executainer is a Rust CLI for running agent work in isolated lanes instead of one giant shared context.
It is for the person who already knows the painful manual version:
- open another session
- restate the task
- copy just enough context over
- keep lane notes somewhere else
- hope the workers do not step on each other
executainer turns that into a repeatable workflow with visible panes, deterministic run artifacts, and a hard stop when lane output is ambiguous.
The sharpest v1 user is an engineer running long or high-risk agent work where shared context makes results worse.
Good first workflows:
- parallel repository audits
- adversarial reviews from multiple viewpoints
- migration planning with isolated read-only lanes
- implementation handoffs where writable work needs human visibility
If you want a giant all-knowing control plane, this is not that.
Most agent tools still assume the answer is more shared context.
That works until every worker sees too much, overreads, drifts, repeats work, and starts sounding confident about the wrong thing.
executainer takes the opposite bet:
- keep each lane bounded
- keep the orchestrator thin
- trust recorded evidence
- block on ambiguity instead of smoothing it over
- runs read-only delegation lanes with
codex exec - uses
tmuxas the operator visibility layer - supports explicit writable lanes for human-supervised work
- archives prompts, pane captures, outputs, and a deterministic
manifest.json - blocks synthesis when parsing fails or lanes collide on proposed files
Not in v1:
- full sandbox isolation
- automatic approval classification
- non-
tmuxbackends - hosted orchestration
Those are future directions, not hidden promises.
The full install guide lives in INSTALL.md.
Fast paths:
- Download a release binary from GitHub Releases
- Install directly from Git:
cargo install --git https://github.com/dgitco/executainer- Or install from local source:
cargo install --path .Verify the local environment:
executainer doctorExpected healthy checks:
- runtime
tmuxcodex_cli- writable temp dir
- parser self-check
Then run a real task:
executainer run --lanes 3 --task "Audit this repository for risky coupling."Inspect the result:
executainer inspect <run-slug>Clean it up:
executainer clean <run-slug> --yesSee docs/sample-run.md for a concrete walkthrough and what the output artifacts look like.
This repo is meant to be boring in the right places.
- CI runs formatting, linting, tests, and build checks
- tagged releases publish release archives and checksums
- docs call out the real support boundary instead of pretending v1 is more complete than it is
Current release target:
- macOS via release binary or source build
- Linux via release binary or source build
- local
tmux - local Codex CLI available as
codex
If your machine already has Rust, tmux, and Codex, time to first useful run should be under 5 minutes.
Common issues and operator mistakes live in docs/troubleshooting.md.
See CONTRIBUTING.md.
See SECURITY.md.