
Sigma Browser
Sigma Browser is a free Chromium-based browser with AI in its core: an agent that completes web tasks, a chat with deep research and image generation, and Eclipse, a local model running offline, plus tracker and ad blocking.
What is Sigma Browser?
Sigma Browser is a desktop and mobile web browser built on the Chromium engine, with artificial intelligence placed at the core of the product rather than bolted on as an extension. From the start page you switch between three modes: Search, AI Chat and Agent.
The AI Agent is the headline feature. Presented as an agentic browser, it works from a goal you describe in plain language, opening pages, clicking, typing into fields, submitting forms, reviewing files and reporting back when the task is done. Every step stays visible and you can take over at any moment. Two agents are selectable in the settings, OpenClaw and Hermes. In standard browsing you connect your own model API key; in Private Mode the agent can run on a local model instead.
AI Chat gathers six functions into one panel: image generation, Deep Research, Text Writer, Web Search, Chat With Page, and TG Search, which searches Telegram posts and summarises the opinions it finds. Deep Research queues sources, clusters the information and returns an organised answer, positioned by the publisher for market analysis and academic review.
Eclipse is the third pillar: a local large language model built into the browser, free, running entirely on the machine and fully offline. Models can be downloaded inside the browser or connected through your own API key, and Eclipse is also published separately on GitHub.
Around the AI sits a privacy and security layer: tracker and fingerprinting-script blocking, a built-in ad blocker, encrypted private profiles, a mnemonic identity backed by a twelve-word phrase, TLS 1.3, HTTPS and DNS-over-HTTPS, real-time phishing detection, end-to-end encryption and a stated defence against prompt injection. The Chromium base keeps ordinary extensions working, and a Web3 layer connects wallets and decentralised applications.
The publisher claims Speedometer 3.0 results of 1.32x faster than ChatGPT Atlas, 1.30x than Safari, 1.17x than Brave and 1.13x than Chrome, and the lowest memory use of six browsers with ten tabs open. Builds exist for macOS, Windows, iOS and Android, with Linux announced. Sigma Browser is published by SigmaBrowser OÜ, an Estonian company registered in Tallinn on 29 May 2024.
What it does
- Hand a multi-step web task to an AI agent that navigates, clicks, types and submits forms on your behalf
- Ask questions about the page you are reading and get an instant summary of it
- Run deep research across several sources and get back a structured, organised answer
- Run a local language model offline, so prompts never leave the device
- Generate images from a text prompt, then restyle or refine them
- Write, rewrite and improve text, and generate or fix code
- Translate highlighted text in place, and block ads and trackers while browsing
When to use Sigma Browser / When not to
A quick filter to help you decide if Sigma Browser is the right fit.
When to use Sigma Browser
- Researchers, analysts and students who compare many sources at once, since Deep Research queues sources, clusters the information and returns a structured answer
- Knowledge workers who want repetitive web chores handled for them: filling forms, booking tickets, tidying an inbox, comparing prices across sites
- Privacy-conscious users who need to think through confidential material without sending it to a cloud provider, thanks to the on-device Eclipse model
- People paying for several AI subscriptions at once who would rather have chat, deep research, image generation and an agent inside one free browser
- macOS and Windows users comfortable plugging their own model API key into an agent, and attached to the Chromium extensions they already use
When not to use Sigma Browser
- Linux users: the .deb and .rpm builds are still listed as coming soon on the download page
- Teams with procurement or compliance requirements, since there is no DPA, no subprocessor list, no documented hosting location and no published position on model training
- Developers looking for an API to build on: Sigma consumes third-party APIs through your key but exposes none of its own and publishes no developer documentation
- Anyone under 13, who is excluded both by the terms of use and by the privacy policy
- Organisations that need meaningful contractual recourse, given liability capped at USD 50 and a perpetual, sub-licensable licence taken on submitted content
How to use Sigma Browser
A typical end-to-end flow, from setup to results.
- Download the installer from the download page: a .dmg for Apple Silicon Macs, a .exe for Windows x64, the App Store for iOS or Google Play for Android
- Install and open the browser; no account is required to start browsing, and the publisher puts setup at under a minute
- Create an account or sign in with Google, Apple, or a twelve-word recovery phrase if you want the mnemonic profile
- Pick a mode from the start page: Search for ordinary browsing, AI Chat for questions, or Agent for tasks
- Choose the engine behind the AI: a cloud model, or a local model downloaded from the settings for fully on-device processing
- To use the agent, open the settings, select OpenClaw or Hermes, then connect an API key or select a local model in Private Mode
- For long research, open the AI Chat panel and select Deep Research before typing your question
- On any article, page or PDF, ask Chat With Page for a summary or a specific answer
- For visuals, open Sigma Chat, choose Create Image, write a prompt, then restyle or resize the result
- Toggle Private Mode from the toolbar when you want tracker blocking and local processing, and highlight any sentence to translate it in place
Pros & Cons
Pros
- The whole AI stack, agent included, is usable at no cost: the free tier meters usage rather than locking features away
- Eclipse runs on the device and offline, so sensitive prompts need never reach a server
- Performance claims come with their protocol, ten Speedometer 3.0 runs averaged on a MacBook Pro M4, and the lowest RAM of six browsers at ten open tabs
- The Chromium base keeps existing extensions, bookmarks and password managers working as before
- You choose the engine: a cloud model, a downloaded local one, or your own API key
- Prompt injection defence is addressed explicitly, which few agentic browsers advertise
- The publisher is established in the European Union under Estonian law, and the browser code is stated to be public
Cons
- The legal documentation is unusually thin for a product sold on privacy: a privacy policy of roughly six thousand characters, undated, with no legal basis and no retention period
- No DPA, no subprocessor list, no named data protection officer, and no hosting country or region disclosed anywhere on the site
- Model training is never addressed: the word does not appear once across the whole site
- A single generic support address is the only contact channel; there is no contact page and no about page
- The terms take a perpetual, sub-licensable licence on submitted content and cap the publisher's total liability at USD 50
- The site contradicts itself: Android is marked coming soon while a live Google Play listing exists, and three different version numbers circulate across the pages
- No Linux build, no free trial of the paid tiers, and in standard browsing the agent needs an API key you supply yourself
Pricing & Plans
Sigma Browser is free to download and use. A permanent free plan is offered, which includes the local Eclipse model, Chat With Page and metered access to the AI Agent, Deep Research, image generation, TG Search and the Text Writer. The lowest paid entry point is the Plus plan at USD 10.00 per month on plain monthly billing, or USD 8.00 per month under an annual commitment. The Pro plan is USD 75.00 per month monthly, or USD 60.00 per month annually. No free trial of the paid plans is offered.
- chat on all platforms
- analysis of text
- images and files
- the local Eclipse LLM
- Chat With Page
- Animate Images
- and limited access to the AI Agent
- image generation
- unlimited prompts and Text Writer
- 50x more file uploads
- 10x more web searches
- 20x more image generations
- 12x more deep researches
- 25x more TG searches
- five additional AI Agent runs
- 5x more Animate Images
- unlimited file uploads
- web search
- image generation
- Deep Research
- TG Search and AI Agent
- 25x more Animate Images
- early access to new features
Data, GDPR & hosting
A consolidated view of how Sigma Browser handles your data.
GDPR overview
The homepage carries a one-line claim, that Sigma complies with GDPR and CCPA, and the publisher is established in Estonia, so the Regulation applies by place of establishment and no Article 27 representative is needed or named. The privacy policy lists rights to know, access, correct, transfer, restrict and delete, extends them to the whole customer base rather than EU residents alone, promises no discriminatory treatment for exercising them, and allows consent to be withdrawn at any time. Grounds for refusing a request are spelled out: legal retention duties, anti-fraud and security investigations, the privacy of others, frivolous or impractical demands. Beyond that the documentation is thin. No legal basis is cited, no data protection officer is named, no retention period is given, no international transfer mechanism is described, no subprocessor is listed, and the policy carries no effective date. The only contact is the general support address.
Who owns the data?
The terms are more expansive than the marketing. Section 5 provides that any content or materials you submit through the Services grant Sigma a perpetual, irrevocable, transferable, sub-licensable, worldwide and royalty-free licence to reproduce, use, modify, display, distribute, translate and create derivative works from them, through multiple tiers and across any media. The privacy policy pulls the other way: users are described as essentially anonymous, credentials and task data are said not to be kept after execution, and the local Eclipse model is presented as technically inaccessible to the vendor. Ownership itself is never asserted in either document, so the licence is the operative text.
Reuse rights
Sigma grants you a non-transferable, non-sublicensable, non-assignable and non-exclusive licence to use the Services for personal or internal business purposes. Within that scope nothing in the terms restricts what you do with the answers, summaries, images or code the tool produces for you: no attribution and no prior permission are required. The restrictions bear on the product itself. You may not sublicense, sell, rent, lease, transfer, reproduce, distribute, host or otherwise commercially exploit the Services, embed Sigma into a service you provide to a third party without written consent, reverse-engineer or decompile it, or reuse Sigma's own content, including its images and text, in another project. On the vendor's side, cookies serve session management and preferences, while third-party cookies collect visitor statistics and measure marketing campaigns and can be refused in browser settings.
Data retention & training
Hosting summary
No hosting country, region or provider is named anywhere on the site. The only infrastructure statement is a negative one, made about the local mode: no servers, no storage, no cost. A cloud side clearly exists, since the AI chat is reachable as a web application and the agent runs through an API key in standard browsing, but where that side is hosted is never documented, and no subprocessor is listed. The publisher is established in Estonia and the terms place the agreement under Estonian law, with arbitration in Tallinn, so the governing jurisdiction is clear even though the data location is not. The site itself is served behind Cloudflare on an anycast address, which describes the delivery network rather than where user data resides and should not be read as a hosting disclosure. Anyone with a data residency requirement will have to ask the publisher directly, as the published documentation does not answer the question.
Things to keep in mind
Risks and trade-offs to weigh before adopting Sigma Browser.
- Submitted content carries a perpetual, irrevocable, transferable and sub-licensable worldwide licence under section 5 of the terms, which is far broader than the privacy messaging suggests
- The publisher's total liability is capped at USD 50, disputes go to binding arbitration in Tallinn under Estonian law, and class actions and jury trials are waived
- Sigma may assign or subcontract its rights and obligations without notice or consent, and may suspend, modify or discontinue any part of the service at its sole discretion
- Delegating web tasks to an agent that logs in, fills forms and submits them means trusting an automated process with credentials and irreversible actions, so supervising each run matters
- The Uncensored AI mode explicitly advertises fewer restrictions, which shifts the burden of judgement entirely onto the user on sensitive subjects
- The Web3 features and the twelve-word mnemonic profile put recovery squarely on the user: losing the phrase means losing the profile, and no one can restore it
- Leaning on instant summaries and a browsing agent for reading and routine decisions can quietly erode the habit of reading sources in full and verifying them, especially since the benchmarks and privacy claims here are self-published and not independently audited
Setup & Integrations
Technical difficulty
Low for ordinary use. Installation is a standard desktop download, a .dmg for Apple Silicon or a .exe for Windows, with the mobile builds coming from the app stores. No account is needed to start browsing and the publisher puts setup at under a minute; signing in uses Google, Apple or a twelve-word recovery phrase. Two friction points remain. Running the agent in standard browsing requires connecting your own model API key, which presupposes an account with a model provider, and using a local model means downloading one from the settings first.
Deployment
Apps stores
Integrations
Supported languages
Behind Sigma Browser
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Sigma Browser.
Frequently asked questions
What is Sigma Browser?
Is Sigma Browser free?
Which platforms does it run on?
What does the AI agent actually do, and do I need an API key?
What is Eclipse and does it really work offline?
Is the tool GDPR compliant, and what happens to my data?
Does Sigma train AI models on my data?
How do I contact support, and who publishes Sigma Browser?
Should you pick Sigma Browser?
Sigma Browser makes an unusually clear promise for this category: an agent that acts on web pages, a chat with deep research and image generation, and a language model that runs on your own machine, all inside a Chromium browser that costs nothing to start using. The performance claims are published with their protocol rather than merely asserted, which is more than most competitors offer, and the free plan meters usage instead of hiding features, so the tool can be judged on real work rather than on a demo.
The reservations are just as clear, and they sit precisely where the product places its strongest claim. For a browser sold on privacy, the written record is remarkably thin: a short, undated privacy policy with no legal basis, no retention period, no data protection officer, no subprocessor list and no disclosed hosting location, and not a single line anywhere about whether models are trained on user data. The terms of use, meanwhile, take a perpetual and sub-licensable licence on submitted content and cap liability at fifty dollars. The publisher is very young and very small, registered in Tallinn in May 2024 with a share capital of one euro, and the site contradicts itself on Android availability and on its own version numbers.
The sensible reading is that Sigma Browser is worth installing and testing, because the core is genuinely free and the local model answers a real need for anyone who thinks aloud about confidential matters. It is not, on the published evidence, ready for a regulated professional context: the gap between what the marketing claims about privacy and what the legal pages actually document is the single thing a prospective user should weigh before committing anything sensitive to it.
- Choosing a selection results in a full page refresh.
- Opens in a new window.