AI browsers have already had their moment. Perplexity’s Comet kicked most of it off, then OpenAI shipped Atlas, and by early this year Google had baked Gemini into a persistent sidebar in Chrome with an agentic auto-browse mode. So even Chrome kind of counts now, which says something about how quickly this category filled up. In any tech or AI boom, you’re always going to end up sifting through a lot of nonsense to find the handful that are actually worth paying attention to.
Sigma is one of the ones worth paying attention to, I think. It’s a privacy-forward Chromium browser that leads with a local LLM called Eclipse, which handles the chat side of things without any of your prompts leaving your machine. But it’s not purist about it either; you can plug in cloud models for the heavier work, so it’s really more of a hybrid than a strictly local-only setup. That flexibility is a big part of what makes it interesting…
Want to stay in the loop with the latest in AI? The XDA AI Insider newsletter drops weekly with deep dives, tool recommendations, and hands-on coverage you won’t find anywhere else on the site. Subscribe by modifying your newsletter preferences!
What is Sigma?
A privacy-forward browser with an AI twist
Sigma is a Chromium browser, so there’s not much to get used to if you’re coming from a browser like Chrome or Brave because extensions still work the same and sites still render properly. What’s different is what happens when you open a new tab. Instead of just a search bar, you get three options at the top of the page – Search, AI Chat, and Agent – and you pick how you want to start whatever you’re doing. Search is a regular browser search, and Sigma supports Google, Bing, Brave, Ecosia, Baidu, and DuckDuckGo. AI Chat drops you into a conversation with an LLM, and Agent is for handing off multi-step tasks that need actual browser actions.
There’s also a Private mode toggle, which is a whole separate environment rather than just an incognito window. The theme changes, the default search engine switches to DuckDuckGo, and the AI paths get locked down to local-only. You get the passphrase flow when you first open Sigma too, and it’s a proper 12-word phrase that encrypts your Spaces (Sigma’s take on profiles). You can save it as a plain text file, which I did, because losing the phrase means losing the data. Bookmark import is optional, but you can also start clean as a guest.
The privacy stack is decent – built-in ad blocker, tracker protection, phishing detection, TLS 1.3 and DNS-over-HTTPS on by default, anti-fingerprinting, end-to-end encrypted AI chats, and a built-in VPN. The codebase is also open-source. The whole thing feels less like a browser with AI features added on, and more like a browser where AI is one of three ways you interact with the web, which is basically its whole philosophy in one line.
The AI chat is the strongest part of the browser
A full workspace with a local LLM at the wheel
The AI Chat mode is where Sigma actually feels different. It’s laid out like Claude or ChatGPT if you’ve spent any time with either – sidebar on the left with New Chat, History, Library, and Projects, and the conversation lives in the middle. It’s not called a workspace but it functions like one, and if you’re used to jumping between browser tabs and separate AI apps, having the chat live inside the browser cuts out those extra steps. Skills are accessible via a slash, helpers via an at-symbol, and there’s a mode picker with buttons like Learn, Summarize, Write, Deep Research, Analyze, and Create Image.
The default engine here is Eclipse, which is Sigma’s own local LLM that ships with the browser and runs on your hardware via llama.cpp. That means no account, no API key, it works offline, and none of your prompts leave the machine. You can also swap in other local engines from the settings if you want something different or a different profile – Gemma 4 E4B is actually the recommended one, but Qwen 3.5, Nemotron 3, GLM 4.7 and the larger Gemma variants are all there too. Each also comes with an optional uncensored toggle that pulls an abliterated build from HuggingFace instead of the default mirror.
One of the features that was a cool surprise was the Canvas Design, which is one of the skills you can trigger in chat. I gave it a prompt for a plant shop onboarding mockup and it planned it out, saved a markdown file describing the design philosophy, then generated a static PNG that honestly looked usable. It’s comparable to Gemini Canvas or Claude Artifacts, just living inside the chat instead of a separate app. The naming and behavior actually looks a lot like Anthropic’s Canvas Design skill.
And there’s a Brave comparison worth making. Leo AI supports local models too, but you have to install Ollama yourself and bring your own weights, which isn’t really doable for anyone who isn’t already in the local-LLM crowd. Sigma bundles the whole thing so it works right from the install.
It has Agents
With a brain of your choice
Sigma’s Agent mode is the third of the three windows on a new tab, and it’s for the tasks that need a browser to do things like navigate, click, fill forms, pull data, cross-reference tabs, and so on. Under it are two open-source agent frameworks, OpenClaw and Hermes, both of which Sigma bundled rather than built. OpenClaw is the default and the one most people should stick with. Hermes is leaner and better for repeat tasks that improve over time.
The framework runs in the browser, but the brain doesn’t. You either plug in an API key from a supported provider – Anthropic, OpenAI, Grok, Gemini, DeepSeek, OpenRouter, or an OpenAI-compatible custom endpoint like LM Studio – or you point it at a local model in Private mode. Eclipse alone won’t run agent tasks, which is worth knowing upfront.
The Agent home page ships with example prompts such as Find cheapest flight, Track a product, Stalk a brand, Scan my inbox, Morning briefing, and Research a place. These are suggestions rather than pre-tuned agents, so you can also just type any browser-based task you’d normally do manually and let it run.
Sigma is coming for the big AI browsers
Sigma is coming for Chrome and Perplexity, and it’s got something neither offers – local AI workflows with your brain of choice, and it’s a properly functional browser too. Perplexity has never really stuck with me anyway, and I think the user experience gap between Sigma and my usual go-tos (Brave and Chrome) is small enough that it’s worth keeping.
