id-agent
A library that generates short, memorable word-based identifiers for AI agents to pass through prompts and logs, using fewer tokens than a UUID and configurable collision resistance.
id-agent generates identifiers meant for the context an AI agent actually works in: prompts, tool calls, logs, traces and memory, where a standard UUID is expensive to encode as tokens and easy for a model to mistype or hallucinate. Instead of hexadecimal characters, it produces IDs built from words drawn from a curated list of 4,096 entries, each of which is exactly one token under a common modern tokenizer.
Because every word contributes exactly 12 bits of entropy regardless of its length, the total entropy of an ID is simply the number of words multiplied by 12 bits, letting a developer choose how many words to use based on the collision risk they can accept, from roughly 12 bits up to 192 bits. The library validates its public inputs with descriptive error messages and is available for both JavaScript and Python.