LLM planning without an API
August 31, 2026
Most LLM prep work is arithmetic and formatting—you do not need a model to guess whether a paste fits a window or what a request might cost. Toolcore's AI hub keeps these helpers in the browser under Client-side — LLM workflow.
Start with size and cost
- Token estimate — rough chars and tokens from any paste (CJK-aware heuristic).
- Context window fit — compare text against 4K–1M presets and see overflow.
- API cost calculator — multiply token counts by your own per-million rates.
- Character budget — reverse direction: pick a token cap, get a character ceiling.
Structured conversations
- Messages JSON breakdown — per-message table from chat API JSON.
- Chat turn breakdown — User:/Assistant: plain-text exports split by turn.
- Context split & budget — delimiter-separated blocks with share percentages.
Trim and split before paste
- Whitespace savings — normalize spacing and see token delta (wording unchanged).
- Long text chunker — numbered slices under a character cap for multi-message chats.
- Paste redact — mask secrets before anything leaves your machine.
None of these pages call a generative API. Counts are planning hints—not billing-grade tokenizers. When you need explanations or rewrites, use the server-assisted section on the same hub and redact first.
Common use cases
- Pick the right browser-only page before pasting a large prompt or chat export elsewhere.
- Share with teammates who need cost and context planning without touching an API key.
- Pair with redaction helpers when logs or tickets might contain secrets.
Common mistakes to avoid
Using rough token counts for invoices
Provider tokenizers differ. Use these pages for planning, not exact billing.
Skipping redaction because processing is local
Local does not mean safe to save or share—mask secrets before you paste anywhere.
FAQ
Do these tools call a language model?
No. They use local heuristics and arithmetic only.
Where are the generative explainers?
On the same AI hub under server-assisted cards—they may send text off-device when configured.