100 LLM workflow tools for embeddings, eval, and rate limits

Toolcore added 100 more client-side helpers for LLM workflow planning—embedding storage, JSONL fine-tune stats, eval overlap, multimodal byte estimates, rate-limit math, and prompt template overhead. They live under AI tools alongside the earlier token and chunk batch. None call a generative API.

Embeddings & vectors

Size float32/float16 indexes, estimate pairwise matrix memory, or compare text overlap before paying for embeddings—start with embedding dim byte size, embedding storage GB, and cosine-like word overlap.

Fine-tune JSONL

Count valid lines, sum rough tokens, split train/val, or find duplicate keys—try JSONL line count, JSONL total token sum, and JSONL train val split.

Eval metrics

Compare response length, bigram/trigram overlap, rough BLEU/ROUGE-L, or readability on pasted outputs—see eval response length compare, eval bigram overlap, and eval BLEU rough.

Multimodal planning

Decode base64 image bytes, estimate audio minutes, or budget PDF/OCR text—open base64 image byte estimate, audio minute estimate, and PDF char heuristic.

Rate limits & batching

Plan RPM/TPM throughput, queue wait, throttle delay, and job duration—use RPM budget planner, TPM budget planner, and throttle delay calc.

Prompt templates

Count few-shot blocks, measure XML/JSON wrapper overhead, or preview filled templates—examples: few-shot example counter, XML wrapper overhead, and template fill preview.

Pair with Batch AI-DP for token stats and chunk math. Redact secrets with paste redact before sharing exports.