Documentation
Welcome to ultFMT. This is a collection of fast, privacy-first developer utilities tailored specifically for AI engineers and researchers.
Core Principles
Zero Data Retention
We don't save your data. Ever. Files uploaded for analysis are processed entirely in-memory and immediately garbage-collected once the response is sent. There is no database storing your inputs.
No Accounts Required
ultFMT is designed for immediate use. You do not need to create an account, log in, or provide an API key. Just open the tool and get to work.
Tool Categories
ML Studio
Tools for exploring, debugging, and visualizing machine learning datasets and training runs.
- Dataset Health: Instantly check CSV datasets for missing values, class imbalances, and duplicate rows before training.
- Feature Intelligence: Calculate correlation matrices, mutual information, and PCA components to identify the strongest predictors in your dataset.
- Learning Curve Plotter: A client-side charting tool to visualize training vs. validation loss/accuracy from CSV logs to spot overfitting.
Prompt Tools
Utilities to debug and optimize complex LLM prompts.
- Token Estimator: Estimate token counts across different tokenizers (OpenAI cl100k_base, Anthropic, etc.) without making API calls.
- Prompt Diff: Visual side-by-side diff tool to track exactly what changed between two prompt versions.
- Context Checker: Verify if your prompt + document payload fits within specific model context windows.
MCP Studio
Tools for building and configuring Model Context Protocol (MCP) agents.
- Visual Builder: Drag-and-drop interface to build complex multi-agent MCP configurations.
- MCP Validator: Static analysis tool to ensure your MCP JSON configurations are valid, have correct env vars, and don't contain conflicting scopes.
Open Source
ultFMT is open source. You can view the code, run it locally, or contribute new tools on our GitHub repository.
View on GitHub