AI & LLM Tools.

Working with large language models requires a different mental model than working with traditional software. Models do not have infinite memory, they have context windows measured in tokens. They do not produce deterministic output, they produce probabilistic completions shaped by how you frame your instructions. They have different cost structures, speed characteristics, and output quality depending on the task type. These tools are built specifically for the developers, researchers, and content creators who work with LLMs daily and need practical utilities rather than yet another chatbot interface.

Tokens are the unit of measure for everything LLM-related: input costs, output costs, context window limits, and rate limits. A token is roughly 0.75 English words, but this varies significantly by language, code, and special characters. The token counter shows the exact token count for any text using the specific tokenizer for each model family, which differs between GPT-4, Claude, Gemini, and open-source models. Knowing your token count before making an API call prevents costly surprises when a long document exceeds the model's context window.

The context window visualizer takes this further by showing how a full prompt, including system message, conversation history, and user input, fills the available context. For models with 128k or 200k token windows, this might seem irrelevant, but context window management becomes critical when building applications that maintain conversation history, retrieve documents, or process long inputs. Seeing the visual fill level helps you architect your retrieval and truncation strategies correctly.

Prompt engineering is a learnable craft, and the prompt builders help you apply proven techniques. A well-structured system prompt specifies the model's role, the output format, the constraints, and examples of the desired behavior. The system prompt builder guides you through each component and generates a complete system prompt you can use immediately. The image prompt builder applies the vocabulary and structural patterns that work well for image generation models, covering subject, style, lighting, composition, and negative prompts.

Choosing the right model for a task is a cost-quality-speed trade-off. GPT-4o, Claude Sonnet, and Gemini Pro have different strengths and price points. For a task that runs thousands of times per day, choosing a model that is twice as expensive doubles your infrastructure cost. The model comparison and LLM pricing calculator put the key numbers side by side, including context window size, input cost, output cost, and relative benchmark performance, so you can make an informed decision before committing to an architecture.

The fine-tuning formatter prepares training data in the JSONL format required by OpenAI's fine-tuning API and similar formats for other providers. Fine-tuning requires pairs of prompts and ideal completions, and the formatting requirements are strict. The tool validates your data structure and exports correctly formatted files, saving the back-and-forth of debugging format errors when submitting a fine-tuning job.

AI content detection tools attempt to identify whether text was written by a human or generated by an AI. These tools work by analysing statistical patterns in word choice, sentence structure, and burstiness, which differs systematically between human writing and LLM output. The results are probabilistic, not definitive, but they give a useful signal when you need to evaluate submitted content for authenticity.