Skip to content
IGRISGitHub

Agents build. Your shell decides.

A builder agent writes it, a reviewer checks it, your tests decide.

01.What IGRIS is

A pipeline, not another agent. You ask Hermes in a chat app. A decision layer picks the agent and the model. Herdr runs them in panes. Your tests, not the agents, decide when it is done.

Nothing runs on a server of ours. There is no server.

02.How a task runs

The decision layer is a switch: plain rules, Laya on your machine, or Jev hosted. A builder agent works in its own worktree and a reviewer reads the result. Then your check command runs.

$ igris run --task "add a retry to the fetch helper"

Exit code 0 means done. Exit code 3 means the task needs you. See it run.

03.Set it up

Run igris setup for a local page that checks your machine and tells you, in plain words, what is ready and what needs a step. It binds to 127.0.0.1 and stops after 30 minutes without use. Prefer to do it by hand? The docs list every step in order.

04.Pick your agents

Claude Code builds and Codex reviews by default. IGRIS knows 24 kinds of agent that Herdr can run, so you can swap either one in igris.toml.

05.What leaves your machine

IGRIS has no telemetry. Only what your own tools send to their providers leaves. With Jev, the task text goes to TypeSafe. With a cloud model behind Hermes, your messages go to that provider.

06.Where it stands

Detection and the local setup page work today. The pipeline editor and the first full release are next. Version 0.0.0 on crates.io only reserves the name.