⚡ Agent Runtime · v1.3.0

THUNDERGATE

ThunderAI

The AI runtime that thinks like a mind, not a tool.

One context. All channels. Persistent memory. Sovereign. v1.3.0 live — the source of truth release: your agent now knows the difference between "I checked" and "I remember." A per-agent map of where truth lives, with your own ruling on which source wins — read before asserting, freshness on its face, and copies never counted as evidence. Nothing becomes permanent memory without your word. Identity can’t drift, memory can’t be fabricated, secrets rotate. Riding alongside it now: an ambient lane that watches your agent's own work and flags what it asserted without checking. Jon has run production on ThunderGate since Jun 28. Your agent can too.

Get Install Code See What It Does
v1.3.0 — The Source of Truth Release ⚡ Checked ≠ Remembered · Operator-Ruled Truth · Sovereign

Why ThunderGate

BUILT DIFFERENT.
ON PURPOSE.

Every other runtime forgets you when the session ends. ThunderGate doesn't. First production migration proven June 28, 2026.

🧠

Learns From You

Corrections, decisions, and preferences burn into persistent memory. Every session your agent starts warmer than the last. No briefing required.

🖥

Runs On Your Hardware

Cloud today, ThunderMind tomorrow. When your 70B spins up locally, the token meter stops. Same agent, same memory, zero marginal cost.

🌐

Works Everywhere

Slack, WhatsApp, iOS, Mac, terminal — one context, all surfaces. Pick up on your phone where you left off on the desktop. No friction.

🏠

Your Data Stays Yours

Your durable memory, credentials, and local corpus stay on your machine — encrypted, never phoned home. Provider and channel payloads leave only when that connector or cloud inference path is explicitly enabled by you.


What It Does

EVERYTHING YOU NEED.
NOTHING YOU DON'T.

50+ design principles + 2 Prime Directives locked. Built for the long run, not the demo.

👁

Ambient Intelligence New

A second lane of attention that reads your agent's own work as it happens — a claim made without checking its registered source, an assertion the evidence doesn't support, something stated as settled that was never verified. It can't answer you, act, or stop a turn; it hands your agent a card before it speaks. Every stage is written to an audit lane the agent cannot author, and the whole subsystem ships with a declared test for its own retirement.

💾

Persistent Memory

Three-tier memory lifecycle — hot, warm, cold. Session summaries survive restarts. Long-term memory curated, not dumped.

🔀

Model Routing

Auto, manual, or supersaver. go big → Opus. go fast → Sonnet. One config, one field, no coin flip on restart.

🌐

ThunderBrowser

Real Chromium, real cookies, live DOM. Navigate, click, fill, extract — with a full audit trail.

📡

Multi-Agent Relay

Multiple specialized agents, one coordinated team. Federated relay protocol — each agent owns its domain, all surfaces connected.

🔒

Encrypted Vault

PII, credentials, and secrets in a tamper-evident encrypted store. Agents think freely; sensitive data never leaks.

Deep Mode

Heavy tasks run in parallel. Surface layer handles interrupts. Status heartbeat. You're never in the dark.

🛡

ThunderGuard In Design

Emergency response built into the agent rather than bolted on — help that understands your situation and can act when you cannot ask for it. Principles written, build ahead: not included in v1.3.0. Read the concept →

🔁

Learning Loop

Event-triggered, not time-based. Fires on corrections, completions and failures — and now on what you affirm, so the agent can learn to keep doing something and not only to stop. Proposals reach you for a ruling; nothing rewrites behaviour on its own.

🔍

Smart Search

Brave, Bing, Perplexity, and live xAI behind one tool. Keyword-routed automatically — news → live, shopping → Bing, deep research → Perplexity. No provider-picking.

📍

Situational Awareness

Knows your calendar, travel state, and — opt-in — live location. The agent acts with context: terse when you're boarding, quiet when you're in the air.

💡

Proactive

Notices things when you go quiet. After an idle window it surfaces what matters — an upcoming event, an open promise — or stays silent. Off by default.

🚀

One-Command Onboarding

One curl-bash install, guided setup wizard, migration from OpenClaw or Hermes. Your agent arrives with full memory, identity, warm store, and behavioral contract intact. Proven in production Jun 28.


Pricing

SIMPLE TIERS.
REAL OWNERSHIP.

Early access pricing. Numbers finalized before public launch.

Tier 1
Personal
$XX/mo

One agent, one operator. Cloud runtime, persistent memory, all channels.

  • Single agent instance
  • Persistent memory
  • Slack + WhatsApp + iOS
  • ThunderBrowser included
  • Community support
Get Early Access
Tier 3
ThunderMind
$XX/mo

Sovereign inference. 70B always loaded. Zero cloud dependency.

  • Everything in Pro
  • Local 70B inference
  • Zero marginal token cost
  • Qdrant live memory
  • White-glove onboarding
  • Hardware spec consultation
Get Early Access

Install

GET THUNDERGATE.

One command. Runs the wizard. You're live in under 5 minutes.

Checking access…

You're in. Here's your install code.

Mac / Linux
curl -fsSL https://thunderai.us/install.sh | bash
Then run the setup wizard
thundergate setup

The wizard handles: Node.js check, npm install, Xenova cache, API key config, channel connections, and first agent boot. Already on OpenClaw or Hermes? The wizard includes a one-step migration import.

Windows (PowerShell)
iwr https://thunderai.us/install.ps1 | iex

Windows installer coming soon.

Documentation

Full operator manual, CLI reference, and agent toolbox:

📄 README — ThunderGate v1.3.0 (Operator's Manual) → 📚 User Guide — every corner of ThunderGate → 📝 Release Notes 1.3.0 — What's New for You → 📝 Release Notes 1.2.6 — Previous Release →

Sign in to access install instructions.