you originally asked. Context health tells you when you're approaching that point, so you can save your work and open a fresh chat before quality drops rather than after.

Coming in a future update. As a conversation grows longer, AI models start "forgetting" your original intent — earlier instructions get pushed out of the active context window. Drift detection will score each response against your stated project goals and flag when the AI starts going off-track, before you waste hours heading the wrong direction.

Coming in a future update. Tips will analyse your usage patterns and surface practical suggestions — things like "this conversation is 60 messages long, context drift risk is high, consider summarising and starting fresh." The goal is to help you work better in long sessions, not just watch a counter.

Send feedback about what's broken or missing, leave a review on the Chrome Web Store, or mention it to someone who works in long AI sessions. All three are on the Support Kontinuity page. None of it unlocks anything — there's nothing to unlock.

Not yet — Chrome only for now. Firefox and Edge support are on the roadmap. The extension is built with modern build tooling which supports cross-browser builds, so it's coming.