Product Kit · 1.0.0
Context Budget Planning Kit
Allocate whole submitted content records against an explicit context budget and estimate cost without inspecting, reflecting, ranking, or rewriting content.
Agent builders planning bounded context assembly and request-volume cost from caller-authored priority, order, reserve, token-ratio, and pricing assumptions.
Live metadata-only envelope
Not generated. Build a new local context budget plan.
Exact inventory, explicit estimates
Record sizes are exact UTF-8 bytes. Token values are labeled estimates computed from Unicode code points and the caller’s characters-per-token assumption. Input USD estimates price record estimates plus reserved system and other tokens; reserved output tokens use the submitted output price. These estimates are not provider quotes.
Whole records only
The whole-record greedy policy uses caller-authored priority descending, then order ascending, and continues after an omission so a later smaller record may fit. There is no semantic ranking, no deduplication, no summarization, and no truncation.
Metadata-only privacy boundary
The omission receipt, assembly plan, live output, copy, and JSON download contain metadata only. There is no content reflection, no model, no tokenizer, no provider call, no network, and no storage. Browser-local input is not uploaded.
Planning limits
- Token values are estimates from ceil(Unicode code points / submitted chars_per_token); no tokenizer, provider quote, model preset, semantic score, summary, embedding, retrieval, or deduplication is used. UTF-8 byte values alone are exact.
- Allocation ranks only submitted priority descending, order ascending, then input index. Records remain whole; an oversized record is omitted and scanning continues so a later smaller record may fit.
- Prices and request volume are submitted assumptions. USD results are labeled estimates and rounded to 12 decimal places. Signed deficits are preserved; requests=0 remains zero.
- Opaque content is accepted but never appears in any success, failure, receipt, plan, report, or download. IDs remain strict safe identifiers. Hosted requests and complete responses are each capped at 262,144 UTF-8 bytes.
Free runnable capability
Plan one bounded context allocation locally with metadata-only output and no signup.
Builder capability
The existing Builder Pilot can coordinate higher bounded planning volume; it does not add model calls, semantic ranking, tokenization, storage, content reflection, execution, or pipeline support.
See the existing Builder PilotPrivacy boundary
Browser planning runs locally. API and MCP are response-only and return metadata only; submitted content may contain URLs, paths, logs, or credential-like material but is never reflected, executed, stored, semantically inspected, or sent to a model or provider.