Legal

Privacy Policy

Effective July 31, 2026

The short version
  • We receive the text you explicitly ask about — nothing is captured in the background.
  • We do not store that text. Attribution is computed and the content is discarded when the response is returned.
  • Your chats are saved in your browser, not on our servers, and you can clear them in one click.
  • We keep usage records for billing. They count tokens; they do not contain your content.
  • We do not sell your data, use it for advertising, or train models on it.

Who and what this covers

This policy explains how TokenPath handles data across the TokenPath API, the tokenpath.ai website and customer platform, and the TokenPath browser extension (“Browse with TokenPath”). TokenPath provides token-level attribution: it maps spans of a generated answer back to the source spans that produced them.

If you use the API directly, you are the controller of the documents you send and are responsible for the notices and permissions your own users require. TokenPath processes that content to answer the request you made.

What the browser extension sends us

The extension reads a page only in response to something you do. There are exactly two ways to start: clicking the TokenPath toolbar icon, or selecting text and choosing Chat with TokenPath from the context menu. Switching tabs, scrolling, and ordinary browsing capture nothing. Its content script has no network access of its own — it cannot send anything anywhere. It waits to be asked for text and returns it to the panel, which is the only component that talks to our API.

One thing worth being precise about: Auto-summarize is on by default, so a toolbar click both captures the page and immediately asks for a summary — that single click sends text to us. A context-menu capture generates nothing on its own; it shows you what it captured and waits for your question. Turning Auto-summarize off in Settings makes the toolbar click behave the same way.

The text you capturedWhen you click the toolbar icon, or capture from the context menu

Whichever one you chose: the text you selected, the readable text of the page, the text of a PDF, or a YouTube video's subtitle transcript. PDF text is extracted locally in your browser and sent as plain text — we never receive the PDF file itself, and neither do our vendors. Long documents are truncated before sending, and the panel tells you when that happened.

The site's originWith each generation request, inside the prompt

The scheme and hostname only — for example https://example.com. This tells the model what it is reading. The rest of the address, including the path, query string, and fragment, is never sent. Local and non-web pages are described only as “the current webpage”. Attribution requests carry no address at all.

Your question and the chat turns before itWhen you send a message

Your typed question, plus the captured text again and the earlier messages in that conversation, so follow-ups make sense. Every follow-up re-sends the captured document; earlier turns are dropped oldest-first to fit the request.

The generated answerAutomatically, right after each answer finishes

Tracing an answer back to its source is a second call that needs the answer, the question, and the source text together in order to map answer spans to source spans. The panel runs it for you as soon as an answer completes — you do not have to click a span first — so the answer text reaches us whether or not you end up tracing it.

Web pages and PDFs can contain personal information and private correspondence. Please treat the pages you send the same way you would treat pasting them into any hosted AI tool, and avoid sending content you are not permitted to share.

What the extension never sends

  • Your browsing history, or the pages you visit but never invoke TokenPath on
  • Full URLs — the path, query string, and fragment never leave your browser
  • Cookies, credentials, passwords, autofill data, or form fields
  • The structure of the page — the DOM stays in the page; only text crosses the boundary
  • Analytics, telemetry, crash reports, or advertising identifiers of any kind — the extension contains no analytics code and contacts no third-party host
  • Anything at all until you capture something with the toolbar icon or the context menu

Why the extension asks for so much access

Chrome warns that TokenPath can “read and change all your data on all websites.” That warning is accurate about what the permission allows, and misleading about what the extension does with it. Here is the whole of it:

  • Access to all websites Lets you capture whichever page you are reading, and lets the extension find and highlight the source passage when you trace part of an answer. It also keeps the current tab's address readable so the panel can restore that page's saved chat and refuse a capture that is out of date after you navigate away.
  • Context menu Adds the single Chat with TokenPath item. It is the only item TokenPath adds.
  • Side panel Shows the chat beside the page you are reading.
  • Active tab and scripting Reaches the tab you invoked TokenPath on, restores capture support on a page that was already open, and checks whether the tab is Chrome's built-in PDF viewer.
  • Storage Holds your API key and your settings, on your machine.

All-site access exists because you may want to capture any page you are reading, and because tracing an answer back to that page means finding and highlighting the passage again later. It is exercised only when you invoke TokenPath. It is not used to observe your browsing.

What stays on your device

Most of what TokenPath knows about your reading never leaves your machine. Three things are stored locally:

Your API keyExtension storage

Saved so you do not have to paste it again. Local to this browser profile; never synced to your other devices.

Saved chats, one per pageIndexedDB

The text that page captured, the questions and answers, and the source maps that let a traced phrase find its passage again. Kept so reopening a page brings its conversation back.

Your settingsLocal storage

Theme, summary style, whether Auto-summarize and follow-up suggestions are on, and any custom summary instructions you wrote.

Saved chats are what make a page you already asked about reopen with its conversation intact, and what keeps attribution working after a refresh. They contain the captured text, so they are worth knowing about. They are pruned automatically: the extension keeps at most 50 pages and deletes anything older than 30 days. Nothing here is synced to another device, and no copy of it reaches us.

To clear it yourself: Clear chat deletes the current page's saved chat. Disconnect removes your API key and every saved chat for every page. Uninstalling the extension removes all of it, settings included.

API keys and account data

Your TokenPath API key is stored in your browser's local extension storage. It is never synced across your devices. It is sent only to TokenPath, as an HTTP Authorization header, over HTTPS — the extension hard-codes the list of API origins it will talk to and rejects every other destination, so your key and your captured text cannot be redirected to another host. The published build talks to api.tokenpath.ai. It is never sent to any other site, never included in page content, and never written to our logs in raw form. Disconnecting in the extension, or uninstalling it, deletes the key from your browser. You can revoke a key at any time from the customer platform, which immediately stops it from working.

If you hold a TokenPath account, we store your email address, your API key records, your token balance, and your purchase history so we can authenticate you and bill correctly.

How long we keep things

Request content — the page or PDF text, the site origin, your questions, and the answers — is processed to serve the request and is not written to a database. It exists in server memory for the life of the request and is discarded afterward. The extension does not use our document-storage feature, so nothing you ask about is saved to an account.

Usage records are kept indefinitely as billing history. Each record contains a customer ID, an API key ID, request and billing IDs, token counts, cost, and a status — and no part of your content.

Operational logs record the request path, HTTP status, latency, customer and key IDs, and billing decisions. They are retained for 30 days and then deleted automatically. Document text, questions, and answers are not logged, and raw API keys are not logged.

Conversation history is kept on your machine, not ours — in your browser, capped at 50 pages and 30 days, and clearable at any time. See what stays on your device. TokenPath holds no copy of it.

Who else processes it

We use a small number of vendors to run the service. Generating an answer necessarily involves sending the page text and your question to a model provider; attribution itself runs on infrastructure we operate.

VendorPurposeData involved
OpenRouterRoutes the generation request to the answering modelPage text, site origin, questions, chat turns
Google (Gemini)Generates the answer for the API and the extension, through OpenRouterPage text, site origin, questions, chat turns
Anthropic (Claude)Generates the answer in the customer platform's playground, through OpenRouterPlayground document text and questions; not extension requests
Amazon Web ServicesRuns the API, the account database, and the log store (us-west-2)Request content in transit and memory; usage metadata at rest
CloudflareFronts and protects the API endpointRequest content in transit; connection metadata
VercelHosts tokenpath.ai and the customer platform, and stores files you upload thereWebsite and dashboard traffic; PDFs and extracted text you upload to the playground; not extension requests
Adobe (PDF Services)Extracts text and layout from PDFs uploaded to the customer platformPlayground PDF files; not extension requests, which never send a file
PostHogProduct analytics for the tokenpath.ai marketing site onlyWebsite page views and clicks; no extension or API request data
ClerkSigns you in to your TokenPath accountEmail address and authentication data
Dodo PaymentsMerchant of record for credit purchasesBilling and payment details

Content passed to a model provider is handled under that provider's terms and privacy policy, linked above. TokenPath does not authorize them to use your content to train their models, but we cannot make guarantees on their behalf — if that distinction matters for your use case, contact us before sending sensitive material.

What we don't do

  • We do not sell or rent your data, and we do not transfer it to third parties except to the vendors listed above.
  • We do not use your content for advertising, profiling, or any purpose unrelated to answering your request.
  • We do not train models on your documents, questions, or answers.
  • We do not use your data to determine creditworthiness or for lending.

Your choices and deletion

Because request content is never stored, there is nothing on our servers to delete for a past question — it is already gone. What you can act on:

  • Stop sending data at any time by closing the panel, disconnecting your API key, or uninstalling the extension.
  • Revoke or rotate API keys from the customer platform.
  • Delete this page's saved chat with Clear chat, or every saved chat and your API key with Disconnect. Uninstalling removes everything TokenPath stored locally.
  • Turn off Auto-summarize in Settings if you would rather a toolbar click never send anything until you ask a question.
  • Request access to, correction of, or deletion of your account data — email address, keys, and usage history — by writing to us. Deleting usage history may be limited where we must keep records for tax and accounting.

Depending on where you live, you may have additional rights over your personal data, including the right to object to processing or to lodge a complaint with your data protection authority. Write to us and we will honor them.

Security and transfers

All traffic to the TokenPath API is encrypted in transit over HTTPS. API keys are stored as hashes, and access to production systems is restricted to the people who operate them.

Our API, account database, and log store run in the United States (AWS us-west-2). The GPU cluster that computes attribution is hardware we operate ourselves, also located in the United States; it receives the document, question, and answer for each attribution request and stores none of them. Our model providers may process requests in other countries. Wherever you use TokenPath from, your data will be transferred to these locations.

Children

TokenPath is a developer tool and is not directed to children under 13, and we do not knowingly collect their personal information.

Changes to this policy

If we change how we handle data, we will update this page and move the effective date above. Material changes to what we collect or how long we keep it will be announced to account holders by email.

Contact

Questions about this policy, or a request about your data, go to support@tokenpath.ai. For enterprise agreements or a DPA, write to sales@tokenpath.ai.

← Back to TokenPath