Skip to content

Latest commit

 

History

History
424 lines (351 loc) · 22.3 KB

File metadata and controls

424 lines (351 loc) · 22.3 KB

Aeon Browser — Services & Systems Reference

DelgadoLogic | Confidential | March 2026

Quick-reference for every named system, service, protocol, and component in the Aeon ecosystem. Status key: 🔴 Planned | 🟡 In Progress | 🟢 Built/Stable


CORE ENGINE

aeon_engine.dll

What it is: Our Chromium fork compiled as a DLL. The rendering engine that powers the entire browser. Chromium provides Blink (HTML/CSS rendering), V8 (JavaScript), and WebGL. We strip Google's telemetry and replace the network stack with our own Rust router. Every website on Earth renders correctly because it is Blink under the hood. Status: 🟢 Built (v20 — 269.5 MB, 57,303 targets, 0 failures) — April 12, 2026 Produced by: ninja -C out\AeonRelease chrome on GCP VM n2-standard-32 Output file: chrome.dll → renamed to aeon_engine.dll (269.5 MB) Location: C:\AeonEngine\engine_package\ (local) + gs://aeon-sovereign-artifacts/ (cloud backup) Google code removed: Safe Browsing, Sync, Crashpad, UMA metrics, Rappor, Google APIs, sign-in, optimization guide cloud calls Cost to run: $0 (compiled locally, redistributed with installer) Dependencies: VS BuildTools 2022, depot_tools, ATL stub (C:\chromium\atl_stub\)


aeon_ai.dll (Year 2 goal)

What it is: Proprietary C++ AI inference engine embedded inside the browser process. No external server, no Python, no Ollama dependency required in production. Built by studying and then replacing llama.cpp + ggml. Runs quantized 3B parameter models in ~2GB RAM. Status: 🔴 Planned (currently using ollama as prototype) Primary model: phi-3-mini (4-bit quantized, ~2.5GB) for chat + reasoning Background model: TinyLlama 1.1B or SmolLM 360M for tab intelligence Cost to run: $0 (all inference is local, user's hardware) Why it matters: Our moat. No other browser has a fully integrated, proprietary, privacy-first AI engine.


NETWORK LAYER

Rust Protocol Router

What it is: Our custom Rust binary that intercepts all browser network requests before they touch Chrome's native network stack. The traffic cop of the browser. Handles:

  • DoH (DNS-over-HTTPS) with 5 fallback resolvers: Cloudflare, NextDNS, AdGuard, our own, quad9
  • .onion URL routing (Tor integration — works in any tab, not a separate window)
  • ipfs:// and magnet: URLs natively
  • CircumventionEngine integration (GoodbyeDPI, zapret, DPI bypass) Status: 🟡 Core binary built; routing logic designed, integration with aeon_engine.dll shell pending Cost to run: $0 (compiled into Aeon.exe, runs locally)

CircumventionEngine

What it is: DPI (Deep Packet Inspection) bypass system that allows Aeon to load sites blocked by ISPs and governments. Combines GoodbyeDPI, zapret, and our own signature-based router. Methods: TCP fragmentation, fake SNI, HTTP header manipulation, QUIC-based bypass (hysteria2 transport) Static signatures: Maintained by us, bundled with each release Dynamic signatures: Supplied by AeonHive (community-discovered bypasses propagate to all users within minutes) Status: 🟢 Built (static signatures), 🔴 AeonHive dynamic layer planned Cost to run: $0 (runs locally). Relay bandwidth for Pro tier relay nodes: included in server costs.


TierDispatcher

What it is: Routes each network request through the appropriate path based on the user's subscription tier and network conditions.

Free:       DoH + direct connection + basic circumvention
Pro:        DoH + Aeon relay nodes + full circumvention chain + faster routing
Enterprise: Company proxy + DLP rules + audit log + Aeon relay fallback

Authorization: Signed JWT from Aeon Account, verified in Rust router. 30-day offline grace period. Status: 🔴 Planned (Rust router exists; tier logic not yet implemented) Cost to run: Free tier costs nothing. Pro relay nodes: included in server infrastructure cost (~$40–50/month for relay VPS fleet).


P2P COLLECTIVE INTELLIGENCE

AeonHive

What it is: The P2P gossip network that makes every Aeon feature smarter as the user base grows. No central server. No user identification. Fully opt-in. Uses GossipSub protocol (same as Ethereum) over a DHT (Distributed Hash Table). Every AeonHive node connects to 8–12 mesh peers. Messages hop exponentially — 10K-user network reached in ~4 hops (~200ms). 1M-user network in ~7 hops. What it shares: Encrypted threat hashes, anonymized circumvention signatures, federated model gradients (differential privacy), anonymized agent task patterns, autocomplete aggregates (100+ user threshold) What it NEVER shares: URLs, user identity, browsing history, passwords, personal data Opt-in default: OFF. User explicitly enables. Status: 🔴 Planned. Research repos cloned to research/p2p/ Foundation library: iroh (Rust P2P, Apache-2.0) + go-libp2p (Go DHT, MIT) Gossip protocol: go-libp2p-pubsub (GossipSub, MIT) Cost to run: Nearly $0 at scale — users carry the bandwidth and storage load. Only Anchor Nodes cost money.


Anchor Nodes

What they are: 5–10 small VPS servers run by DelgadoLogic that serve as DHT bootstrap points. They are phonebook operators — they tell peers who else is online, then get out of the way. They hold zero user data, see zero user traffic, cannot read any gossip messages (all end-to-end encrypted). Phase 1 (0–10K users): Required for peer discovery Phase 2 (10K–100K users): Optional fallback, network self-discovers Phase 3 (100K+ users): Shut off — network doesn't notice Cost: $6/month per node × 7 nodes = **$42/month total** Recommended provider: Hetzner (EU) or DigitalOcean (US) — $6/month for CX11 instances Legal exposure: Zero. They're phonebook operators. Snowflake (Tor) has run this model for years with no legal action.


AeonHive Gossip Message Format

{
  "type":       "threat_sig | circ_sig | model_grad | agent_pattern | autocomplete",
  "hash":       "sha256:a3f9c2...",
  "geo_region": "US-East",
  "weight":     0.4,
  "ttl":        48,
  "pow":        "0000ab3f...",
  "sig":        "Ed25519:..."
}
  • weight = reporter's trust score (0.0 new node → 1.0 after 1 year of honest participation)
  • pow = HashCash proof-of-work — stops spam flooding
  • sig = Ed25519 — proves message authenticity, prevents forgery
  • ttl = 48 hours — stale data auto-purges

Sybil Protection (4-Layer Defense)

Layer Mechanism
1 — Temporal Weight New nodes: weight 0.0. Takes 1 year of honest participation to reach 1.0
2 — Geographic Quorum Threat must be confirmed from 3+ distinct regions independently
3 — STAR Threshold Individual reports hidden until consensus threshold (50+ reports) via Cloudflare's sta-rs
4 — Pro Node Anchoring Verified Pro subscribers get 2× vote weight — buying thousands of Pro subs is not economical

Aeon Credits (Internal Economy)

What they are: Non-transferable, non-cash reward tokens earned by contributing to AeonHive. Cannot be sold, transferred, or converted to cash. Not a security. Not crypto.

Action Credits Earned
Relay 1GB of AeonHive gossip 50
Cache 200MB of content for 30 days 10/month
Confirmed threat report (threshold reached) 5
Verified circumvention signature 20
Agent task pattern accepted 3

Redeemable for:

  • +5GB Aeon Sync storage (per 500 Credits)
  • Priority relay routing (faster TierDispatcher)
  • Early access to new AI models
  • Extended offline cache quota Cost to us: Credits are softcurrency. No cash cost. Redeemed benefits are bucketed into existing infrastructure costs.

USER DATA SERVICES

AeonSync

What it is: Cross-device browser data sync using a Firefox Sync-style zero-knowledge architecture. How it works: All data is AES-256 encrypted with the user's local sync key before leaving the device. Our server stores encrypted blobs it cannot read. Even if our servers are breached, attackers get gibberish. Synced data: Bookmarks, settings, AeonVault password list (encrypted), browsing history (last 30 days), open tabs, extension list (not extension data) NOT synced: AI model weights (P2P only for Pro), raw browsing history older than 30 days

Tier differences:

Tier Sync Method Server
Free Encrypted cloud sync, 100MB cap Cloudflare R2
Pro Encrypted cloud + direct P2P device-to-device sync via AeonHive Cloudflare R2 + AeonHive DHT

Pro P2P sync: Devices discover each other via DHT, authenticate with shared device key (derived from master sync key via HKDF), merge data using CRDT (automerge library). No server involved. Works across networks via AeonHive relay.

Storage cost:

  • 1K users × 1MB avg = 1GB = $0.015/month (Cloudflare R2)
  • 100K users × 1MB avg = 100GB = $1.50/month
  • The sync backend is essentially free.

Your sync key: Never leaves your devices. If lost, we cannot recover it. This is intentional. Zero-knowledge. Status: 🔴 Planned (Phase 3)


AeonVault

What it is: Built-in password manager. Zero-knowledge, DPAPI-encrypted locally. Hooks directly into Blink's autofill pipeline (not a floating overlay — works on sites that block standard autofill). Encryption: AES-256 + DPAPI. Master password never stored — only its hash used as key derivation input (Argon2id). Features: Auto-fill, password generator, breach detection, 2FA TOTP support Breach detection: Via AeonHive gossip (peers share domain breach hashes — not your passwords, just the domain name hashes) Biometric unlock: Windows Hello integration Status: 🟡 Architecture designed, not yet implemented Cost to run: $0 (fully local)


Aeon Account

What it is: Optional user account system for Pro subscription management and sync key backup. What it stores: Email (for account recovery), hashed password (Argon2id), encrypted sync key backup (user's password encrypts this — we can't read it), subscription status. What it does NOT store: Browsing data, search history, AI conversations, anything from the browser. Status: 🔴 Planned (Phase 3) Backend: Lightweight stateless API on Cloud Run. ~$5–15/month at early scale.


AI FEATURES

AI Sidebar

What it is: A persistent right-side panel providing chat, summarization, and context-aware AI assistance for any webpage. Invocation: Keyboard shortcut or toolbar button Features: Summarize page, explain selected text, chat with page context, rewrite/simplify content, code explanation Model: phi-3-mini (local, 4-bit quantized) via ollama initially, aeon_ai.dll eventually Architecture: Renderer (sandboxed) → Mojo IPC → aeon_ai utility process → inference → streamed response Status: 🔴 Planned (Phase 2, after engine build completes) Cost to run: $0 (local inference only)


Agent Mode (Auto-Browse)

What it is: Give Aeon a goal in natural language. The agent autonomously navigates, clicks, reads, and returns results without sending your data anywhere. Architecture:

User intent → llama.cpp (local) → task plan
              → Blink DOM bridge (direct DOM access, no scraping hacks)
              → Execute steps → Return result

Advantage over competitors: Direct Blink DOM bridge (we ARE the browser — no computer vision, no extension injection). On-device inference = 150–200ms per step vs 800ms–2s cloud agents. 45 seconds faster on a 30-step task. Memory (2 layers):

  • Personal (local, encrypted, never shared): Your preferences, your task history, AeonVault
  • Collective (AeonHive, anonymized): Successful navigation patterns across all users Status: 🔴 Planned (Phase 3 - P2 for architecture, P3 for full release) Cost to run: $0 (local inference) Research repos: semantic-kernel, autogen, langgraph, agentic-platform-engineering

Smart Address Bar (AeonOmni)

What it is: Replaces Google Suggest with a 3-layer privacy system that's as useful without any privacy compromise.

Layer 1: Local History         — Your URLs, searches, bookmarks. Zero network.
Layer 2: AeonHive Collective   — Top completions from 100+ other Aeon users this week only. Anonymous aggregate.
Layer 3: Local AI Intent       — phi-3-mini classifies intent, enriches result, previews answer. Zero network.

Layer 2 minimum: A completion only surfaces if 100+ independent users matched it this week. Differential privacy noise added to counts. You can never identify an individual. Intent examples: "tesla stock" → mini chart | "pizza near me" → map pin | "cancel amazon" → direct cancellation link Status: 🔴 Planned (Phase 2) Cost to run: $0 (all local + anonymous AeonHive aggregates are already in network)


Privacy Inspector

What it is: Real-time detection and blocking of browser fingerprinting, tracking scripts, and surveillance techniques on every page load. Detected vectors: Canvas, WebGL vendor/renderer, AudioContext, font enumeration, screen API, timing attacks, battery API, network connection API, keyboard layout, … With AeonHive: Collective threat intelligence — "1,247 Aeon users visited this site. New technique detected this week: WebGL vendor string extraction. We blocked it before you loaded." Zero-day detection: 5 users report anomalous behavior → AeonHive flags it globally before most users are affected Status: 🔴 Planned (Phase 2) Cost to run: $0


AeonSleep (Tab Memory Manager)

What it is: Suspends inactive tabs at the C++ renderer level (not just JS freeze). Sleeping tabs use <5MB RAM vs Chrome's 80–300MB.

Tier 3 (budget hardware):   Sleep after 30 seconds inactive
Tier 2 (mid-range):         Sleep after 90 seconds
Tier 1 (high-end):          Sleep after 5 minutes

Restored instantly when you click the tab — 200–400ms cold start from disk cache. With AeonHive: Sleep timing calibrated by collective behavior. "Gmail compose: never sleep — 12% of users lost drafts." Learned from millions of tabs, not hardcoded. Status: 🟡 Tab sleep manager coded. Blink-level suspension pending engine build. Cost to run: $0


TELEMETRY & COMPLIANCE

Zero-Identify Telemetry

What it is: Opt-in only crash reporting. First-run explicit consent. Nothing leaves your machine without permission.

Collected (with consent) NEVER Collected
Stack trace (crash type + version) URL at time of crash
Crash count by version Machine ID or hardware fingerprint
Anonymous feature opt-in rates Any user identifier
Aggregate tab counts (numbers only) Browsing history, search queries

Implementation: Crashpad stripped of PII extraction. Only stack hash + version number sent. With AeonHive enabled: crash data propagates through gossip instead of hitting our server — we never even see it. 50-report threshold before a crash pattern surfaces to engineering. Marketing: "Aeon is the only browser where you can read exactly what leaves your machine — because the code is open and the answer is almost nothing." Status: 🔴 Planned (Phase 2 - strip Crashpad, add opt-in dialog)


Optimization Guide (Hijacked from Google)

What it is: Chrome's components/optimization_guide/ drives ~30 features (tab discard, preloading, omnibox predictions, anti-phishing ML). Most de-Google forks gut it, breaking those features. Aeon keeps the API surface, replaces the data source.

Google: optimization_guide/ → Google CDN (phones home constantly)
Aeon:   optimization_guide/ → Local Aeon Cache (stays on device)
        + AeonHive collective seeds it on day 1 for new installs

Cold start problem: New install has zero optimization data. AeonHive seeds it from peers with similar hardware specs on day 1. Status: 🔴 Planned (Phase 2 — high priority differentiator vs ungoogled-chromium)


INFRASTRUCTURE COST SUMMARY

System Component Monthly Cost
AeonHive 7× Anchor Node VPS (Hetzner CX11) ~$42
AeonSync Cloudflare R2 storage (per GB) ~$0–$2
Aeon Account Cloud Run stateless API ~$5–$15
Aeon Relay (Pro) 3× relay VPS (Hetzner CX21) ~$30
Total (Beta) ~$50–$90/month
Total (1M users) ~$200–$300/month

Why costs barely scale: AeonHive users carry their own bandwidth + storage. Model updates distributed P2P (BitTorrent-style) — not from our CDN. Sync data is tiny and compressed. Anchor nodes cost the same whether you have 100 or 100K users.


LIVE INFRASTRUCTURE SERVICES (April 14, 2026)

Cloud Run Services (aeon-browser-build project)

Service Purpose Region Status
aeon-update-server Sovereign update manifests (/v1/update/{product}/{channel}) us-east1 🟢 Live
aeon-ai-scanner CVE/threat intelligence scanning us-central1 🟢 Live
aeon-dns Sovereign DoH resolver us-central1 🟢 Live
aeon-relay Encrypted WebSocket tunnel relay us-central1 🟢 Live
aeon-evolution-engine Autonomous patch system (dual-region) us-central1, us-east1 🟢 Live

Domain Infrastructure

Domain Backend Status
delgadologic.tech Firebase Hosting (manuel-portfolio-2026) 🟢 HTTP 200
www.delgadologic.tech Firebase Hosting (CNAME) 🟢 HTTP 200
status.delgadologic.tech Firebase Hosting 🟢 HTTP 200
api.delgadologic.tech Firebase Hosting (API portal) 🟢 HTTP 200
docs.delgadologic.tech Firebase Hosting (docs) 🟢 HTTP 200
audit.delgadologic.tech Firebase Hosting (Command Center) 🟢 HTTP 200
aeon.delgadologic.tech Firebase Hosting (aeon-browser-delgado) 🟢 HTTP 200
browseaeon.com Firebase Hosting (consumer brand) 🟢 HTTP 200
aeonbrowse.com Firebase Hosting (secondary alias) 🟢 HTTP 200

GCS Buckets

Bucket Role Access
aeon-sovereign-artifacts Engine builds, signed manifests, private CI artifacts Private
aeon-public-dist Installer downloads, public release binaries Public CDN

Update Server Routes

Route Response
/health Service health check
/v1/update/logicflow/stable LogicFlow v1.0.0 manifest (Ed25519 signed) ✅
/v1/update/aeon/stable Aeon stable channel (no release yet)
/v1/update/aeon/nightly Aeon nightly channel (no release yet)
/v1/manifest/publish CI publish endpoint (Bearer auth + GCP Secret Manager)

REVENUE MODEL

Stream Description Potential
Search Partnership Default search deal (DDG easiest first) $$$$ at scale
Aeon Pro $4.99–$9.99/month — Sync + AI + Premium relay $600K/yr at 10K users
Aeon Shield / Enterprise $15–$50/seat/month — managed browser, zero telemetry, audit logs Island Browser charges ~$20/seat; IBM/Goldman use it
AeonVPN $3–$5/month — premium routing via CircumventionEngine + relay Bundled into Pro
Privacy Ads Opt-in ads, revenue share with user as Aeon Credits Brave model

RESEARCH REPOSITORY INDEX

AI Engines (research/ai-engines/)

Repo License Role
llama.cpp MIT Core inference engine to study/fork
ggml MIT Tensor math backend
ollama MIT Local model server (prototype)
gemma.cpp Apache Single-file C++ model packaging pattern
mistral.rs Apache Rust-native inference (ties to our router)
LocalAI MIT OpenAI-compatible local wrapper
whisper.cpp MIT Voice navigation ("Hey Aeon")
llama-models MIT Meta LLaMA architecture spec
TinyLlama Apache 1.1B background tab model
phi-3-mini MIT 3.8B primary chat model
SmolLM Apache 135M–1.7B background intelligence
MiniCPM Apache 2B multimodal (privacy inspector)
mediapipe Apache Gemini Nano runtime SDK
LiteRT Apache Gemini Nano execution engine
ai-edge-torch Apache Model export to Gemini Nano format
agentic-platform-engineering MIT MCP agent patterns blueprint
autogen MIT Multi-agent orchestration
semantic-kernel MIT C# AI SDK — our stack
langgraph Apache Stateful agent graphs
openai-swarm MIT Agent handoff patterns
crewAI MIT Role-based multi-agent automation
graphrag MIT Knowledge graph + LLM retrieval
Jan AGPL Local AI chat UX reference
chatbot-ui MIT Streaming chat UI patterns

P2P Networks (research/p2p/)

Repo License Role
iroh Apache-2.0 Primary: Rust P2P networking for AeonHive
go-libp2p MIT DHT peer discovery + NAT traversal
go-libp2p-pubsub MIT GossipSub protocol layer
libtorrent BSD-3 AI model P2P distribution
automerge MIT CRDT state sync (conflict-free shared state)
yjs MIT JS CRDT (agent task memory sync)
privacypass MPL-2.0 Anonymous token system (study only)
veilid MPL-2.0 Anonymous P2P framework (study only)
yacy_search_server GPL-2 Proof P2P collective intelligence works (study only)

Federated Learning (research/federated/)

Repo License Role
flower Apache-2.0 Primary: Federated learning framework
PySyft Apache-2.0 Secure aggregation protocols
federated (TF) Apache-2.0 FedAvg reference implementation
secretflow Apache-2.0 Secure multiparty computation
opendp MIT Rust differential privacy library
google/differential-privacy Apache-2.0 C++ DP reference
sta-rs BSD-3 Cloudflare's STAR protocol (AeonHive threshold reporting)

Circumvention (research/circumvention/)

Repo License Role
dnscrypt-proxy ISC DNS encryption
GoodbyeDPI Apache-2.0 DPI bypass (Windows)
zapret MIT DPI bypass (cross-platform)
outline-server Apache-2.0 Shadowsocks relay architecture
snowflake BSD-3 P2P bridge (Pro contribution model)
hysteria2 MIT QUIC-based censorship evasion
psiphon-tunnel-core GPL-3 Study only — production circumvention reference

Last Updated: April 14, 2026 | DelgadoLogic / Aeon Browser