Evaluate a math expression precisely and return the exact result. Supports + - * / % ^ (power), parentheses, and decimals, e.g. '(12.5% of 349)' should be written as '349 * 12.5 / 100', or '2 ^ 10', or '(3 + 4) * 5'.

Decode a JWT (JSON Web Token) and return its header, payload, and signature. Does NOT verify the signature — use this to inspect what's inside a token. Useful for debugging auth issues.

Compute an EXACT line-by-line diff between two texts, with true added/removed counts. Models miscount and invent changes when comparing by eye — this computes it. Use to compare two versions of text, config, or code.

Estimate the number of tokens a piece of text will consume when sent to an LLM. Uses a simple heuristic (roughly 1 token per 4 characters), not a real tokenizer. Useful for rough budgeting before sending large texts to Claude or ChatGPT.

Get the real current date and time, optionally in a specific IANA timezone (e.g. 'Asia/Kolkata', 'America/New_York', 'Europe/London'). Defaults to UTC. Use this whenever the actual current time matters — the model's own sense of time is unreliable.

Compile and ACTUALLY RUN a regular expression against test strings, returning the real matches, positions and capture groups. Models routinely guess regex behaviour wrong — this executes it. Use whenever a regex is written, debugged, or claimed to work.