UltraSushiTron mascot
UST-001 /
PRODUCT 製品

Your AI. Your Secrets.
Your Control.

A security-first AI agent framework with zero-knowledge architecture, multi-provider LLM routing, and full autonomy across desktop, mobile, and CLI. Humans approve every secret access — your AI never sees plaintext.

type ai agent framework
runtime rust / wasm
security zero-knowledge, e2e encrypted
license open source
Get Started View Docs

CAPABILITIES 機能

Built for Security, Designed for Developers

Every layer is designed to keep secrets safe while letting AI agents do their work.

—

Isolated Secrets Agent

Secrets live in a separate Rust microservice with no network access. Unix socket only. Air-gapped by design.

—

Context-Aware Approvals

See exactly what your AI wants to do with a secret before granting access. Full context, no blind trust.

—

Zero-Knowledge Core

The AI core never sees plaintext secrets. Encrypted values are injected only after user approval, then wiped.

—

23 Verified Plugins

Seven categories — security, cloud, CI/CD, AI, monitoring, chat, and dev tools — running in a capability-enforced WASM sandbox with automated security scanning.

—

Hash-Chained Audits

Every action is recorded in a tamper-proof, Ed25519-signed, hash-chained audit log. Full accountability.

—

Dual-Path Memory

AES-256-GCM encrypted store with both full-text search and embedding-based recall. Degrades gracefully to lexical-only mode when embeddings are unavailable.

—

Smart Scheduler

Natural language cron parsing — "every weekday at 8am" — with LLM fallback. Unified scheduling across desktop, mobile, and CLI with execution logging.

—

18 Built-in Tools

Read, Write, Edit, Bash, Grep, Find, HttpFetch, WebSearch, Diff, Patch, ListDir, Browser, ComputerUse, MemoryRecall, MemoryStore, Skill, Task, and more.

AGENT エージェント

Enterprise-Grade Agent Features

Full autonomous agent capabilities with planning, research, computer use, and multi-provider LLM routing.

—

Browser Automation

Playwright-based web automation with 20+ actions. Navigate, click, type, screenshot — via a dedicated HTTP microservice with browser pooling.

—

Computer Use & Vision

Native screen capture, desktop interaction, and visual document processing with policy-enforced safety controls across platforms.

—

Explicit Planning

Goal decomposition before execution. Break complex tasks into atomic, executable steps with progress tracking.

—

Multi-Agent Orchestration

Spawn sub-agents for parallel execution. Category-based delegation: quick, deep, ultrabrain. Remote task envelopes with Ed25519 signatures.

—

Parallel Research

Concurrent research across web, GitHub, and docs. Aggregate results from multiple sources with learning engine integration.

—

Multi-Provider LLM

Route across 7+ providers — OpenAI, Anthropic, Xiaomi MiMo, MiniMax, Alibaba Qwen, Kimi, and more. Provider opt-in by default with per-provider overrides.

—

MCP Integration

Model Context Protocol client with JSON-RPC 2.0 tools, schema discovery, and bridge implementation for external tool exposure.

—

Skill Packaging

Auto-capture successful agent workflows as reusable skills. Record, package, share, and replay across your organization via the team registry.

ARCHITECTURE アーキテクチャ

Multi-Zone Security Architecture

Defense in depth with three isolated security zones.

Zone 0 — Secure Enclave
Mobile App
Secret approval, admin dashboard, push notifications, biometric auth, scheduler control
WebSocket + E2E Encryption
Zone 1 — Air-Gapped
Secrets Agent
SQLite vault, AES-256-GCM encryption, Unix socket only
Unix Socket (gRPC)
Zone 2 — Full Access
AI Core
Multi-provider LLM routing, 18 tools, WASM sandbox, computer use, MCP bridge

INTEGRATION 統合

See It in Action

A tool requests a secret — the user approves on their phone — the agent continues.

Agent requests API key json 
{
  "tool": "HttpFetch",
  "arguments": {
    "url": "https://api.example.com/data",
    "method": "GET",
    "headers": {
      "Authorization": "Bearer {{secret:api_key}}"
    }
  }
}
User receives approval request text 
Secret Request
------------------------------
Secret: api_key (API_KEY)
Tool:   HttpFetch
Reason: GET https://api.example.com/data
Duration: One-time use

        [Approve]  [Deny]

SECURITY セキュリティ

Enterprise-Grade Cryptography

Ed25519 Signatures
AES-256-GCM
X25519 ECDH
Argon2id KDF
WASM Sandbox
SHA-256 Hash Chains

CHANNELS チャンネル

Manage Secrets From Your Chat

Bidirectional chat connectors let your team request, approve, and rotate secrets without leaving their favorite platform.

/

Slack

Slash commands with Block Kit responses, thread replies, and interactive approval workflows.

Channel Plugin
/

Telegram

Inline bot commands with HTML formatting, reply threading, and real-time notifications.

Channel Plugin
/

Discord

Slash commands with rich embeds, ephemeral responses, and color-coded status indicators.

Channel Plugin
/

Mobile App

Manage sessions, approve secret requests, and chat with your AI agent directly from your phone.

Mobile App

PLATFORMS プラットフォーム

Run Everywhere

Native experiences across desktop, mobile, CLI, and web with full feature parity.

/

Desktop Shell

macOS and Linux native app with tray management, onboarding, crash recovery, process supervision, and local ML routing.

macOS / Linux
/

Mobile App

React Native app with iOS Keychain and Android Keystore integration. Approve secrets, manage schedules, browse memory, and chat with your agent.

iOS / Android
/

CLI & TUI

Full-featured command line with interactive setup, natural language scheduling, provider management, and status monitoring.

Terminal
/

Web Approval Client

Browser-based approval UI with WebCrypto, non-extractable keys, Ed25519 verification, and X25519 ECDH encryption.

Browser

GET STARTED 始める

Ready to Take Control?

Get started with UltraSushiTron in minutes.

Install bash 
git clone https://github.com/jaikoo/UltraSushiTron.git
cd UltraSushiTron
cargo build --release
Read the Docs