LeanCTX
Local Rust binary that sits between your agent and the model. It compresses what the agent reads and shell output (60 to 90 percent fewer tokens), caches re-reads at about 13 tokens and keeps a signed ledger of what gets saved.
Strengths
- 60 to 90 percent fewer tokens on reads and shell output
- Local-first: Rust binary on your machine, nothing transits a third-party service
- Caches re-reads at about 13 tokens, the agent does not re-pay for a file it already saw
Limitations
- An extra layer to wire in, one more friction point in the pipeline
- Recent, the real gain depends heavily on the share of reads in your session
Best for
- Long agentic sessions where repeated reads dominate the token budget
- Sensitive contexts where code must not transit through an external service