Your team's wiki,
finally answers back.An AI reading assistant that chats with your PDFs, docs, and wikis — and cites every answer.

NotebookLM is a research workspace you upload sources into. Sidenote lives on the page you're reading. We optimise for one thing: in-browser, with citations that scroll the source. We don't do audio overviews; we do “click this chip, see the paragraph”.

Yes. Confluence reads through the Chrome extension on the page you can already see — no integration to set up, no admin install, no SSO scopes to grant, and your wiki content goes through your own account, not a shared pipeline. Notion, Google Docs, Slack canvases, and SharePoint & OneDrive connect through each provider's official API with read-only access to the documents you choose to open, again through your own account. SharePoint and OneDrive sign in with Microsoft (delegated, read-only Graph scopes); on locked-down Microsoft 365 tenants a one-time admin approval may be needed.

No. Anthropic and Voyage AI run with no-training defaults on the API tier we use, and we never fine-tune models on user content. Your documents are used only to answer your own questions.

Web articles (Readability extraction) and browser-rendered PDFs work today, alongside dedicated integrations for Confluence, Notion, Google Docs, Slack canvases, and SharePoint & OneDrive. For everything else, drop a PDF, DOCX, PPTX, TXT, or MD file onto the side-panel and Sidenote ingests it the same way. Uploads chat headlessly — citation chips highlight passages in the upload's text rather than scrolling a tab you don't have.

Every answer is checked server-side against the retrieved passages before it leaves the API. If a claim has no source, the chip is dropped — you only ever see citations the model could actually back up. Click a chip and Sidenote scrolls the document to the exact passage and pulses it in amber. When the model bookends a multi-sentence topic by quoting first and last sentence, each cited sentence is highlighted on its own — the gap in the middle stays plain, so you see exactly what was used to ground the answer.

Chat runs on Anthropic Claude — Sonnet on Pro for fast multi-doc chat, Opus on Pro+ and Pro Max for sharper reasoning across long, dense docs. Chunk summaries and glossaries use Claude Haiku. Embeddings come from Voyage AI's voyage-3. We pick the model so you don't have to.

Data lives in Supabase's eu-west-2 (UK) region, encrypted at rest, with row-level security so no two accounts can read each other's chunks. Store mode keeps the document for the lifetime of your account, or until you delete it. Discard mode purges the document — including its chunks and any orphaned shared content — within 24 hours of upload via an hourly background sweep.

Sign up with email, click the magic link, and the 7-day Pro trial starts — no card required. Trial caps are 3 documents, 10 explanations, 3 glossary builds, and 10 chat turns across the week. We only ask for a card when you convert. After that, one click in the dashboard opens the Stripe Customer Portal — cancel, change plans, or update your payment method any time. You keep Pro access until the end of the current billing period.