How It Works

How LeanCTX controls what your AI sees.

LeanCTX is the context engineering layer for AI agents: one small Rust binary that runs the same loop on every read, command and search. It decides what your agents read, remembers what they learn across sessions, guards what they touch, proves what they save in a signed ledger, and replays what they saw. Reads are parsed for structure across 26 languages and stripped of noise, so what reaches the model is intentional, compressed, and provable. You install it once. Everything else is automatic.

The context pipeline.

Every request flows through four stages. Your tools stay the same — LeanCTX intercepts silently between your AI tool and the LLM.

Intercept

Your agent calls ctx_read, ctx_shell, or ctx_search. LeanCTX intercepts the call before it reaches the filesystem.

Analyze

Tree-sitter parses the AST. Shell patterns match the command. The content-addressed cache checks for re-reads.

Deliver

Only the signal reaches the model. Signatures, errors, relevant matches — structured, deduplicated, compressed.

One Rust binary. Zero config. Your agent gets better context without knowing LeanCTX exists.

What your agent actually sees.

When your agent reads a file, LeanCTX returns only the structural signature — function signatures, type definitions, and key logic. The full file never enters context.

Explore read modes

Shell output, compressed.

A raw git diff produces hundreds of lines. LeanCTX pattern-matches 95+ commands and returns only the meaningful delta — file names, hunks, and error lines.

See all patterns

Security built into every call.

Before any file read or shell command executes, LeanCTX validates the path against PathJail, checks the shell allowlist, and redacts detected secrets. lean-ctx doctor shows the full posture.

Security docs

How it works, answered.

The practical details behind the compression layer.

See itin action.

Install LeanCTX and watch your agent's context usage drop.

curl -fsSL https://leanctx.com/install.sh | sh
Support this project