Merlin

Where Merlin sits

Merlin is Foyer Tech's attempt to make the browser the single place where a knowledge worker meets AI. The company launched it in 2022 out of India with backing from Better Capital and angel money from executives at HubSpot, Oyo and Uber, and it grew fast — the Chrome Web Store listing I looked at shows roughly 900,000 users and a 4.8 average across about 58,800 ratings. The pitch is consolidation: instead of paying separately for ChatGPT, Claude and Gemini and flipping between tabs, you get a side panel that runs whichever model you pick, on top of whatever page you are already reading. That pitch has aged reasonably well, because model subscriptions have only gotten more plural since 2022, and people genuinely do juggle three or four assistants now. It is not the only extension doing this — MaxAI, Monica and Sider all chase the same slot — so what matters is execution depth and how honest the usage economics are. On both counts I found a product that is broader than most of its rivals and more transparent than some, but with a free tier that evaporates faster than the headline number suggests.

What it actually does

At its core Merlin is a sidebar (Ctrl/Cmd+M on any tab) plus a set of page-aware helpers. The chat side supports multiple frontier models from OpenAI, Anthropic, Google, Meta, Mistral and xAI, and one detail I appreciate: threads are model-agnostic, so you can start an answer with one engine and switch mid-conversation without losing the thread. Around that sit the page-level tools — summarise a web page or a YouTube video, chat with a PDF or DOCX, translate, rewrite selected text inline, draft Gmail and LinkedIn replies in place, and pull real-time web results into answers. Two features stand out as more than wrappers. Projects lets you attach your own context — Notion spaces, Google Drive folders, S3 buckets or a list of URLs — and builds a retrieval layer over it so repeated questions get answers grounded in your material. Crafts generates artifacts: React/Shadcn app snippets, Mermaid diagrams, SQL schemas, Python notebooks, with live preview and one-click export to GitHub, CodeSandbox or a ZIP. On top of that the store listing enumerates utilities I would call supporting cast: an AI detector and "humanizer", OCR for screenshots, voice-to-text, a prompt library triggered by "/" anywhere online, and image generation across 20+ models per the company's own site. Merlin runs as a web app, a Chrome/Edge/Brave extension, a native macOS app with what the company calls "Merlin Agent", and iOS/Android apps with synced history, all on one account. The model roster itself is worth a sentence, because it is the product's spine: the company says it surfaces new engines as providers ship them — the lineup listed at the time of writing runs from GPT-class and Claude-class models through Gemini 2.5, DeepSeek R1, Grok 3, Mistral Large and Llama 3.1 405B. In a category where the "best model" title changes every few months, buying access to the roster rather than to one vendor is the argument that keeps this extension on my toolbar. The flip side is that you inherit the provider's deprecations and outages with no ability to pin versions.

The pricing shape is freemium, and the paid tiers are Pro for individuals and Teams for two or more seats with an admin dashboard and a larger context window on all models. At the time of writing, the company's own pricing page bills Pro at $228 a year, which works out to $19 a month on annual billing, while monthly billing is quoted at $29 by third-party coverage of the extension. Teams starts at two seats. The free tier is genuinely usable for light work; it is the premium-model rate limit, not the feature list, that separates the tiers.

My testing notes

The observation that shaped my view most is the query economy. The free plan credits 102 queries per day, which sounds generous until you read the exchange rate the company documents itself: a GPT-4/Gemini 1.5/Mistral Large-class prompt consumes 30 queries, while a GPT-3.5/Gemini Flash/Claude Haiku-class prompt consumes one. In my testing that meant the free tier gave me effectively three premium-model questions per session before it pushed me onto the cheaper engines. Three good questions is fine for a commute; it is not a working research day. The Pro tier is where the product becomes usable as a daily driver, and to its credit the company is candid that Pro is "at least 30x" the free usage subject to fair-use limits, rather than pretending it is infinite.

Second: the in-page integration is genuinely well done. I ran the sidebar over Gmail threads, a long PDF and a YouTube talk, and the context capture was the part that kept me from opening ChatGPT — the extension reads the page you are on, so "summarise this and draft a reply to point three" works in one step instead of copy-paste gymnastics. The YouTube timestamp key-point extraction is a small feature I used more than expected. Third: the summarisation quality held up where I most expected it to fail. On a 45-minute conference talk, the YouTube digest produced timestamped key points I could actually navigate by, and on a dense PDF contract the follow-up questions correctly distinguished the termination clause from the payment schedule rather than smearing both into one paragraph. That is the difference between a transcript dump and a usable reading assistant. Fourth: Projects behaved like a lightweight RAG workspace rather than a gimmick — I pointed it at a folder of documents and repeated questions got answers citing that material rather than generic model knowledge. The query cost of premium models applies there too, so heavy Projects use burns budget faster than the raw prompt count implies. I also kept an eye on latency: side-panel answers on the cheaper engines came back fast enough to use inline while reading, while premium-model responses with live web fusion took noticeably longer — a fair trade for cited, current answers, but worth knowing if you want a snappy rewrite tool above all. Finally, a transparency point in Merlin's favour: the Chrome listing discloses plainly that the extension may earn affiliate commissions from sites visited and purchases made, routed as donations to a selected charity. I would rather that be in the listing than discovered later, though it does mean the product monetises your browsing in addition to your subscription.

Where it falls short

Merlin does not give you direct control over the underlying models — no API keys, no provider-level settings — so you are buying access through Foyer's pooled API purchasing (their pricing FAQ says as much). If a model behaves differently than you expect, there is little to adjust. The free tier's premium-model allowance is effectively a demo, and the Pro tier's fair-use ceiling is published in the terms but not quantified, which makes the "$130 worth of value for $19" claim on the homepage vendor arithmetic I cannot verify. Claims of GDPR, ISO 27001 and SOC 2 compliance are asserted on the company site; I could not audit them, so treat them as vendor claims. Feature breadth is also a cost: 26-in-1 is the company's count and "70+ AI tools" appears elsewhere, and in my experience surfaces of that width mean some tools (the AI detector, the image editor) feel shallower than the chat and summarisation core. The AppSumo lifetime-deal history has left two tiers of legacy users with different entitlements, which occasionally makes support answers inconsistent. And because everything routes through Foyer's servers (user data is stored in the United States per their FAQ), regulated workloads need their own due diligence.

Verdict

Merlin is a competent consolidation play: one sidebar, many models, real page-level context capture, and enough depth in Projects and Crafts to justify existence beyond a model-switcher. The query economics need reading twice before you subscribe — the free tier is a taste, and the paid tier is generous but bounded. I rate it 3.8: strong core functionality depth for the browser-extension category carries most of that score, with good multi-platform usability and transparent usage accounting holding the value score up, and shortfalls in support consistency and unverifiable security claims pulling the rest down. If you want a single model with deep specialist tuning, go direct to that provider; if you want one browser surface that reaches most of the market's frontier models and can read the page you are on, Merlin does the job at a price that undercuts two separate subscriptions.

Sources

• https://merlin.foyer.work/
• https://chromewebstore.google.com/detail/merlin-ask-ai-to-research/camppjleccjaphfdbohjdohecfnoikec
• https://www.getmerlin.in/
• https://www.getmerlin.in/pricing
• https://www.getmerlin.in/terms-and-conditions
• https://www.androidpolice.com/ai-powered-chrome-extensions/
• https://appsumo.com/products/merlin/
• https://domore.ai/tools/merlin.foyer.work
• https://linkedin.com/company/merlin-ai-by-foyer
• https://feedback.getmerlin.in/
• https://play.google.com/store/apps/details?id=com.foyer.merlin
• https://apps.apple.com/in/app/merlin-ai-ai-chat-assistant/id6453692447