ASTRONOM Whitepaper

Abstract

The next generation of AI agents — those that act autonomously on the web on behalf of users — is bottlenecked not by model architecture or compute, but by access to diverse, real, consented human web behavior data. Closed labs accumulate proprietary tool-use traces internally; public datasets are stale and narrow. ASTRONOM is a decentralized network that captures, augments, and trains on consented browser trajectories, with on-chain provenance and contributor rewards. This document outlines the architecture, economics, and roadmap of the network.

The agentic data crisis

2025–2026 saw the emergence of capable browser-using agents (Claude Computer Use, Operator, in-house OpenAI/Anthropic systems). Each requires massive volumes of demonstration data from real users navigating real websites with real intent. The supply chain for that data is currently:

• Closed and centralized — top labs collect internally, sharing nothing.
• Stale — public datasets like Mind2Web were collected in 2023; the modern web has moved on.
• Narrow — single-task or single-domain, not generalizable.
• Unethical — scrapers don't compensate the humans whose work generated the data.

This is the same data crisis Physical AI faced in 2024 — and the solution is the same: simulation-first capture, distributed augmentation, on-chain provenance.

Browser-native capture pipeline

Each contributor installs the ASTRONOM Capture Client. During opt-in capture sessions, the client records:

• DOM snapshots at action timestamps
• User actions (clicks, scrolls, key inputs, navigation)
• Screen coordinates and viewport state
• Task intent (user-provided, optional)

The trajectory is locally redacted — passwords, payment details, personal IDs are stripped before upload. The contributor reviews and signs the submission. Validators verify quality. Synthesis engines augment the trajectory across thousands of domain-randomized variants. Final outputs feed Vision-Language-Action policy training.

Three layers, one loop

┌─────────────────────────────────────────────────────────────┐
│  Capture Client  →  Validator  →  Synthesis  →  Training    │
│   (browser)         (stake)       (GPU)          (VLA)      │
│       ↑                                            ↓        │
│       └──────────────  Token Rewards  ─────────────┘        │
└─────────────────────────────────────────────────────────────┘
                            ↓
              Every trajectory: data_id on-chain
                            ↓
                  Verifiable. Tradable. Owned.

$ASTRO token

The $ASTRO token coordinates the four-sided market: contributors, validators, compute, consumers. Detailed tokenomics, supply schedule, and incentive curves will be published in v1.0 of this whitepaper before mainnet launch.

Roadmap

• Q1 2026 — Project founded. Initial team formation, core protocol architecture finalized.
• April 2026 — Public site launch. Early Access dashboard opens for community signups.
• May 14, 2026 — 7 PM UTC — $ASTRO TGE on Solana. Mainnet activation. Capture Client alpha for design partners.
• Q3 2026 — Public Capture Client beta. First technical report published. Validator network expansion.
• Q4 2026 — Full validator decentralization. Tokenomics v1.0 finalized.