AI Web Tools

Working with large language models means spending a lot of time on prompt design, token counting, and output formatting. These tools help with the practical side of that workflow — tokenising prompts, building prompt chains, checking for hallucinations, stress-testing safety filters, and formatting structured data for model input. Everything runs client-side, so your prompts and API keys never touch a server.

The Cross-Model Prompt Architect & Tokenizer is the starting point: it counts tokens using multiple tokenisation schemes (cl100k, p50k, o200k) so you can see exactly how many tokens a prompt will consume across GPT-4, Claude, and other models. This matters because token limits vary and costs are calculated per token — knowing your count before sending saves money and prevents truncation.

For more complex workflows, the Prompt Chain Builder lets you design multi-step chains where the output of one prompt becomes the input to the next. This is useful for tasks like research summarisation (search → extract → summarise → translate) or content generation (outline → draft → edit → format). Each step can specify the model and temperature, and the chain can be exported as a runnable script.

The AI Hallucination & Fact-Check Guard takes a model's output and identifies claims that should be verified against external sources, flagging statements that are likely factual (and thus checkable) versus opinion or speculation. The System Prompt Safety Stress-Tester systematically probes a system prompt for injection vulnerabilities by testing adversarial inputs that attempt to override instructions.