LeanCTX | OSS

LeanCTX is the Context Gateway for AI agents. Control what your AI can see.

The open-source Engine brings context selection, supported access and content controls, local knowledge and observable evidence to existing AI tools through CLI, MCP, hooks and supported proxy paths. Your host owns the task, agent loop and model.

Install and connect an agent

cargo install lean-ctx --locked
lean-ctx wrap codex     # or: wrap claude / wrap cursor
lean-ctx doctor

Select · Control · Learn · Prove

Select: search and read the detail a task needs, with compact views and source references for recovering more.
Control: apply configured path permissions, content filters and budgets on supported integration paths.
Learn: retain local task, finding and decision records and connect code and knowledge across sessions. This does not train the model.
Prove: inspect the source, operation, policy and usage evidence the integration can observe with lean-ctx inspect.

A context-only integration records its prepared result; the host owns the subsequent model call. Filters have configured coverage limits. Recovery depends on current permissions, retention and source or archive availability. Usage estimates and calculated costs are separate from answer quality.

Build from source

From the repository's rust/ directory:

cargo check --locked
cargo fast

Development builds use the configured jobs and shared target directory. Use cargo build --release --locked for explicit release validation. Keep the installed runtime running while building or testing; lean-ctx dev-install performs the supported atomic replacement after a successful build.

Validate a change

cargo test --lib
cargo clippy --all-features -- -D warnings
cargo fmt --check

Integration boundaries

Attach: use the installed CLI, MCP server, hooks or supported local proxy.
Embed: use the standalone LeanCTX | SDK's released lifecycle and Agent Tools contracts. Preview operations and older in-tree experiments keep their own limits.
Commercial deployments: Enterprise and SDK/OEM have separate scope and licensing; consult the applicable release and agreement.