How it works

Persistent context,
specialized execution.

Your deck holds project memory, writes code, runs scripts, and handles daily work across eight chat platforms. When a job outgrows the workspace — full repo cloning, multi-file refactors, running test suites — you tell it to spawn an isolated worker that plans, codes, tests, and ships a PR, then shuts down.

The deck · always on
↩ Answers arrive in the same conversation
Heavy engineering?
Workers · on demand
↩ events stream back to chat; you can jump in and steer anytime
01 · you message

Send a message from any connected platform — Telegram, Discord, or the web UI. Your deck is always on, same context regardless of which channel you use.

Architecture

One workspace.
Workers when you need them.

The Deck · Moltis

Always-on AI workspace.

Moltis is an open-source, single-binary Rust application — no npm dependencies, no plugin runtime, no supply chain to audit. It runs in a sandboxed container, maintains long-term hybrid memory (vector + full-text search), searches the web, writes code, runs shell commands, creates its own skills at runtime, and connects to eight chat platforms. Most tasks — including coding, scripting, and automation — complete entirely inside the deck.

Workers · OpenHands + Browser

Isolated environments for heavy work.

When a task outgrows the deck — full repo cloning, multi-file engineering, running full test suites, or browser automation — your deck spawns an isolated worker container. The worker has its own resources so your workspace stays clean: it can install dependencies, trash its own environment, and get destroyed when it's done. Three reasons to use a worker instead of the deck: resource isolation (workers get 1-3GB RAM vs the deck's 512MB), autonomous multi-step execution (workers iterate on failures without you steering each step), and disposability (nothing the worker does can damage your persistent workspace).

Autonomous PRs
Plans, edits, tests, commits, and pushes — streaming events back to your deck until the PR is ready. $0.05/hr.
Interactive workspace
Web-based VSCode + browser. Watch the worker live, take over the terminal, or drive the browser alongside it. $0.15/hr.
Browser session
Isolated Chromium
Scraping, form automation, or JS-heavy sites that need real rendering. $0.05/hr.
Compute

The deck does most of the work.
Workers handle the rest.

Your deck runs code, scripts, and automations natively — 24/7, no extra cost. When a task needs its own isolated environment (full repo cloning, multi-file engineering, browser automation), spawn a worker container. Per-second billing, no idle charges. Workers are completely optional.

// always-on · included

The Deck $7 / month

Your persistent workspace. Chat, coding, scripting, shell execution, eight chat channels, voice, MCP tools, web search, cron jobs, webhooks. Handles most tasks natively — no worker needed.

ContainerDedicated, sandboxed
IsolationgVisor runtime
Uptime24 / 7
// on demand · optional

Coding Worker from $0.05 / hr

Autonomous multi-file engineering in an isolated container. Clones repos, installs deps, runs tests, opens PRs. Headless or with a full VSCode IDE.

Headless (OpenHands CLI)$0.05/hr
GUI (VSCode + browser)$0.15/hr
Billingper second, 60s min
// on demand · optional

Browser Worker $0.05 / hr

Isolated Chromium for scraping, form automation, or JS-heavy sites. Full Chrome DevTools Protocol control.

IsolationSandboxed container
Billingper second, 60s min
Inside Moltis

One binary.
Every capability.

The deck isn't a thin wrapper around an LLM. It's a full AI server — sandboxed shell execution, coding, hybrid memory, eight chat channels, voice, encryption, self-extending skills. Here's what runs under the hood.

// full list
Security
  • Passkeys (WebAuthn)
  • Scoped API keys
  • Encryption-at-rest vault
  • Secrets zeroed on drop
  • Human-in-the-loop approval
  • No unsafe Rust
LLM Providers
  • Anthropic · OpenAI · Gemini · OpenRouter
  • DeepSeek · Mistral · Groq · xAI
  • Custom OpenAI-compatible
  • Provider fallback chains
  • Per-provider metrics
Memory
  • Hybrid search (vector + FTS)
  • Local GGUF embeddings
  • OpenAI batch API (50% off)
  • Embedding cache
  • File watching · live sync
  • Session export
Extensibility
  • MCP server support
  • GraphQL API
  • CalDAV calendar
  • Lifecycle hook system
  • Cron job scheduling
  • TOML configuration
  • Webhook triggers
Channels
  • Web UI
  • Telegram · WhatsApp · Discord
  • Slack · Teams · Matrix · Nostr
  • JSON-RPC + GraphQL API
  • Push notifications
Observability
  • Prometheus metrics
  • OpenTelemetry tracing
  • Structured logging
  • Per-provider charts
  • SQLite persistence
  • Real-time WebSocket
Managed hosting

We run the infra.
You review the PRs.

Both Moltis and OpenHands are open source — you could self-host them. Nebula Deck is for people who'd rather not. We handle provisioning, networking, sandboxing, updates, credential management, and the glue between Moltis and the coding worker layer.

01 / One-click deploy

Sign up, plug in a key, go.

We provision your Moltis workspace, wire up your chat bridges, and hand you a private subdomain. No servers, no config files, no Docker daemon on your laptop.

02 / Updates on your terms

Update now, schedule for tonight, or roll back.

When a new release is ready, you choose when it lands — update instantly, schedule it for off-hours, or stay on the current version. Something feel off? Roll back to the previous version in one click. Your memory, skills, settings and all other files carry over every time.

03 / Encrypted credentials

Keys encrypted at rest, zeroed on drop.

Your deck stores API keys in an encrypted vault — same as self-hosted Moltis. When coding workers need model access, the orchestrator injects the key server-side so ephemeral containers never hold credentials. Vault contents are redacted from all logs and zeroed from memory the moment they're no longer needed.

04 / Tenant isolation

Every workspace runs in its own world.

Each customer gets a dedicated container with restricted syscalls and isolated networking. Untrusted LLM-generated code can't touch other tenants or the host. Workers run in their own ephemeral containers with the same isolation.

05 / Webhooks + automation

Your deck watches. Workers act when needed.

GitHub PR opened? Stripe payment received? Your deck picks up the event, triages it, and decides whether it needs attention — or whether to spawn a worker for real work. You stay in control of what triggers what.

06 / Per-second billing

The meter stops when the worker does.

Your deck runs 24/7 for a flat fee. Workers are billed per second with a 60-second minimum — no rounding up to the next hour. LLM tokens are charged at your provider's rates. No markup, no opaque credits.

Your AI workspace.
Always on, one bill.