agent-stack
Glue that turns one Linux machine into a team of coding agents, with Jev making typed calls like who builds a task and whether it may merge.
Overview
agent-stack turns one Linux computer into a software team of AI coding agents. You talk to an operator agent, which onboards a repository and sets up a project team (lead, architect, test authors, builders, QA, reviewers from two model families and a merge owner). The team plans, writes tests first, builds, uses the result like a person would, gets cross-family review and merges through a gate; you approve the plan and risky steps.
It is configuration around existing tools: OpenRig runs the teams, Claude Code and Codex CLI are the workspaces, CLIProxyAPI pools subscriptions, Playwright provides a browser, and TypeSafe Jev makes the small typed decisions such as who builds this and may this merge. The merge gate combines CI, other-family review, a QA ship verdict and a live Jev check. A read-only rig-console shows every team.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which agent should build this task? | — |
| Yes / no | May this pull request merge? | yes · no |
How it works
- Operator onboards a repo and creates a team
- Team plans; you approve
- Jev routes work to builders
- PR passes CI, cross-family review, QA verdict and a live Jev gate
- A fresh agent uses the deployed result
Strengths
- Jev handles small routing and gate decisions cheaply
- Human approval reserved for plans and risky steps
- Read-only console with fixture mode for trying it
Caveats
- Heavy stack of dependencies and subscriptions
- No licence stated
Getting started
Try the console with node console/src/main.ts --fixture console/fixtures/v3.json (no daemon or account needed).At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Linux host with OpenRig, Claude Code and Codex CLI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Turns one Linux machine into a team of coding agents that plans, writes tests first, builds, uses the result and has a second model family review it, with Jev making the routing decisions.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Small typed calls such as which agent builds a task and whether a PR may merge.
What else is in the merge gate?
CI, review by a second model family and a QA ship verdict.
Can I preview it without setup?
Yes, run rig-console against the bundled fixture.
Related projects
Desktop workspace for coding agents that runs tasks in parallel worktrees, listed as using Jev to pick the best agent per task.
ProjectbetaTypeScript, RustLocal Kanban workspace for Codex where Jev recommends a model and reasoning effort for each ticket before execution.
ProjectbetaTypeScriptThree Italian-language Claude Code prompts that build Jev-powered model routing, context compaction and code review into your project.
ProjectprototypeThree Spanish-language demos where Jev decides and Claude reasons: Jira triage, code pre-flight and a PR review router.
AppprototypeTypeScript- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript Open-source team agent platform for Slack, GitHub and more that asks Jev once per session which agent runtime and model to use.
ServicebetaTypeScript