IJAI Platform

This project is currently on hold due to personal circumstances.
Development is expected to resume after December 6, 2025.

( Work in progress )

IJAI is a modular AI platform that integrates Large Language Models (LLM), Speech-to-Text (STT), and Text-to-Speech (TTS) with a cross-platform Electron + React desktop application. It provides a unified environment for model management, dataset handling, configuration control, and plugin-based extensions.

Features

Cross-platform desktop application (Electron).
Integrated LLM, STT, and TTS pipelines.
Config-driven architecture for models and assistant policies.
Support for custom datasets and corpora.
Extensible plugin system (OpenAPI-based).
Prebuilt model installers and Python integration scripts.

Join Us

Scan the QR code to connect to IJAI.

Repository Structure

Directory	Purpose	Notes
`App/`	Main Electron application — UI, launch, and packaging.	Build/dev scripts: `npm run build`, `npm run dev`.
`IJAI-configs/`	Assistant and model configuration files (profiles, launch parameters).	Keep example configs under `examples/`.
`dataset/`	Datasets for LLM training/fine-tuning, corpora for STT, collections for TTS.	Specify format and license next to each dataset.
`models/`	Bundled models and installers (GGUF, ONNX, etc.).	Add a README inside each model folder with download instructions.
`plugins/`	Plugin system: OpenAPI integrations, extensions, and sample plugins.	Examples live in `plugins/examples/`.
`share/lexicons/`	Lexicon resources: dictionaries, transcription tables, localization files.	Version lexicons and cite the source.

File Overview

Path	Description
App/src/assets/icons/IJAI-logo.png	Application logo
App/src/assets/icons/.gitkeep	# gonna be released
App/src/components/ChartCard.js	Chart visualization component
App/src/components/DataTable.js	Tabular data component
App/src/components/Header.js	Application header
App/src/components/Sidebar.js	Sidebar navigation
App/src/styles/dashboard.css	Dashboard styling
App/src/styles/material.css	Material Design overrides
App/index.html	Main HTML entry point
App/renderer.js	Electron renderer process
App/main.js	Electron main process
App/preload.js	Secure preload bridge
App/README.md	Documentation for app module
App/electron-builder.yml	Build configuration (electron-builder)
App/forge.config.js	Forge build configuration
App/package.json	NPM dependencies and scripts
IJAI-configs/assistant.conf.yaml	Assistant runtime configuration
IJAI-configs/models.conf.json	Model registry definitions
IJAI-configs/policies.yaml	Policy rules
dataset/llm/set1/doc.md	LLM training notes
dataset/llm/set2/.gitkeep	# gonna be released
dataset/llm/set8/doc.md	Supplemental dataset
dataset/llm/articles_names.txt	Corpus index of articles
dataset/llm/doc.md	General dataset description
dataset/stt/book1.txt	STT training text
dataset/tts/voice1_set/.gitkeep	# gonna be released
dataset/tts/voice2_set/.gitkeep	# gonna be released
dataset/tts/book1.txt	TTS training corpus
dataset/important.jl	Julia dataset file
dataset/magnificent.asm	Assembly dataset (experimental)
dataset/examples/Текстовый документ.txt	# gonna be released
dataset/files/info/Текстовый документ.txt	Metadata resources
models/codellm/README.md	Model documentation
models/codellm/codellama.py	Python integration
models/codellm/installer.sh	Installer script
models/ollama-deepseekr1:8b/README.md	Model documentation
models/ollama-deepseekr1:8b/deepseekr1.py	Python integration
models/ollama-deepseekr1:8b/installer.sh	Installer script
models/llm/gpt-neo/add_info.md	Additional notes
models/llm/gpt-neo/gptneo.py	GPT-Neo integration
models/llm/gpt-neo/merges.txt, vocab.json	Tokenizer files
models/llm/gpt-neo/special_tokens_map.json	Special tokens
models/llm/gpt-neo/tokenizer.json, tokenizer_config.json	Tokenizer configs
models/llm/gpt2-medium/config.json	Model configuration
models/llm/gpt2-medium/gpt2m.py	GPT-2 medium integration
models/llm/gpt2-medium/tokinizer.json	Tokenizer config (naming typo)
models/ollama-llama3/DISCLAIMER.md	Usage disclaimer
models/ollama-llama3/installer.sh	Installer script
models/ollama-llama3/llm3.py	Integration script
models/phi3mini/README.md	Model documentation
models/phi3mini/added_tokens.json	Model tokenizer additions
models/phi3mini/config.json	Model configuration
models/phi3mini/configuration_phi3.py	Model configuration script
models/phi3mini/generation_config.json	Generation parameters
models/phi3mini/model.safetensors	Model weights
models/phi3mini/modeling_phi3.py	Model architecture
models/phi3mini/script.py	Helper scripts
models/phi3mini/special_tokens_map.json	Special tokens map
models/phi3mini/tokenizer.json, tokenizer.model, tokenizer_config.json	Tokenizer files
models/stt/Coqui/init.py	Python package initialization
models/stt/Coqui/coqui_stt.py	STT integration script
models/stt/Silero/README.md	Model documentation
models/stt/Silero/init.py	Python package initialization
models/stt/Silero/silero_stt.py	STT integration script
models/stt/whisper-small/config.json, preprocessor_config.json, tokenizer.json	Whisper model configs
models/stt/whisper-small/whisper-small.py	Whisper integration script
models/tts/tts-small/Текстовый документ.txt	# gonna be released
models/vocoder/Текстовый документ.txt	# gonna be released
plugins/weather/manifest.json	Plugin manifest
plugins/weather/openapi.yaml	OpenAPI schema
share/lexicons/Текстовый документ.txt	# gonna be released
.gitignore	Git ignore rules
COMMERCIAL-LICENSE.md	Commercial license terms
CONTRIBUTORS.md	Contributor acknowledgments
LICENSE	Open source license
README.md	Main documentation

Benchmarks (LLM Inference Speed)

Model	Params	GPU (RTX 2080s)	CPU (i7-10700)
Phi-3 Mini	3.8B	~45 tok/s	~6 tok/s
LLaMA 3	8B	~22 tok/s	~3 tok/s
DeepSeek R1	7B	~25 tok/s	~3.5 tok/s
CodeLLaMA	7B	~24 tok/s	~3.2 tok/s
GPT-Neo	2.7B	~40 tok/s	~5.5 tok/s
GPT-2 XL	1.5B	~60 tok/s	~8 tok/s
Falcon-7B	7B	~26 tok/s	~3.5 tok/s
Mistral-7B	7B	~28 tok/s	~3.7 tok/s
Yi-6B	6B	~30 tok/s	~4 tok/s
OPT-6.7B	6.7B	~23 tok/s	~3 tok/s
StableLM-7B	7B	~22 tok/s	~3 tok/s

System Requirements

Large Language Models (LLM)

Model / Folder	CPU	RAM	GPU	Notes
codellm	4+ cores	16 GB+	Optional, recommended for 8B+ models	PyTorch / Ollama compatible
codellama	4+ cores	16 GB+	NVIDIA GPU (RTX 2060+) for smooth inference	8-bit/16-bit quantization recommended
ollama-deepseekr1:8b	4+ cores	16 GB+	NVIDIA GPU for fast generation	Pretrained 8B model, uses Ollama runtime
llm/Falcon-7B	4+ cores	16 GB+	GPU optional, faster with CUDA	TII Falcon, strong general LLM
llm/GPT-NeoX-20B	8+ cores	32 GB+	High-end GPU (24 GB VRAM+) required	Very large model, slow on CPU
llm/Gemma	4+ cores	12 GB+	GPU optional	Compact Google LLM
llm/LLaMA-2	4+ cores	16 GB+	GPU recommended	Meta LLaMA 2 family
llm/MPT	4+ cores	16 GB+	GPU optional	MosaicML transformer family
llm/Mistral-7B	4+ cores	16 GB+	GPU recommended	Highly efficient dense model
llm/OPT	4+ cores	16 GB+	GPU optional	Meta OPT series
llm/Qwen-1.5	4+ cores	16 GB+	GPU optional	Multilingual Alibaba model
llm/StableLM	4+ cores	12 GB+	GPU optional	StabilityAI open family
llm/Yi-1.5-6B	4+ cores	16 GB+	GPU recommended	Bilingual efficiency
llm/gpt-j-6b	4+ cores	16 GB+	GPU optional	EleutherAI GPT-J classic
llm/gpt-neo	4+ cores	12 GB+	GPU optional	EleutherAI GPT-Neo
llm/gpt2-medium	4+ cores	8 GB+	GPU optional	Classic GPT-2 medium
ollama-llama3/phi3mini	4+ cores	12 GB+	GPU recommended	Small LLaMA3 variant

Speech-to-Text (STT)

Model / Folder	CPU	RAM	GPU	Audio Requirements
stt/Coqui	4+ cores	8 GB+	NVIDIA CUDA GPU recommended	WAV, 16 kHz, mono
stt/DeepSpeech	4+ cores	8 GB+	Optional	WAV, 16 kHz, mono
stt/Faster-Whisper	4+ cores	8 GB+	GPU recommended	WAV/OGG, 16 kHz
stt/Nemo ASR	4+ cores	8 GB+	GPU strongly advised	WAV, 16 kHz
stt/Silero	4+ cores	8 GB+	Optional	WAV, 16 kHz, mono
stt/Vosk	4+ cores	4 GB+	Optional	WAV, 16 kHz, mono
stt/whisper-small	4+ cores	8 GB+	GPU recommended	WAV/OGG, 16 kHz preferred

Text-to-Speech (TTS)

Model / Folder	CPU	RAM	GPU	Notes
tts/Parler-TTS	4+ cores	8 GB+	GPU strongly recommended	HuggingFace Parler, realistic voices
tts/tts-small	4+ cores	8 GB+	Optional, recommended for faster synthesis	Lightweight TTS
vocoder	4+ cores	8 GB+	GPU recommended	Converts spectrograms to waveform

Minimal Set

Models included:

LLM: llm/gpt2-medium
STT: stt/Silero
TTS: tts/tts-small
Vocoder: vocoder

Requirements:

Resource	Recommended
CPU	4 cores modern x86_64
RAM	8 GB
GPU (optional)	NVIDIA CUDA GPU (RTX 2060) for faster STT/TTS inference
Storage	~1 GB for all models
Audio format	WAV, 16-bit PCM, mono, 16 kHz

Full Set

Models included:

LLM: codellm, codellama, ollama-deepseekr1:8b, llm/Falcon-7B, llm/GPT-NeoX-20B, llm/Gemma, llm/LLaMA-2, llm/MPT, llm/Mistral-7B, llm/OPT, llm/Qwen-1.5, llm/StableLM, llm/Yi-1.5-6B, llm/gpt-j-6b, llm/gpt-neo, llm/gpt2-medium, ollama-llama3/phi3mini
STT: stt/Coqui, stt/DeepSpeech, stt/Faster-Whisper, stt/Nemo ASR, stt/Silero, stt/Vosk, stt/whisper-small
TTS: tts/Parler-TTS, tts/tts-small
Vocoder: vocoder

Requirements:

Resource	Recommended
CPU	8+ cores modern x86_64
RAM	16 GB+ (32 GB recommended for multiple LLMs)
GPU	NVIDIA CUDA GPU (RTX 3060+ recommended) for smooth inference across LLM, STT, and TTS
Storage	20+ GB depending on models downloaded
Audio format	WAV/OGG, 16-bit PCM, mono, 16 kHz

Minimal Set

Models included:

LLM: llm/gpt2-medium
STT: stt/Silero
TTS: tts/tts-small
Vocoder: vocoder

Requirements:

Resource	Recommended
CPU	4 cores modern x86_64
RAM	8 GB
GPU (optional)	NVIDIA CUDA GPU (e.g., RTX 2060) for faster STT/TTS inference
Storage	~1 GB for all models
Audio format	WAV, 16-bit PCM, mono, 16 kHz

Minimal set runs on CPU, but GPU improves transcription and TTS speed. Suitable for lightweight testing and small projects.

Full Set

Models included:

LLM: codellm, codellama, ollama-deepseekr1:8b, llm/gpt-neo, llm/gpt2-medium, ollama-llama3/phi3mini
STT: stt/Coqui, stt/Silero, stt/whisper-small
TTS: tts/tts-small
Vocoder: vocoder

Requirements:

Resource	Recommended
CPU	8+ cores modern x86_64
RAM	16 GB+ (32 GB recommended for multiple LLMs)
GPU	NVIDIA CUDA GPU (RTX 3060+ recommended) for smooth inference across LLM, STT, and TTS
Storage	10+ GB depending on models downloaded
Audio format	WAV/OGG, 16-bit PCM, mono, 16 kHz

Full set allows full functionality: large LLMs, multiple STT engines, and high-quality TTS. GPU is strongly recommended for smooth experience.

System Architecture

flowchart TB
    %% === SYSTEM CHECK ===
    SYS[System Profiler: CPU GPU RAM]

    %% === OPTIMIZATION PIPELINE ===
    OPT[Auto Configurator: Quantization + Model Selection]
    BENCH[Benchmark Runner: Tokens per sec]
    DEPLOY[Runtime Deployment: Optimized Models]

    %% === RAW MODELS ===
    subgraph RAW [Raw Models]
        M1[CodeLLM or CodeLLaMA]
        M2[DeepSeek R1]
        M3[GPT-Neo and GPT-2]
        M4[LLaMA-3]
        M5[Phi-3 Mini]
        M6[Whisper Small STT]
        M7[TTS Small]
        M8[Vocoder]
    end

    %% === DATASETS ===
    subgraph DATA [Datasets]
        D1[LLM corpora]
        D2[STT transcripts]
        D3[TTS voices]
    end

    %% === CONFIGS ===
    subgraph CFG [Configuration Layer]
        C1[assistant.conf.yaml]
        C2[models.conf.json]
        C3[policies.yaml]
    end

    %% === RUNTIME SERVICES ===
    subgraph RUNTIME [Runtime Services]
        MON[Monitoring and Logging]
        SEC[Policy Engine]
        API[Runtime API]
    end

    %% === UI & PLUGINS ===
    subgraph UI [User Interface]
        U1[Electron Desktop UI]
        P1[Plugins: Weather etc]
    end

    %% === FLOWS ===
    SYS --> OPT --> BENCH --> DEPLOY

    %% DATA → RAW MODELS
    D1 --> M1
    D1 --> M2
    D1 --> M3
    D1 --> M4
    D1 --> M5
    D2 --> M6
    D3 --> M7 --> M8

    %% RAW MODELS → OPTIMIZER
    M1 --> OPT
    M2 --> OPT
    M3 --> OPT
    M4 --> OPT
    M5 --> OPT
    M6 --> OPT
    M7 --> OPT
    M8 --> OPT

    %% OPTIMIZED DEPLOY → RUNTIME SERVICES
    DEPLOY --> API
    DEPLOY --> MON
    DEPLOY --> SEC

    %% CONFIGS
    C1 --> API
    C2 --> API
    C3 --> SEC

    %% UI LAYER
    U1 --> API
    U1 --> MON
    U1 --> SEC
    U1 --> P1
    P1 --> API

Configuration

assistant.conf.yaml – Core assistant runtime configuration.
models.conf.json – Central model registry.
policies.yaml – Execution and safety policies.

Models

Supported families include:

CodeLLM – Code generation.
DeepSeek R1 – Reasoning model.
GPT-Neo & GPT-2 – General language models.
LLaMA-3 – Open-source LLM.
Phi-3 Mini – Lightweight transformer model.
Whisper Small – Speech recognition.
TTS Small + Vocoder – Speech synthesis.

Each model folder provides:

Installer (installer.sh)
Python integration (*.py)
Tokenizer/configuration files
Model weights (.safetensors)

Datasets

LLM corpora – Markdown/text resources.
STT corpora – Speech recognition text files.
TTS corpora – Voice datasets, transcripts.
Experimental – Julia (.jl) and Assembly (.asm) files.

Plugins

Weather plugin as reference implementation.
- manifest.json – Plugin declaration.
- openapi.yaml – API specification.

Installation

# Clone repository
git clone https://github.com/your-org/IJAI.git
cd IJAI

# Install frontend dependencies
cd App
npm install

# Run in development
npm start

# Build production desktop app
npm run build

Model installation is handled by individual installer.sh scripts inside each model folder.

CLI - version

Usage

Start the Electron application.
Configure assistant and models via IJAI-configs/.
Place datasets in dataset/.
Install and load required models from models/.
Extend functionality with plugins/.

Contributing

Follow established coding standards.
Submit pull requests for review.
See CONTRIBUTORS.md for acknowledgments.

Project Stats

Star History

Contributors

Roadmap

✅ Completed (Ready for Release)

🧪 Demo / Experimental

Model fine-tuning workflow (UI + CLI prototype)
Interactive prompt playground for LLMs
Voice cloning demo for TTS
Speech-to-speech pipeline (STT → LLM → TTS)
Minimal plugin marketplace (manual install)

🚧 In Progress

Auto config saver on your flash drive
GPU acceleration benchmarks (CUDA / ROCm)
Model caching & optimized loading (disk + RAM)
Dataset versioning & tagging
CLI tool (ijai-cli) for headless workflows
Enhanced error logging & monitoring dashboard
Extended plugin APIs (beyond Weather)

🎯 Planned for Full Release

Cloud sync & model sharing
Advanced policy engine (safety, filtering, sandboxing)
Fine-tuning UI (drag-and-drop datasets)
Multi-language UI (EN, RU, etc.)
Integration with external APIs (translation, search, etc.)
Plugin marketplace (in-app browsing & install)
Mobile companion app (view results, run lightweight tasks)

License

Open Source License: AGPLv3.0
Commercial License: COMMERCIAL-LICENSE.md

Usage Policy

All models, datasets, and configurations provided in this repository are released strictly for research and educational purposes.
By using this project, you agree to follow these rules:

❌ No malicious usage — It is strictly forbidden to use IJAI or any of its models for harmful purposes, including but not limited to:
- spreading disinformation,
- generating hateful or violent content,
- surveillance or harassment,
- assisting in illegal activities.
❌ No malicious fine-tuning — Fine-tuning IJAI models on datasets intended for harmful, discriminatory, or illegal applications is prohibited.
✅ Allowed usage — You may use, extend, and fine-tune the models for positive, ethical, and constructive purposes such as:
- research,
- accessibility,
- education,
- productivity,
- creativity.

Important: Any violation of these policies voids your right to use or redistribute this project under its license.
We trust the open-source community to act responsibly and improve IJAI in ways that benefit everyone.

💝 Support the Developer

You can support me via crypto, Steam trade or Click. Any help is appreciated!

💳 Crypto Donations

Currency	Network	Address
USDT	TRC20	`TPutzJ12Bs4jAPLT9rkQhvg6PdwHhQfJVB`

🎮 Steam Trade

📲 Click (P2P)

You can send via Click using the app:

Thank you for your support. It helps me keep working on projects.

Prebuilt model weights and installers are available via the official distribution channel: Telegram.

The project is currently in the development stage, with core functionalities being implemented and tested.
Additional features, optimizations, and refinements are planned for subsequent development phases.

Name		Name	Last commit message	Last commit date
Latest commit History 441 Commits
App		App
CLI		CLI
IJAI-configs		IJAI-configs
dataset		dataset
examples		examples
imgs		imgs
models		models
plugins/weather		plugins/weather
share/lexicons		share/lexicons
site		site
.gitignore		.gitignore
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
COMMERCIAL-LICENSE.md		COMMERCIAL-LICENSE.md
CONTRIBUTING.md		CONTRIBUTING.md
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
installer.sh		installer.sh
requirements.txt		requirements.txt

License

IbrokhimN/IJAI

Folders and files

Latest commit

History

Repository files navigation

IJAI Platform

( Work in progress )

Features

Repository Structure

File Overview

Benchmarks (LLM Inference Speed)

System Requirements

Large Language Models (LLM)

Speech-to-Text (STT)

Text-to-Speech (TTS)

Minimal Set

Full Set

Minimal Set

Full Set

System Architecture

Configuration

Models

Datasets

Plugins

Installation

CLI - version

Usage

Contributing

Project Stats

Star History

Contributors

Roadmap

✅ Completed (Ready for Release)

🧪 Demo / Experimental

🚧 In Progress

🎯 Planned for Full Release

License

Usage Policy

💝 Support the Developer

💳 Crypto Donations

🎮 Steam Trade

📲 Click (P2P)

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Contributors 2

Uh oh!

Languages

Packages