Cutok
demo · prompt compressionPaste a verbose prompt and compress it before sending it to an LLM. This page demos one feature — the full engine does a lot more, shown below.
Save more with the library & MCP
Prompt compression is one lever. Wrap your LLM calls with the Python library, or expose the engine to agents over MCP, to also get:
Attachment optimization
Minify JSON/CSV, strip code comments (AST-safe for Python), and extract then compress text from PDF, Word, and more.
JSON 33% · code 24% · prose 22%
Repeated-context dedup & delta
Re-sent context collapses to a reference across agent turns, and sections shared between documents are deduped.
up to ~99% on re-sends
Cache optimization
Reorders the payload for prefix stability and injects cache-control markers so provider prompt caches keep hitting.
measured from real usage
Response budgeting
Injects a provider-correct max_tokens cap and an optional brevity directive so replies stay tight.
caps the reply
Get the full engine
pip install cutokuv run cutok mcp