Sider is a browser sidebar from Vidline Inc. that puts ten AI models one click away on any page, then adds summarizing, translation, a cross-site browser agent and page-level customization. It runs on Chrome, Edge, Windows, macOS, iOS and Android.
Used this tool? Rate it
Used this tool? Rate it
Sider is an AI sidebar that lives inside your browser and follows you from page to page. Instead of copying text into a separate chat website, you open a panel beside whatever you are reading and ask questions about it directly. The product is operated by Vidline Inc., registered at 335 Huntington Ave in Boston, Massachusetts, and the same account works across a Chrome extension, an Edge extension, Windows and macOS desktop applications, and iOS and Android apps.
The positioning has widened considerably. What began as a chat panel now describes itself as an AI agent for the browser, organized around three verbs: Chat to understand, Claw to automate, Code to customize. Chat summarizes, explains and translates the page in front of you. Claw carries out multi-step tasks across real websites using your own logged-in sessions. Code rewrites how a site looks and behaves from a plain-language instruction and keeps the change for your next visit. Whether all three are equally mature is a fair question, and the limitations section addresses what can and cannot be verified about them.
The commercial proposition is simplest to state as model access. A single subscription reaches ten selectable models spanning GPT, Claude, Gemini, DeepSeek and GLM families, split into Basic and Advanced tiers — and only one of them, Sider Fusion, is the company's own name. Everything else is a third-party frontier model accessed through one interface and one bill. That, more than any individual feature, is what people are buying.
The model picker is the product's centre of gravity. Basic covers Sider Fusion alongside GPT-5.4 mini, Claude Haiku 4.5, Gemini 3.0 Flash, DeepSeek-V4-Flash and GLM-5; Advanced covers GPT-5.5, Gemini 3.1 pro, Claude Sonnet 4.6 and Claude Opus 4.6. Switching between them is a dropdown rather than a different subscription, which is the practical argument for using an aggregator instead of buying model subscriptions individually. Sider Fusion is the one name that is not a published third-party model; the site does not explain whether it is a trained model of its own or a routing layer over the others, so this page does not guess.
As a webpage summarizer the Chat layer summarizes, explains, translates and answers follow-up questions about web pages, PDFs and YouTube videos. Because the panel sits beside the page, the content is already in context — no copying, no tab switching. This is the oldest and most heavily used part of the product, and it is what the five million Chrome installs were mostly acquired on.
Claw is the automation layer, and one detail matters more than any other: it works across real websites and uses your logged-in sessions. That is what lets it compare prices while shopping, fill out forms, research across several sites, or triage a Gmail inbox — tasks that are impossible for a tool locked outside your authenticated accounts. It is also what makes the permission model worth thinking about before you turn it loose, since an agent operating with your credentials acts with your authority.
Code takes a plain-language instruction and edits the live page: restyle it, simplify it, add or remove elements, then keep those changes for future visits. It is an unusual capability for this category — closer to a user-script manager driven by natural language than to a chat assistant — and it is the clearest expression of the "agent for the browser" framing.
Each paid tier includes a cloud computer running OpenClaw, sized by plan: 2 vCPU and 2 GB of memory on Lite, 2 vCPU and 4 GB on Pro, 4 vCPU and 8 GB on Max, each with a 40 GB solid-state disk and each priced by the company at a stated monthly value. Bundling actual compute rather than only model credits tells you the agent features are meant to run somewhere other than your laptop.
Beyond the three main verbs, the site lists a long roster of separate entry points: Wisebase, Deep Research, Scholar Research, Math Solver, Rec Note, Audio to Text, Gamified Learning, Interactive Reading, ChatPDF, Web Creator, AI Slides, AI Essay Writer, an image generator and a family of image editors, plus AI Translator, Image Translator and PDF Translator. Breadth on this scale is a genuine convenience and also a reason to check that the specific tool you care about is more than a thin wrapper.
The Create layer produces images, videos, slides and documents, and the pricing page names the underlying engines: GPT Image 2, Nano Banana Pro and Seedance 1.5 Pro among others. As with chat, the value is aggregation — several generation models under one subscription — rather than proprietary technology.
The Chrome extension ships an interface in 55 languages, from major European and Asian languages through Arabic, Hebrew, Persian, Hindi, Bengali, Tamil, Telugu, Malayalam, Thai and Amharic. Very few tools in this category localize that broadly, and the testimonials the company displays — in Turkish, Korean and Arabic — are consistent with a genuinely multi-region user base.
The most common use is also the simplest: open a long article, report or documentation page, ask for a summary, then ask follow-up questions where the summary is thin. Because the sidebar reads the page you are on, this works on paywalled-in-browser content, internal wikis and anything else your browser can already display.
PDFs are treated as first-class input, both inside Chat and through the dedicated ChatPDF and PDF Translator entry points. For anyone reviewing contracts, papers or manuals, asking questions of the document beside the document is a materially different workflow from uploading it to a separate service.
Multi-model AI chat is the point here: because switching models is a dropdown, you can put the same question to a Claude model and a GPT model within seconds. For high-stakes writing or analysis, seeing where two frontier models disagree is often more informative than either answer alone.
Between the in-chat translation, the standalone translator, image translation and PDF translation, the product covers most of the ways translation actually comes up while browsing. The 55-language interface means the tool itself is usable in your own language, not just capable of translating into it.
Deep Research and Scholar Research exist as separate entry points, and Claw is described as researching, comparing and organizing information across the websites you already use. For literature scans or competitive research, the appeal is having retrieval and synthesis happen without you shuttling between twenty tabs.
Price comparison while shopping, form filling and inbox triage are the examples the vendor itself gives for Claw. These are low-stakes, high-frequency tasks where an agent failing is annoying rather than costly — a sensible place to start before trusting automation with anything consequential.
If a site you visit constantly has an irritating layout, Code lets you describe the fix and keeps it applied. This suits people with a small number of heavily used web applications more than casual browsing.
Math Solver, Gamified Learning and Interactive Reading are aimed squarely at students, and the mobile app description mentions helping children with homework. Note the age considerations in the limitations section before deploying it that way.
Start with the Chrome or Edge browser extension, since the sidebar is where nearly all of the product's value lives. The Chrome listing is version 5.32.2, updated 23 August 2026 — a useful signal that the extension is actively maintained rather than parked.
The same account carries across extension, desktop and mobile. Set it up once; the mobile apps are most useful as a continuation of desktop work rather than as a primary surface.
Understand the Basic and Advanced split immediately, because it governs both answer quality and how fast your allowance depletes. Use a Basic model for summarizing and routine questions, and reserve Advanced models for reasoning-heavy work. This single habit determines whether a plan feels generous or tight.
On any long page, ask for a summary first and read the original only where the summary raises a question. This inverts the usual reading order and is where most of the time saving comes from.
Because the agent runs with your logged-in sessions, begin with tasks whose failure costs nothing — price comparison, drafting rather than sending, reading rather than writing. Build a sense of its reliability before letting it touch anything that submits, purchases or deletes.
The pricing FAQ includes a dedicated question about checking credit usage and consumption details, which tells you it is a common source of confusion. Find that screen in your first week rather than at the moment you run out.
If you plan to publish or sell what you generate, read both the terms and the privacy policy before you do — they say different things about who owns generated content, and the limitations section below explains the conflict.
Match the model tier to the task. Advanced models cost more of your allowance per call. Summarizing a news article does not need Claude Opus; a legal analysis might. Treating every query as equally important is the fastest way to exhaust a plan.
Ask the sidebar about the page, not in general. The reason to use an in-page panel rather than a chat website is context. Questions phrased as "what does this section claim about X" get better answers than questions you could have asked anywhere.
Verify anything factual before you rely on it. The terms state plainly that the company does not guarantee the accuracy or suitability of generated content and that verification is your responsibility. For anything consequential, treat output as a draft.
Be deliberate about which sites the agent may touch. Claw acting through your logged-in sessions is exactly what makes it useful and exactly what makes it risky. Think about banking, email and admin panels specifically rather than in general.
Do not use it where regulated compliance applies. The terms say the services are not designed to comply with industry-specific regulations, naming HIPAA and FISMA. If you handle protected health information or federal systems data, this is a stated exclusion rather than a gap you can manage around.
Check the retention terms if you work with sensitive material. The stated retention period for account and communication data is measured in decades, not months. That is a legitimate reason to keep genuinely sensitive content out of the tool entirely.
Save Code changes only where they help. Persistent page rewrites are powerful on sites you use daily and confusing on sites you visit rarely — especially if you later forget a page is modified.
Test on the free surface before subscribing. Entry pricing starts at $25 a month, which is above the category's usual $10 to $20 band. Confirm the models you actually want are available and the workflow suits you before committing at that level.
Heavy browser users who read, research and write inside a browser all day get the most from it, because the sidebar's advantage compounds with the number of pages you process.
People who want several frontier models without several subscriptions are the clearest fit. If you would otherwise pay separately for two or three model providers, one subscription reaching ten models is straightforward arithmetic.
Students and researchers are directly targeted by Math Solver, Scholar Research, Interactive Reading and the study-oriented tools, and the PDF handling suits reading-heavy work.
Multilingual workers benefit from both the translation tooling and the 55-language interface, which is unusually broad for this category.
Early adopters of browser agents who want to experiment with cross-site automation will find Claw plus a bundled cloud computer a more complete package than most competitors offer.
Who should be cautious: anyone in a regulated environment, given the explicit HIPAA and FISMA exclusion; anyone who needs refund flexibility, since fees are stated as non-refundable; anyone uncomfortable with decades-long data retention; and anyone whose commercial publishing plans depend on unambiguous content ownership, given the contradiction between the two policy documents.
Sider distributes across six channels under one account. Figures below were verified first-hand on 24 August 2026 and store data changes daily.
Chrome extension — the flagship surface, rated 4.9 stars across 114,200 ratings and 5,000,000 users, carrying a Featured badge, with Google noting the publisher has no history of violations and follows recommended practices. This is by a wide margin the largest evidence base for the product.
Edge extension — listed on the Microsoft Edge Add-ons store and linked from the official download page.
iOS — listed as Sider: Chat & Work with All AI, published by Vidline Inc., rated 4.73 stars across 4,257 ratings. Version 3.9.0 released 17 August 2026; first published 11 April 2023. Free, categorized under Productivity, and rated 17+.
Android — 4.5 stars across 19,699 ratings with more than a million downloads.
macOS — a separate App Store listing, Sider: AI Chat & Search, version 3.0.4 from 31 July 2026, rated 4+. It currently shows zero ratings, so no conclusion about the desktop experience can be drawn from user feedback.
Windows — a native application downloadable from the site.
Interface languages — the Chrome extension ships 55 localized interfaces, including Simplified and Traditional Chinese, Japanese, Korean, Russian, Arabic, Hebrew, Hindi, Bengali, Tamil, Telugu, Malayalam, Thai and Amharic.
Three individual tiers are published, all priced monthly, with a team view also available and yearly billing advertised at 17% off. Figures below are from the official pricing page as read on 24 August 2026.
Lite — $25 per month. Described as suitable for most everyday tasks. Includes a cloud computer with 2 vCPU, 2 GB of memory and a 40 GB disk, which the company values at $10 per month.
Pro — $50 per month. Positioned as the higher-quota option, with a 2 vCPU, 4 GB cloud computer valued at $15 per month.
Max — $100 per month. The top individual tier, with a 4 vCPU, 8 GB cloud computer valued at $20 per month.
The company's own framing is that plans differ only by monthly credits and cloud computer power, and that all plans include all features — so the decision is about volume, not capability. That makes the credit allowance the number that matters most, and here is the honest limitation: the credit figure renders blank on the public pricing page when you are not signed in. The page shows the label without a number. Rather than repeat figures from affiliate review sites that could not be verified, this page simply advises checking the allowance after signing in, before subscribing.
Two further points. First, fees are non-refundable unless otherwise stated according to the terms, so treat any subscription as a committed spend. Second, the pricing FAQ's own question list — invoices, refunds, credit usage checks, cross-platform use, when credits reset, account sharing, monthly-to-yearly upgrades — is a fair map of where billing questions cluster.
For context, $25 as an entry point sits above the typical $10 to $20 of single-model consumer subscriptions. The bundled compute and multi-model access are the justification; whether that trade works depends on how much you would otherwise be paying separately.
Monica, Merlin and MaxAI occupy the same niche — a browser sidebar aggregating multiple models — and compete primarily on price, model roster and credit generosity. If aggregation is all you need, these are direct substitutes worth pricing against each other.
Going direct to a model provider is the obvious alternative if you use one model overwhelmingly. A single first-party subscription is usually cheaper and gives you that vendor's newest features first; you give up model switching and in-page context.
Browser-native AI built into Edge, Chrome and newer AI-first browsers covers basic page summarizing at no cost. For light users this may be sufficient, and no extension permissions are involved.
Dedicated agent tools are worth comparing if cross-site automation is your actual goal rather than a bonus. Sider's advantage is that the agent sits beside a general assistant; a specialist tool may be more capable but does less.
Reader and note tools such as dedicated PDF assistants or knowledge bases will beat a generalist sidebar on depth if your work is concentrated in one of those areas. The Wisebase feature aims at that territory but is one component among many rather than the whole product.
The two policy documents contradict each other on content ownership. The terms of use, last updated 8 May 2025, state that you retain ownership of AI-generated content with full rights to use, modify, distribute and commercialize it, and that the company does not claim ownership over content generated from your inputs. The privacy policy, meanwhile, states that the user, Sider.AI and potentially third-party providers such as Stability, OpenAI and Claude may jointly hold the copyright to generated content, while reserving to the company a right to use that content for internal research, product improvement and demonstration. Both documents are currently published. The practical advice is to assume the more restrictive reading if you intend to publish or sell the output, and to seek written clarification for anything commercially significant.
Stated data retention is measured in decades. The GDPR section says account-related information is retained for 99 years after the account is closed and communication information for 99 years from the last communication, with only usage and analytics data on a three-month clock. Ninety-nine years is an unusual figure to find inside a section about data minimization, and it is quoted here verbatim rather than paraphrased because it materially affects anyone weighing what to put into the tool.
Self-reported scale does not match the store. The site claims 10M+ active users and 100K+ five-star ratings, while the Chrome listing shows 5,000,000 users and 114,200 total ratings, not five-star ratings specifically. The site figure may aggregate every platform, but it is vendor-reported and unaudited either way, and the two should not be read as the same claim.
Trustpilot is negative but the sample is far too small to conclude anything. The profile shows 2.4 out of 5 from just 15 reviews, well under any threshold for a quality judgment, and it is unclaimed with no history of the company inviting reviews. It is worth reading for themes rather than score: complaints cluster on subscriptions and support, and one named reviewer holding an annual plan for a team of seven describes being locked out across devices after travel and receiving slow, templated support replies. Weigh that against more than 138,000 store ratings across three platforms averaging well above 4.5.
Regulated use is explicitly excluded. The terms state the services are not designed to comply with industry-specific regulations such as HIPAA and FISMA and should not be used where such compliance is required. This is a stated boundary, not an oversight.
Credit allowances are not publicly visible. Because the credit numbers do not render for signed-out visitors, the single most decision-relevant figure — how much you actually get for $25, $50 or $100 — cannot be established before creating an account. That is a genuine transparency gap for a product whose tiers differ only by volume.
The agent runs with your credentials. Claw's use of logged-in sessions is a design choice that trades security surface for capability. An agent acting inside your authenticated accounts can do anything you can do there, and the site does not publish guardrails, site allowlists or failure rates for it.
Age ratings are inconsistent across the company's own apps. The iOS build is rated 17+ on iOS while the macOS build is rated 4+, for what is broadly the same assistant. Apple's listing returns no specific content advisories explaining the 17+ classification; unrestricted web access is the usual reason for such a rating on AI chat apps. The privacy policy separately states that the services are not intended for use by children under the age of 13 and that the company does not knowingly collect their data. Parents deploying it for homework help should note both facts.
The site is inaccessible to ordinary tools. sider.ai returns HTTP 403 to ordinary clients, including its own legal pages, and the sitemap it advertises returns no entries. Every official fact on this page was obtained through a rendered browser session. This does not affect normal users but it does make independent verification harder than it should be.
Company history could not be verified. Widely repeated claims about a 2012 founding in Tokyo, an origin as a code review tool, and specific revenue figures appear only in secondary review blogs and database summaries; they could not be confirmed first-hand, and the domain associated with the supposed predecessor product no longer resolves. Given how commonly the name "Sider" is reused, this page attributes only what the Chrome Web Store developer record and the official terms establish: Vidline Inc., Boston.
No independent technology press coverage exists. Searches of major technology outlets returned no independent review or reported article. The available third-party evidence is app store data — which is substantial — plus a small Trustpilot profile. Affiliate review blogs were deliberately excluded as sources.
Ten selectable models across two tiers. Basic includes Sider Fusion, GPT-5.4 mini, Claude Haiku 4.5, Gemini 3.0 Flash, DeepSeek-V4-Flash and GLM-5. Advanced includes GPT-5.5, Gemini 3.1 pro, Claude Sonnet 4.6 and Claude Opus 4.6. For image and video generation the pricing page names GPT Image 2, Nano Banana Pro and Seedance 1.5 Pro. All of these except Sider Fusion are third-party models; Sider Fusion is the company's own name and its composition is not explained publicly.
Overwhelmingly the latter, and the site is open about it — every model in the picker except Sider Fusion carries a third-party vendor's name. The value being sold is the interface layer: the in-page sidebar, session-aware automation, page rewriting and a single bill covering many providers. That is a legitimate product, but it means model quality is inherited rather than proprietary, and features may follow upstream releases rather than lead them.
Three individual tiers: Lite at $25 per month, Pro at $50 and Max at $100, with yearly billing advertised at 17% off. Every tier includes all features and a cloud computer, differing only in monthly credits and compute size. The catch is that the credit numbers do not display to signed-out visitors, so you cannot compare allowances until you have an account. Fees are stated as non-refundable.
Each paid tier bundles a cloud machine running OpenClaw that acts as the execution environment for agent tasks: 2 vCPU and 2 GB of memory on Lite, 2 vCPU and 4 GB on Pro, 4 vCPU and 8 GB on Max, each with a 40 GB solid-state disk. The company assigns each a stated monthly value of $10, $15 and $20 respectively. Bundling real compute rather than only model credits indicates the automation features are meant to run off your own machine.
This is genuinely unclear from the published documents. The terms of use say you retain ownership and full commercial rights, and that the company claims no ownership over content generated from your inputs. The privacy policy says copyright may be jointly held by you, Sider.AI and third-party providers, while reserving company use for internal research, product improvement and demonstration. Until that is reconciled, assume the more restrictive position for commercial work and ask support in writing.
Per the privacy policy, account-related information is retained for 99 years after the account is closed unless law requires longer, and communication information for 99 years from the last communication. Usage and analytics data is kept three months and then anonymized or deleted. EEA residents retain GDPR rights of access, correction, deletion, portability, objection and restriction.
That depends entirely on which accounts. Claw is explicitly designed to work through your existing sessions, which is what enables useful cross-site tasks and also means it operates with your full authority on any site it touches. The company publishes no allowlist, guardrail description or reliability data for it. A sensible approach is to start with reversible, low-stakes tasks and to think carefully before exposing financial, email or administrative accounts.
The terms answer this directly and negatively: the services are not designed to comply with industry-specific regulations, naming HIPAA and FISMA, and should not be used where such compliance is required. Combined with the 99-year retention terms and the unresolved content ownership question, organizations with compliance obligations should treat this as a consumer tool and seek contractual assurances before any regulated deployment.
On the platforms with meaningful samples, well. The Chrome extension holds 4.9 stars across 114,200 ratings with 5 million users and a Featured badge; Android is 4.5 across 19,699 ratings; iOS is 4.73 across 4,257. Trustpilot shows 2.4, but from only 15 unclaimed-profile reviews, which is too small a sample to weigh against the others — read it for recurring themes such as support responsiveness rather than as a score. No major technology publication has reviewed it independently.
Yes. One account covers a Chrome extension, an Edge extension, Windows and macOS desktop applications, and iOS and Android apps. In practice the Chrome extension is where the product is most mature — it holds essentially all of the user base and review volume — while the macOS App Store listing currently has no ratings at all, so desktop maturity is harder to judge from public evidence.
