A local-first AI workbench that connects chat, coding, runs, routines, refactoring, notebooks, Telegram, and provider routing into one practical workflow.
Updated: 2026-05-21
omnux is not just another chat UI. It is a local-first AI workbench that keeps conversations, coding runs, generated files, validation logs, routines, logic graphs, notebooks, Safe Refactor previews, and Telegram control in the same operational flow.
It does not assume that one expensive model or one expensive machine is the answer. You can use Groq, Gemini, Cerebras, NVIDIA NIM, Copilot, and Codex from the same Tauri desktop UI, then keep the result as files, logs, previews, and recoverable run snapshots.
Start here: Quickstart.
A lot of AI tools look good until the first answer is over. Then the trail disappears.
- Generated files and commands are hard to find later.
- The Tauri desktop UI and the Telegram bot do not share the same capabilities.
- Coding output is hard to rerun, compare, or inspect.
- Refactoring often means overwriting a file without a real preview.
- Scheduled jobs, chat context, search, and handoff notes live in separate places.
omnux turns those loose pieces into a working loop.
| Common problem | omnux approach |
|---|---|
| Chat and execution are separate | Conversations, coding results, and logs stay connected |
| Provider comparison is messy | Single, orchestration, and multi-LLM modes work in chat and coding |
| Desktop UI and Telegram drift apart | Both use the same CommandService layer |
| Refactoring feels risky | Preview, stale validation, and guarded apply |
| Ops state is unclear | /healthz, /readyz, and doctor --json |
| Chat | Coding |
|---|---|
 |
 |
| Safe Refactor | Routines |
|---|---|
 |
 |
| Logic Graph | Notebooks |
|---|---|
 |
 |
| Plans | Skills |
|---|---|
 |
 |
| Settings | Mobile closed |
|---|---|
 |
 |
| Mobile composer open |
|---|
 |
- Chat: single model, orchestration, multi-LLM comparison, Think+, URL/search/memory context, TTS
- Coding: single completion loop, orchestration loop, multi-provider independent runs, preview and rerun support
- Search: Gemini grounding, query rewrite, evidence pack, guard, cache fallback
- Routines: natural-language routine creation, immediate/scheduled runs, browser agent, Telegram delivery
- Logic Graph: connect chat, coding, routines, and tools on a node canvas
- Notebooks: learnings, decisions, verification notes, and handoff documents
- Safe Refactor: anchor edits, LSP rename, ast-grep replacement with preview and guarded apply
- Skills: project/global
SKILL.mdfiles, sticky activation, shared chat and Telegram behavior, single-skill guard, and quick aliases - Remote Tauri UI: LAN access toggle with no OTP prompt for remote clients; read-oriented views, routing policy, and model selection remain available while chat, coding, routines, logic graph execution, auth, secrets, and external-access settings stay blocked
- Security boundaries: remote Tauri UI OTP requests stay blocked, while remote sessions enter limited mode automatically. The permission table, categorized remote-block messages, WebSocket Origin checks, pre-auth message allowlists, local image path limits, attachment count/size rejection, and Markdown raw HTML blocking are documented and covered by the test contract.
macOS/Linux global launcher:
omnux setup
omnux
omnux shutdownFrom a fresh checkout, run ./scripts/omnux setup first. It checks or installs dependencies, builds the middleware, runs npm test, and registers the launcher. A first omnux start also attempts automatic setup if the setup marker is missing.
Manual run:
dotnet run --project apps/omnux-middleware/Omnux.Middleware.csprojWindows:
.\scripts\omnux.ps1 setup
dotnet run --project apps\omnux-middleware\Omnux.Middleware.csprojOpen:
- Tauri UI: http://127.0.0.1:1420/
- Remote UI:
http://<LAN-IP>:1420/ - Middleware API/WS: http://127.0.0.1:41880/
- Health: http://127.0.0.1:41880/healthz
- Ready: http://127.0.0.1:41880/readyz
| provider key | Label | Integration |
|---|---|---|
gemini |
Gemini | Google API and grounded search |
groq |
Groq | OpenAI-compatible HTTP |
cerebras |
Cerebras | HTTP API |
nvidia |
NVIDIA NIM | OpenAI-compatible https://integrate.api.nvidia.com/v1 |
copilot |
Copilot | gh/copilot CLI |
codex |
Codex | codex CLI or API key |