Skip to content

Squeez

Squeez Logo

Squeeze verbose LLM agent tool output down to only the relevant evidence blocks.

LLM coding agents waste 80-95% of context tokens on irrelevant tool output. Squeez prunes a single tool observation down to the minimal verbatim evidence block the agent should read next.

Quick Example

$ cat django/middleware.py | squeez "Find the referer validation block in the CSRF middleware"
class CsrfViewMiddleware(MiddlewareMixin):
    def _check_referer(self, request):
        referer = request.META.get('HTTP_REFERER')
        if referer is None:
            raise RejectRequest('No referer')
        good_referer = request.get_host()
        if not same_origin(referer, good_referer):
            raise RejectRequest('Bad referer')

From 42 lines of middleware code, Squeez extracts only the block relevant to the CSRF referer check.

Highlights

  • 92% compression, 0.86 recall — keeps only the evidence block the agent needs
  • CLI + Python API — cat file | squeez "task" or ToolOutputExtractor
  • Four backends — vLLM server, local transformers, encoder, and pooled classifier
  • Config-driven — YAML config, env vars, or CLI args
  • Agent-ready — works with Claude Code, Codex CLI, OpenCode via instruction files
  • 27 tool types — trained on real SWE-bench workflows and synthetic multi-ecosystem outputs