Open SourceMIT License

IIMAGINE Desktop

An open-source desktop app for running local AI models. Full chat UI, voice, plugins, and an OpenAI-compatible API — everything runs on your machine with zero external dependencies.

Three privacy levels.

Built for professionals in health, finance, and legal who have strict data compliance requirements.

Local
Maximum privacy

Models run on your machine via the bundled inference engine. Data never leaves your device.

Regional Cloud
Compliant

Access GPT, Claude, or Gemini through your organization's approved cloud (Azure, AWS, GCP) with your own credentials.

Third-party Cloud
Convenience

Direct API calls to providers using your own API keys, or via an IIMAGINE subscription.

Features.

Full chat UI

Streaming responses, conversation history, and model switching — all running locally.

Local inference engine

Built on llama.cpp. Run any GGUF model with GPU offloading and hardware-based recommendations.

Voice & TTS

Local text-to-speech via Kokoro model. Voice input, voice chat mode, and standalone audio generation.

Plugin system

WordPress-style plugin architecture. Build or install plugins that extend every part of the app.

OpenAI-compatible API

Localhost endpoint that other apps can connect to. Use your local models from any tool that speaks OpenAI.

Encrypted storage

Conversations stored in SQLite + SQLCipher. Your data is encrypted at rest on your own machine.

MCP client

Model Context Protocol for connecting external tool servers. PATH-aware for Electron environments.

Agent runner

Agentic task decomposition and sequential execution with tool calling — entirely local.

Skills manager

Structured markdown skills injected into system prompt. Customize model behaviour without code.

Technology.

App framework

Electron 33

Inference

llama.cpp

Storage

SQLite + SQLCipher

Vectors

sqlite-vec

Plugins

Node.js modules

Models

GGUF (HuggingFace)

TTS

Kokoro (local)

Platforms

macOS / Windows / Linux

License

MIT

Privacy is an architecture decision.

The desktop app runs entirely on your machine. No accounts, no telemetry, no network calls. When you run a local model, the boundary closes and zero bytes leave your device. This isn't a settings toggle — it's how the app is built.