From LMBridge

Private AI.
On Your Terms.

Cipher runs powerful language models on your iPhone — or connects to models on your own hardware. No subscriptions. No cloud. No compromises.

Download on the App Store See what it can do
On-device inference
Remote models via LMBridge
40+ MCP tools
No data leaves your hands
9:41 ●●●
DeepSeek-R1 — Remote
LMBridge ›
Explain quantum entanglement simply.
Thinking…
Quantum entanglement links two particles so that measuring one instantly determines the state of the other — no matter how far apart they are.
Run a quick benchmark on my server.
Using tool: terminal
Running sysbench… done. 4,821 events/sec.
Ask anything…

Everything you expect.
Nothing you don't.

Truly private, on-device inference

Run quantized GGUF models directly on your iPhone. Your conversations never touch a server. No account required, no usage tracking, no API keys.

  • Powered by llama.cpp — the fastest on-device inference engine
  • Supports 1B to 7B+ models depending on your device
  • Full support for reasoning / thinking models
  • Works completely offline
Models
📱
Mistral 7B Q4
On-device
📱
Phi-3 Mini Q8
On-device
🌐
DeepSeek-R1 70B
LMBridge
🌐
Llama 3.3 70B
LMBridge

Tap into your own big models

When you need more power, connect Cipher to models running on your own Mac or server via LMBridge. Get frontier-scale inference without paying per token.

  • Stream any model exposed by LM Studio
  • Simple 6-digit PIN pairing — no config files
  • Auto-reconnects on return without re-pairing
  • Switch between local and remote mid-conversation
Connected to Chris's Mac
DeepSeek-R1 70B
Running on LM Studio · streaming
LMBridge Relay
End-to-end WebSocket · wss://
🔒 Your model weights never leave your machine

Models that can actually do things

Give your AI real capabilities with tool calling. Built-in tools run on-device, while MCP servers unlock terminal access, databases, Docker, web browsing, and more.

  • Calculator, datetime, unit converter built-in
  • MCP server catalog with 40+ integrations
  • Per-tool approval — you decide what runs
  • Auto-approve mode for trusted workflows
⚙️ Tool Request
terminal · run_command
Execute: docker ps --format table
Deny
Approve
You stay in control of every tool call. One tap to approve or deny before anything runs.
🧠

Thinking Models

Displays reasoning traces from DeepSeek R1 and similar models. Watch the model think before it answers.

Token Streaming

Real-time token-by-token output. No waiting for the full response — start reading as the model writes.

🔄

Auto-Reconnect

Background the app, come back later — Cipher reconnects to your LMBridge server automatically.

🎛️

Full Control

Tune temperature, token limits, system prompts, and context window per conversation.

📶

Offline First

On-device models work with no internet connection. LMBridge adds remote power when you want it.

🔒

Zero Telemetry

No usage analytics, no conversation logging, no accounts. What you ask stays with you.

40+ tools, one app

The Cipher MCP catalog gives your models the ability to act — not just respond. Add any server in seconds from the built-in store.

💻
Terminal
Run shell commands
🐳
Docker
Manage containers
☸️
Kubernetes
Control clusters
🗄️
Databases
Query SQL & NoSQL
🌍
Web Browse
Search & fetch pages
📁
Filesystem
Read & write files
⚙️
System Auto
Automate your OS
Custom
Add your own MCP server

Your hardware, your models,
available everywhere

LMBridge is the relay that connects Cipher on your phone to large models running on your own Mac or server — securely, over WebSocket.

🖥️
LM Studio
Your Mac · any model
🌉
LMBridge Relay
Secure WebSocket
📱
Cipher
iPhone · anywhere

Pair with a 6-digit PIN. Model weights stay on your machine. The relay only passes tokens — nothing is stored.

Private.
Powerful. Yours.

Download Cipher and start running AI on your own terms today.

Download on the App Store LMBridge on GitHub