NELA

NELA is a local-first desktop AI workspace for chat, documents, vision, speech, podcast generation, and mindmaps.

It runs on-device using Tauri (Rust + React), so your inference pipeline can work without cloud APIs for normal usage. Internet is only needed for downloading models (and optionally searching Hugging Face inside the app).

Why NELA

• Local intelligence with no required cloud inference backend.
• Workspace-oriented UX with import/export via .nela project files.
• Multi-mode assistant in one app:
  • Chat with optional document grounding (RAG).
  • Vision chat over images.
  • Audio mode for TTS output and voice input transcription.
  • Podcast Studio to generate two-speaker audio episodes from your docs.
  • Mindmap generation for study/planning/brainstorming.
• Model management UI with compatibility scoring and runtime parameter controls.
• Guided onboarding tours built into the app.

Repository Layout

• genhat-desktop: Main NELA desktop app (Tauri + React).
• models: Model storage (gitignored in normal development).
• benchmark: Runtime benchmark suite and plotting tools.
• The-Bare: Standalone experiments/prototypes.

Quick Start For Users

If you already have a release build, install and launch it directly.

If you want to run from source, follow the steps below.

Run From Source

1. Prerequisites

• Node.js 24 or newer
• npm
• Rust stable toolchain (Cargo), rustc 1.77.2+ recommended

Linux system packages (Ubuntu/Debian):

sudo apt-get update
sudo apt-get install -y \
  libwebkit2gtk-4.1-dev \
  libgtk-3-dev \
  libayatana-appindicator3-dev \
  librsvg2-dev \
  libasound2-dev \
  pkg-config

2. Install Dependencies

cd genhat-desktop
npm ci

3. Start NELA In Development Mode

npx tauri dev

The app opens with the startup modal where you can create a new workspace or import an existing .nela project.

First-Time In-App Flow

1. Create a workspace (or import a .nela file).
2. Open Settings and download required models for your workflow.
3. Choose a mode from the input bar: Chat, Vision, Audio, Podcast, or Mindmap.
4. Select an installed model for that mode.
5. Start interacting.

Tip: Use Help -> Tours if you want a guided walkthrough.

Core Features

Workspaces And Project Files

• Workspaces isolate chats, documents, podcasts, and generated assets.
• Export/import supported via .nela archives.

Chat + RAG

• Ask normal LLM questions.
• Add files/folders to build a local knowledge base.
• RAG pipeline supports chunking, retrieval, enrichment, and source-backed answers.

Single-file ingestion supports document/code/audio types including:

• pdf, docx, pptx
• txt, md, csv, json, toml, yaml/yml, xml, html, css, log
• rs, py, js, ts, c, cpp, h, java, go, rb, sh, bat
• mp3, wav, m4a, ogg, flac, aac, wma, webm

Vision Mode

• Select an image and run multimodal Q&A with a vision-capable model.

Audio Mode

• Generate speech from text with selectable voices and speed.
• Use microphone input to transcribe voice prompts.

Podcast Studio

• Generate a two-speaker scripted discussion from ingested docs.
• Produces per-line audio segments and a combined episode track.

Mindmaps

• Generate concept trees from model knowledge or document-grounded context.
• Saved mindmaps are attached to workspace sessions.

Model Controls

• Install/uninstall models inside the app.
• Configure runtime parameters (context size, max tokens, temperature, top-p, top-k, repeat penalty, and backend-specific controls).
• Built-in compatibility estimator helps prevent downloading models that are likely too heavy for the current machine.

Models And Storage

By default, models are resolved from the repository-level models directory during development.

You can override model location with an environment variable before launching the app:

Linux/macOS:

export GENHAT_MODEL_PATH=/absolute/path/to/models

Windows PowerShell:

$env:GENHAT_MODEL_PATH="D:\\path\\to\\models"

Developer Section

This section focuses on running and compiling source code manually.

A. Fast Dev Loop

Use Tauri dev mode with live frontend updates:

cd genhat-desktop
npm ci
npx tauri dev

B. Manual Compilation (No Tauri Dev Runner)

This path compiles frontend and backend explicitly, then runs the compiled desktop binary.

1. Build frontend assets:

   cd genhat-desktop npm ci npm run build

2. Compile Rust backend manually:

   cd src-tauri cargo build

3. Run compiled app from source tree:

   cargo run

Notes:

• Step 1 is required because Tauri loads frontend files from genhat-desktop/dist in non-dev runs.
• In debug mode, model discovery falls back to repository models automatically.

C. Create Installable Bundles

From genhat-desktop:

Linux deb:

npx tauri build --bundles deb

Windows msi/nsis:

npx tauri build --bundles msi,nsis

macOS dmg:

npx tauri build --bundles dmg

Bundle outputs are written under:

• genhat-desktop/src-tauri/target/release/bundle

Benchmarking

Use the benchmark suite in benchmark to capture startup, memory, CPU, process-tree stats, per-model load metrics, and plots.

Quick setup:

python3 -m venv .venv-benchmark
source .venv-benchmark/bin/activate
pip install -r benchmark/requirements.txt

Quick run (launch mode):

python3 benchmark/run_benchmark.py \
  --repo-root . \
  --mode launch \
  --launch-cmd "cd genhat-desktop && npx tauri dev" \
  --interactive \
  --shutdown-after-benchmark \
  --sanitize-launch-env

Full details are in benchmark/README.md.

Troubleshooting

• App fails to open on Linux: verify the Linux packages listed in prerequisites are installed.
• Models not visible: check your model directory and GENHAT_MODEL_PATH.
• RAG answers are weak: verify embedding/grader/classifier models are installed in Settings.
• Very large model is slow or unstable: use the compatibility hints and reduce context size/max tokens.

Additional References

• Main desktop app: genhat-desktop
• In-app help guide source: genhat-desktop/src/content/help-guide.md
• Model compatibility notes: genhat-desktop/docs/MODEL_COMPATIBILITY.md
• Benchmark docs: benchmark/README.md

Status

This repository is actively evolving. If you are onboarding a new team member, start with this README, then open the in-app tours after launching NELA.