Play the video in the post belowcodejunkie99/keel on GitHub
- Stars
- 41
- Forks
- 4
- Language
- Rust
- License
- MIT
- Last push
- 23 Sep 2026
Post on X
- Views
- 116,738
- Likes
- 699
- Reposts
- 34
Read from GitHub and X on . Counts change daily.
Reported by the author
- Cost/time cut
- ~80%
in my workflow, this cut costs and time by ~80%
x.com
Keel is a local-first macOS coding workspace built with Rust and GPUI. For a fresh, unpinned task, it lets a local Laya selector, running through Core ML, choose from host-prepared candidate routes, or lets an opt-in hosted Jev call make the same kind of choice; Keel then validates the selected route before applying it. Pinned routes and live or resumed sessions keep their existing route instead of triggering a new selection.
The workspace also uses bounded tool-focus selection inside an embedded DeepSeek loop, and a computer-use decide command that selects a prepared action ID without performing the click or keystroke itself. The README is explicit that decision records support evaluation but automatic training is not implemented, and that build checks do not by themselves establish better coding outcomes.
The author reports that this jev-plus-Opus workflow, where Opus handles the harder reasoning and Jev or Laya picks from the options the harness prepares, cut their own costs and time by about 80%.
How it uses Jev
pub const MODEL: &str = "jev-1.13.0";
const ENDPOINT: &str = "https://api.typesafe.ai/v1/systemone";
const LEGACY_SECRET: &str = ".codex/codex-router/typesafe-api-key.secret";
const MAX_RESPONSE_BYTES: usize = 64 * 1024;
/// The host decides which credential sources this installation is allowed to use.
/// `app_owned` makes onboarding Skip genuinely leave Jev disabled.
#[derive(Clone, Debug)]View in crates/jev-core/src/lib.rs
