For Confluence

AI assistant for Confluence — built for the page you’re on.

Sidenote is a Chrome extension that turns your Confluence wiki into something that answers. Open a page, ask a question, click a citation — Sidenote scrolls the live page and highlights the exact passage that proves the answer. No admin install, no server-side ingestion, works on pages behind your company login.

The problem

Confluence search wasn’t built for answers.

Every fast-growing team has the same dirty secret. Twelve thousand pages, half of them stale, the same five answers re-asked in standup every week. The current runbook lives somewhere between Deploy runbook (NEW) and Deploy v1 (DEPRECATED), and Confluence's built-in search ranks them roughly the same.

Sidenote doesn't replace your wiki. It teaches it to answer — in your browser, on the pages you already have open, with citations you can verify in one click.

Confluence's built-in AI surfaces results from the wiki you already have, complete with whatever staleness it has. Sidenote shows you the passage so you can decide if it's still true.
How it works

How Sidenote works on Confluence.

Open a Confluence page.Sidenote rides along in the side panel of any Chrome tab. As soon as you open a Confluence page, the extension extracts it through a dedicated wiki extractor — clean text, headings, tables, links.
Ask the question you'd ask a senior.“Where do we deploy hotfixes?” “Which spec is current?” “What's the SLA for refunds?” Sidenote queries across every Confluence page you've opened — not just the one in front of you.
Click a citation, see the passage.Every claim is grounded in a quoted chunk of the source page. Click the chip; Sidenote scrolls the live Confluence tab to the exact paragraph and highlights it in amber.
Working across a handful of related pages? Bundle them into a Collection and query the whole set in one go — citations name the source page.
Use cases

What teams ask Confluence — and how Sidenote answers.

  • “Where do we deploy hotfixes?”Sidenote answers from the current runbook and quotes the exact step — not the deprecated v1 page two hits below.
  • “Who's on call this week, and how do they escalate?”It reads the rota page and the escalation matrix in the same answer, with a citation back to each.
  • “Why did we pick Postgres over DynamoDB?”ADRs are gold buried under a year of Slack threads. Sidenote surfaces the decision and the trade-off paragraph.
  • “How do we handle refunds for enterprise customers?”Support reps stop refreshing the policy page. Sidenote quotes the clause that applies and shows the date it was last reviewed.
  • “What's the rollout plan for feature X?”PRDs, milestone pages, and roadmap docs queried together — citations point at the row in the rollout table.
  • “I'm new — how do I get set up on day one?”New hires query the wiki directly instead of pinging seniors. Onboarding stops being an interrupt.
Private by design

Built for private pages — because it runs in your browser.

Server-side Confluence AI has a structural problem with restricted pages: it fetches your wiki on someone else's infrastructure, with someone else's OAuth credentials, against an API that returns whatever those credentials are allowed to see. Restricted spaces, draft pages, pages behind extra ACLs — most of them are invisible to that pipeline, or require an admin to grant broader scopes than the organisation is comfortable with.

Sidenote sidesteps the whole question. It reads the same DOM you read, on the same session you're already signed into. If you can open the page, Sidenote can answer about it. If you can't, neither can Sidenote. Permissions stay exactly where Confluence put them.

Your wiki content stays in your account. Each user's pages live behind row-level security in our UK (eu-west-2) Supabase region — no two accounts can read each other's chunks. Anthropic and Voyage AI both run with no-training defaults on the API tiers we use; we never fine-tune models on user content.
Compared

Sidenote vs other Confluence AI tools.

Three honest framings. The shorthand each time: Sidenote is the lightweight, per-user, browser-native option — designed to coexist with heavier platforms, not replace them.

vs Atlassian RovoAtlassian's own AI — tightly integrated with Cloud, sold as a per-seat add-on. Server-side, Cloud-only. Sidenote runs in your browser, so it works with Cloud, Server, and Data Center, and there's nothing for an admin to install.
vs eesel AIIngests your Confluence server-side via OAuth and builds a separate index. Sidenote stays per-page and opt-in: nothing is read until you open a page yourself. There's no global crawl of your wiki and no third-party-hosted copy of every doc.
vs GleanEnterprise search across many tools — powerful, expensive, and heavyweight to deploy. Sidenote is a £10/month-per-user extension your team can try this afternoon, and the citation chips highlight the exact passage inside the live Confluence page.
FAQ

Common questions about Sidenote for Confluence.

Does Sidenote need a Confluence admin install?

No. Sidenote is a Chrome extension — your team installs it from the Chrome Web Store the same way as any other extension. There's nothing to deploy in Confluence itself, no Forge app to approve, no OAuth scopes for an admin to grant. Each user signs into Sidenote with their email; Confluence is read locally from their own browser session.

Can it read private Confluence pages?

Yes — and this is the part most server-side tools cannot do. Sidenote runs inside your browser, on the session that's already signed into Confluence. It reads the same DOM you can see. Pages behind your company SSO, restricted spaces, draft pages — anything you can open in your browser, Sidenote can answer questions about.

Does Sidenote ingest our entire Confluence?

No. Sidenote is per-page and opt-in: it only indexes a page once you've opened it with the side panel active. There's no server-side crawl of your wiki, no Confluence API integration, no global content store. If a page is never opened, it's never read. Use Collections to bundle the pages you do open into a group you can query in one go.

Is our Confluence content used to train AI models?

No. Anthropic (the model provider) and Voyage AI (the embedding provider) both run with no-training defaults on the API tiers Sidenote uses, and Sidenote never fine-tunes models on user content. Your wiki content is used only to answer your own questions and is stored in a UK (eu-west-2) region with row-level security isolating every account.

How is this different from Atlassian Rovo?

Rovo is Atlassian's own AI for Confluence Cloud — tightly integrated, but locked to Atlassian Cloud, sold as a per-seat add-on, and operated server-side. Sidenote is a lightweight per-user extension that works with Cloud, Server, and Data Center because it reads from your browser rather than the Atlassian API. The two can coexist; the value Sidenote adds is cross-tab, browser-native answers and citation chips that scroll the live page.

Does Sidenote work with Confluence Cloud, Server, and Data Center?

Yes — because Sidenote reads what your browser renders, not what an integration API exposes. Confluence Cloud, Server, and Data Center all serve the page DOM in the same way to your browser; Sidenote's Confluence extractor works against all three. The same applies to self-hosted intranets that render HTML.

Try it on your wiki

Make your Confluence answer back.

Add Sidenote to Chrome, open a Confluence page, and ask the question you keep asking your senior engineer. No admin install. Free tier forever, 7-day Pro trial — no card required.

Works with Confluence Cloud, Server, and Data Center