Interface, consent, permissions and OS integration.
OPEN ON-DEVICE COGNITIVE SYSTEM
Give your device
a cognitive foundation.
Build local AI applications with a Rust-owned foundation for model execution, conversations, tools, knowledge and workflows. Use Omnix to compose capabilities, or use its modules independently through Rust and Dart APIs.
Now powers OmniAsk. Explore the same foundation in your own applications. APIs may change during the preview; check compatibility for your runtime and target platform.
Composable capabilities. Explicit boundaries.
Choose the capabilities your application needs, with shared behavior implemented in Rust. Keep your own interface, consent flow and OS integration while Omnix coordinates local models, tools and tasks.
Rust-owned orchestration over explicitly selected modules, exposed through Rust and Dart APIs.
Explicitly selected model runtimes, stores and protocols.
Start with one module. Build your system.
Core
Select LiteRT-LM or llama.cpp for local inference. Streaming, cancellation, runtime controls and embeddings; multimodal input depends on the model and adapter.
Agents + capabilities
Rust-owned streaming model/tool loops with host-supplied tools, skills and execution policies. Optional composition in Omnix, not a separate inference runtime.
Workflow
Durable execution state, step results, retry rules and recovery. Hosts provide triggers and OS scheduling; Android has an example scheduling adapter.
Knowledge
Index local sources, select embedding profiles and retrieve relevant context with source attribution for your model.
Nexus
Paired A2A communication and authority checks; Matter controller integration remains experimental.
Storage
Persist structured records with SQLite and embedding vectors with Qdrant. Choose either backend independently or use both.
One runtime. One conversation. Your interface.
Omnix stays headless. Bring your own presentation and adopt only the capabilities you need. This native Core example requires a configured build and trusted local model.
Read the quick startimport 'package:omnix_core/omnix_core.dart';
// Select runtimes in the app's native build configuration.
final core = OmnixCore();
final model = await core.loadModel(modelPath: trustedModelPath);
try {
final chat = await model.startConversation();
try {
await for (final text in chat.sendStream('Hello, locally.')) {
print(text);
}
} finally {
await chat.close();
}
} finally {
await model.close();
}Neutral infrastructure. A shared direction.
OmniAsk uses Omnix for its local AI backend. Try the example application to explore capabilities before integrating them into your own product. You can use Omnix without OmniAsk, an account or a marketplace, or choose standalone Omnix Core for model execution.