Repository navigation
First public release
nutsheLLM is a quality-aware context optimization workbench for AI applications and agents. It compresses eligible context with PariTok, preserves exact facts, validates task outcomes, and automatically falls back to safer context when compression is not acceptable.
Measured live result
- 52.2% median input-token savings
- 3/3 deterministic scenario validations passed (100%)
- 2,883 of 6,880 input tokens saved in aggregate (41.9%)
- the checkout safety case honestly reports 0% savings because its compressed result was not accepted
Included
- FastAPI backend and React workbench
- Docker and Railway production configuration
- curated benchmark scenarios and redacted evidence
- Python, frontend, and Docker CI
- documentation, submission assets, Nuto mascot, and Apache-2.0 license
Live demo: https://nutshellm-production.up.railway.app