Now available — v1.0.27

The AI orchestrator that runs on your machine.

Coordinate Claude, Codex, Gemini, and local models from one desktop app. Works fully offline with Ollama. Your data never leaves your machine.

$curl -fsSL https://nymiria.com/install.sh | bash
3
Platforms
100%
Local-first
0
Gateways exposed
CAPABILITIES

Built for people who live in their tools.

Orchestrates every model you have

Route Claude, Codex, Gemini, and local models from one desktop. Handoff preserves context; subagents run in parallel.

Runs fully offline when you need it to

Built-in Ollama and llama.cpp. Flip to Local Inference mode and your transcript never touches the network.

Remembers what matters

Per-project memory, decisions, and manifest. Nothing is lost between sessions — and nothing syncs to a cloud unless you ask.

Automates what repeats

Native crons, skills, and a Kanban board live in the app. Schedule briefings, code reviews, or anything else without leaving.

Shares state with the CLI you already use

First-class integration with Claude Code. One channel, one context, one source of truth.

Bring your own keys

BYOK for every provider. Keys stay in your OS keychain. No middleman, no per-token markup.

LOCAL INFERENCE MODE

The only AI orchestrator that works with zero network.

Flip one switch. Nymiria routes every request to Ollama or llama.cpp on your machine. No tokens leave the device, no keys required, no rate limits. Drop into an airplane, a sensitive site, or a poorly-connected cafe — Nymiria keeps working.

01

Every chat, every tool call

No provider fallback. If a request would touch the network, it doesn’t fire.

02

Models you choose

Llama, Qwen, DeepSeek, Mistral — anything your hardware can run.

03

Air-gapped visible proof

Status badge in the window chrome shows network = 0 bytes out while the mode is on.

Privacy isn’t a feature. It’s the floor.

Every product tells you privacy matters. We designed the runtime so we couldn’t violate yours even if we wanted to.

We do
We don’t
Store your chats on your machine
Store them in a cloud we control
Use your own API keys (BYOK)
Resell you tokens with a markup
Support fully offline mode
Require a network to function
Keep keys in your OS keychain
Keep keys in plaintext config
Ship with zero telemetry by default
Phone home for product analytics
Open the source for what matters
Hide the boundaries of our runtime
SECURITY ARCHITECTURE

No gateway. No exposure. By design.

Tools like OpenClaw open a local WebSocket server to bridge their agent to the browser. That gateway became the attack surface for ClawJacked — a vulnerability that lets any website you visit silently take full control of your AI agent with zero clicks.

Nymiria has no gateway. Agent execution lives inside the app process. There is no port, no WebSocket server, and no loopback interface for malicious code to reach.

Nymiria
Gateway-based agents (OpenClaw, etc.)
No local server or gateway — nothing listens on a port
Exposes a local WebSocket server any website can reach
No cross-origin attack surface inside the app
Localhost exempted from rate limiting — brute-forceable in seconds
No device pairing over loopback with auto-approval
Visiting one malicious site = full agent takeover, no click required
Credentials stored in OS keychain, never in config files
Auth tokens accessible via config dump after gateway breach
Zero network-exposed surface for agent commands
Tens of thousands of instances exposed to the internet by default

OpenClaw vulnerability references: The Hacker News · Oasis Security

EARLY ACCESS

Stay in the loop.

Get notified about new releases, local inference updates, and early access to Nymiria Pro.

No spam. Unsubscribe any time.