First impressions: Unsloth just destroyed LMStudio, Ollama, Open WebUI, and Lemonade

Unsloth Desktop App: First Impressions Review

This video presents a first-impressions review of the newly released Unsloth desktop application (currently in beta). The creator, who typically covers meta-analysis and systematic review topics, evaluates the app as a potential replacement for their existing local AI toolset, including LM Studio, Ollama, and Open Web UI.

Key Features Highlighted

Folder-Based RAG Integration

• Users can link a local folder containing PDFs directly to a project, and the app automatically embeds and syncs all documents
• Adding or removing files from the folder updates the knowledge base instantly—a feature the creator had previously struggled to get working with Open Web UI

Model Hub and Context Management

• The model discovery interface closely resembles LM Studio, allowing users to browse and download models from Hugging Face
• Context window settings can be adjusted to maximum capacity with an auto-fit option, eliminating the guesswork required in other tools
• The app defaults to Unsloth-optimized models but supports searching for any model on Hugging Face

Deep Research Capability

• Generates a multi-step research plan (e.g., 12 steps) that users can review and edit before execution
• The model actually incorporates user edits into the final plan
• On the creator's 8GB VRAM system, a full research run takes roughly 20 minutes
• Approximately 1 in 5 runs encounters a tool-call error, but a simple retry typically resolves the issue

Additional Tools and Integrations

• Includes image, video, audio, and recipe generation capabilities
• The recipe feature is particularly useful for creating supervised fine-tuning datasets
• Built-in model training GUI for local fine-tuning
• Exposes an API similar to LM Studio for external connections
• Supports connecting to cloud-based models (e.g., Chat LM) with response times around 3.8 seconds
• Tool permission settings allow users to choose between full autonomy or requiring approval for high-risk actions

Limitations Noted

• Connected cloud models currently lack access to the tool features (likely a beta limitation)
• The creator could not locate settings to enable tools for connected models

Installation and Availability

The app is fully open source and available on GitHub. Installation on Linux was flawless via a downloaded .deb package.

Key Takeaway

The Unsloth desktop app consolidates chat, RAG, model management, deep research, and training into a single, polished interface. Despite being in beta, it has convinced the creator to uninstall LM Studio, Ollama, and Open Web UI, positioning itself as a compelling all-in-one alternative to fragmented local AI workflows.