Build • Fine-tune • Benchmark • Secure • Evaluate • Deploy
fef A desktop-first IDE for Local AI development.
Train, benchmark, secure, and manage your local AI models from one unified workspace.
RedForge is a Local AI Engineering Platform designed for developers, researchers, students, and AI enthusiasts who want a professional environment for building AI applications without relying on the cloud.
So about why I devoloped something like this, a while ago I watched my favourite youtuber go fromm a beginner to advanced level AI engineer in less than a year. Pewdiepie fine tuned a model into something of a size of GPT. He even made Odysseus, something i ws very fascinated by. I cant really finetune an ai (not much with my comp), then I thought of making something that'll help people Train, Evaluate, Benchmark and Access thier model. So thats how the devolopment of Redforge started.
Think of it as:
VS Code + Docker Desktop + MLflow + LM Studio + Postman
for Local AI.
Everything runs locally by default.
No accounts.
No subscriptions.
No telemetry.
No vendor lock-in.
No data leaves your computer unless you explicitly connect a cloud provider.
Modern AI development is fragmented.
You download models using one application.
Run them from another.
Fine-tune using command-line tools.
Benchmark somewhere else.
Perform security testing using different scripts.
Manage datasets manually.
Track experiments in spreadsheets.
RedForge brings the entire Local AI workflow into one application.
From downloading a model to training, benchmarking, evaluating, securing, and exporting it—everything happens inside one professional workspace.
Your workspace for Local AI projects.
Organize everything related to an AI project in one place.
- Projects
- Models
- Datasets
- Experiments
- Benchmarks
- Reports
- Evaluations
- Training Runs
- Security Assessments
Browse, download and manage AI models directly inside RedForge.
No terminal required.
Supported sources include:
- Hugging Face
- Ollama
- GGUF Models
- Future providers
Browse models by:
- Small Models
- Coding Models
- Chat Models
- Vision Models
- Embedding Models
- Fine-tuning Friendly Models
Each model displays:
- Parameters
- Download Size
- VRAM Requirements
- RAM Requirements
- Training Suitability
- Benchmark Suitability
- Recommended Hardware
One unified runtime abstraction.
Supported providers include:
- Ollama
- LM Studio
- llama.cpp
- vLLM
- OpenAI
- Anthropic
- Gemini
- Groq
- OpenRouter
Switch between providers without changing your workflow.
Experiment with prompts and models.
Features include:
- Multi-model chat
- Streaming responses
- System prompts
- Temperature
- Top-p
- Max Tokens
- Seed
- Context Management
- Conversation History
Run evaluations directly from the Playground.
Create production-ready datasets.
Supports:
- CSV
- JSON
- JSONL
- TXT
- Markdown
- DOCX
Features:
- Dataset Preview
- Quality Analysis
- Duplicate Detection
- Prompt Leakage Detection
- Language Detection
- Cleaning Pipeline
- Versioning
- Train / Validation / Test Splitting
Fine-tune local language models using LoRA / QLoRA.
Designed for beginner-friendly local fine-tuning.
Features:
- Guided Training Wizard
- Foundation Model Registry
- Hardware Compatibility Checks
- Live Progress Dashboard
- Loss Graphs
- Checkpoints
- Training Logs
- Artifact Tracking
- Training History
Current status:
🧪 Experimental
Training infrastructure continues to evolve while maintaining a production-quality user experience.
Compare models using repeatable benchmarks.
Features:
- Custom Benchmark Suites
- Performance Comparison
- Side-by-side Results
- Historical Runs
- Leaderboards
- Exportable Reports
Originally the core of RedForge.
Evaluate language models against adversarial attacks.
Includes:
- Prompt Injection
- Jailbreaks
- Roleplay
- Prompt Extraction
- RAG Attacks
- Encoding Attacks
- Multi-turn Attacks
- Policy Evasion
- Custom Attack Suites
Generate comprehensive security reports.
Automatically evaluate AI models.
Pipeline:
Profile
↓
Plan
↓
Execute
↓
Judge
↓
Analyze
↓
Report
Outputs include:
- Accuracy
- Robustness
- Security
- Reliability
- Performance
- Overall Score
Generate professional reports.
Format:
- Markdown
- JSON
Share benchmark and security results easily.
Track every generated artifact.
Including:
- Checkpoints
- Reports
- Benchmarks
- Training Runs
- Logs
- Datasets
- Evaluations
Every long-running operation is managed centrally.
Examples:
- Model Downloads
- Training
- Benchmarks
- Evaluations
- Dataset Imports
- Security Scans
Background execution allows you to continue working while tasks run.
Inspired by VS Code.
Quickly access every feature using your keyboard.
Examples:
> Download Model
> Train Model
> Run Benchmark
> Import Dataset
> Open Task Manager
> Generate Report
> Run Security Scan
> Restart Runtime
RedForge is built around one principle:
Your AI models belong on your machine.
By default:
- No telemetry
- No analytics
- No cloud dependency
- No accounts
- No external storage
Your projects stay on your computer.
RedForge follows a modular architecture.
Core systems include:
- Runtime Platform
- Foundation Model Registry
- Model Hub
- Dataset Platform
- Training Platform
- Benchmark Platform
- Security Platform
- Evaluation Engine
- Experiment Tracking
- Artifact Registry
- Global Task Manager
Every subsystem is designed to be replaceable and extensible.
- React
- TypeScript
- Vite
- Tailwind CSS
- shadcn/ui
- Python
- FastAPI
- SQLAlchemy
- SQLite
- Ollama
- llama.cpp
- LM Studio
- Hugging Face
- Unsloth
- Transformers
- AI Developers
- Machine Learning Engineers
- Students
- Researchers
- Security Engineers
- Prompt Engineers
- Open Source Contributors
- Local AI Enthusiasts
- AI Studio
- Model Hub
- Runtime Manager
- Playground
- Dataset Lab
- Benchmark Center
- Security Center
- Evaluation Engine
- Reports
- Global Task Manager
- Command Palette
- Improved Fine-tuning
- Distributed Training
- Better Export Pipeline
- Experiment Tracking
- Model Versioning
- AI Agents
- Workflow Automation
- Visual Pipelines
- Plugin Marketplace
- Enterprise Collaboration
Contributions, issues, and feature requests are welcome.
Feel free to fork the repository and submit a pull request.
MIT License
Built for the Local AI community ❤️