Tersely — LLM token and context optimizer

Configure

Content type Target model Compression Quality need Total token budget Output length

Task / retrieval query Needed for relevance pruning and RAG selection

Optimization layers

0 tokens

Original context

Optimized package0 tokens

From word trimming to context compilation.

Lexical + semantic dedupe

Condenses wordy phrasing and catches similar repeated instructions while protecting negations.

Relevance retrieval

Scores paragraphs against the current task and sends only the strongest context blocks.

Rolling conversation memory

Turns older dialogue into durable project state while retaining recent turns verbatim.

Code and HTML cleanup

Removes webpage chrome, scripts, dividers and low-value comments without pretending to minify executable code.

Budget compiler

Reserves output space, fits input to the remaining budget and protects constraints.

Cost strategy

Creates a cache-friendly prefix, suggests a model tier, estimates cost and caps output.