What is an Agentic Browser?
A new category of software where an AI agent does the browsing for you — not by automating clicks, but by speaking directly to services through a structured protocol. This is agent-driven browsing.
The traditional browser is dead weight for AI
The web browser was designed in 1991 for humans reading hyperlinked documents. It renders HTML, runs JavaScript, and paints pixels. That was the right design for 1994. It is the wrong design for 2026.
When an AI agent tries to use a traditional browser, it's doing one of two things: either scraping the rendered HTML (fragile, breaks constantly, produces noise) or using a browser automation framework like Playwright (slow, resource-intensive, brittle). Neither approach is sustainable at scale.
The fundamental mismatch is this: agents don't need pixels. They need structured data. A traditional browser gives them a wall of HTML when all they need is a JSON object.
How an agentic browser is different
An AI agent browser doesn't render HTML at all. It communicates with services through a protocol that returns structured data — typed, versioned, and ready for agents to act on. The result is an interface where you talk, and the agent handles everything else.
Traditional browser
- You navigate to each service manually
- You fill out forms, click buttons
- Data is locked inside rendered HTML
- AI must scrape or automate clicks
- Dozens of tabs, zero memory
Agentic browser (Astral)
- You tell the agent what you want
- Agent fetches, filters, and acts across services
- Data arrives as typed JSON via UADP
- AI reads structured objects, not HTML noise
- One interface for every service you use
How Astral works
Astral's architecture has four layers that work together to make agent-driven browsing possible:
You speak or type
"Show me my last three bank transactions over $200" or "Reply to all unread emails from my team." Natural language, not URLs.
The agent plans
Astral's agent runtime (Claude claude-sonnet-4-6 via Anthropic API) classifies your intent, determines which services to query, and plans the sequence of UADP calls.
UADP fetches structured data
The UADP client issues signed requests using your passkey token. Services return typed JSON — not HTML. The agent reads schema types like uadp:transaction or uadp:email directly.
The UI renders from config
ComponentConfig — serializable JSON you can customize by voice — tells the Solid.js render engine how to display the results. No hardcoded layouts, no per-site CSS.
Why UADP matters
UADP (Universal Agentic Data Protocol) is the contract between services and agents. Without a shared protocol, every agentic browser would implement its own scraping adapter for every website — a maintenance nightmare that makes agent-driven browsing fragile by definition.
UADP defines 52 canonical schema types across 10 service categories: social posts, financial transactions, articles, calendar events, messages, products, and more. When a service speaks UADP, an agent can query it reliably, compare data across services, and take actions — all without reading a single line of HTML.
Services that don't yet support UADP are handled by scraping adapters that normalize HTML into UADP schema types. The agent always sees structured data. The protocol is open — explore it on GitHub.
Agentic browser vs AI chatbot
This distinction matters. An AI chatbot like ChatGPT or Claude.ai generates answers from its training data. It doesn't connect to your live bank account, your actual inbox, or your real calendar. It can't take actions in the world.
An AI-first browser like Astral is different. It authenticates to real services using your passkeys, fetches live data, and executes actions — send that email, pay that bill, place that order — on your behalf. The LLM is the reasoning layer, but the browser is the execution layer. Both are required.
The simplest test: can it check your actual bank balance right now? An agentic browser can. A chatbot cannot.
Frequently asked questions
What is an agentic browser?
An agentic browser replaces the traditional point-and-click interface with an AI agent that acts on your behalf. Instead of you navigating websites, the agent fetches data, fills forms, and orchestrates services — all through a conversational interface. Astral is the first browser built around this model.
How is an agentic browser different from a regular AI chatbot?
An AI chatbot answers questions using its training data. An agentic browser actually connects to live services — your email, bank, calendar, shopping history — using signed passkey tokens and the UADP protocol. It takes real actions, not just provides information.
Is an agentic browser safe?
Astral's security model is device-first: private keys never leave your hardware. Authentication uses WebAuthn passkeys backed by your device's Secure Enclave or TPM. The agent only acts on what you explicitly authorize, and all network calls are signed.
Does Astral replace my current browser?
Astral is a complement and eventual replacement for routine browsing tasks. For reading structured data from services you've connected — email, banking, social feeds — Astral is faster and more capable than any traditional browser.
What is UADP and why does an agentic browser need it?
UADP (Universal Agentic Data Protocol) is the structured JSON protocol that lets services expose data in a format agents can read directly. Without a protocol like UADP, agents must scrape HTML — fragile, slow, and easily broken. UADP gives agents a reliable, type-safe interface.
When will Astral be available?
Astral is currently pre-alpha. Architecture and core protocol work is in progress. Desktop builds for macOS, Windows and Linux are available from the downloads section.
Keep exploring
See agent-driven browsing in action
Download Astral and ask its agent to fetch data, compare services, or explain how UADP works.
Download Astral