Organization repository on GitHub
Deterministic model routing for coding agents.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Tools & CLIs
Switchloom
Experimental deterministic model routing for coding agents with Jev in the loop: assign capabilities to models and copy one workflow prompt into Codex Desktop. The author's own Pokédex pilots did not show a cost win and he recommends against adopting it.
- Stars
- 1
- Language
- TypeScript
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- instructa
- Last verified
- Oct 2026
- Languages
- TypeScript
- Platforms
- CLI
- Capabilities
- Open sourceRuns locallyRequires API key
Highlights
- Default cards: GPT-5.6 Luna (coordination), Sol (implementation), GPT-6 Astra (planning and review), each on a persistent Codex task with its own context.
- The author recommends against it: in the second Pokédex run the Jev team cost almost as much as Astra alone and hit the time limit; Astra passed all 14 checks.
- Vanilla mode sets an explicit capability and needs no API key; Jev mode is on by default and needs a local TypeSafe key exported as TYPESAFE_API_KEY.
- Setup: cargo install --path . --locked, then copy skills/switchloom/SKILL.md into ~/.agents/skills/switchloom/.
- The website playground has a routing graph, probabilities and a 4,000-character prompt limit; 1.0 is an unpublished local preview.
Quickstart
bash
cargo install --path . --locked
mkdir -p ~/.agents/skills/switchloom
cp skills/switchloom/SKILL.md ~/.agents/skills/switchloom/SKILL.mdWatch out
Grounded on the truncated README and GitHub metadata reviewed 2026-10-15; the README is cut inside the playground section and links benchmark limitations at switchloom.ai/benchmarks/pokedex-round-2, which this digest does not include. The author explicitly recommends against adopting the routing workflow based on two pilots on one application.
Questions about this entry
Short answers drawn from the linked project or page.
Experimental deterministic model routing for coding agents: you assign capabilities to models, copy one workflow prompt into Codex Desktop with your task, and Jev optionally decides routing while Codex owns execution.
Jev mode is enabled by default and the CLI prepares routing decisions and handoff arguments while Codex creates tasks, sends messages and runs the work. Vanilla mode sets an explicit capability without calling TypeSafe and needs no key.
Not per the author's own pilot: in the second Pokédex run the Jev team cost almost as much as Astra alone and hit the time limit before final acceptance, while Astra finished and passed all 14 independent functional checks. He recommends against adopting it.
No. 1.0 is an unpublished local preview, unit, CLI and browser checks are distinct from a complete multi-task desktop run, and desktop verification plus representative Jev-versus-vanilla measurements remain open in TASKS.md.
Install the CLI with cargo install --path . --locked, copy skills/switchloom/SKILL.md to ~/.agents/skills/switchloom/, and export TYPESAFE_API_KEY for Jev mode. The copied desktop workflow needs the CLI and skill in both modes.
The default board lists GPT-5.6 Luna (max effort, coordination and mechanical tasks), GPT-5.6 Sol (medium, implementation and tests) and GPT-6 Astra (high, planning and review); you can choose another suggested model or enter a custom GPT ID.
The website includes a Jev Playground with a routing graph, probabilities and a 4,000-character prompt limit; the public version evaluates routing only and keeps the TypeSafe key on the server, while pnpm site:local enables local benchmarks.
More Jev resources like this
Write end-to-end tests for web apps in plain English; jev-e2e turns each case into a plan, has Jev pick the page controls, and runs it with Playwright, returning PASS, FAIL or BLOCKED with HTML, JSON and masked screenshot evidence.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
perixtar/jev-e2eTypeScript
perixtar/jev-e2eTypeScriptAn open-source, local-first model router for coding harnesses: Jevia asks Jev for a capability tier per task, applies your safety policy, and keeps recorded execution history in your project or database so later routes learn from outcomes.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
assistant-ui/jeviaRust
assistant-ui/jeviaRustBrowser QA where Playwright drives and films, and Jev judges: natural-language assertions, self-healing locators and goal-driven click loops, with a JSON-flow CLI for agents, a local dashboard and an agent skill.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
jonymusky/jev-browser-qaTypeScript
jonymusky/jev-browser-qaTypeScriptA Rust CLI and service for log triage at ingest: it collapses a stream into Drain templates, asks Jev a policy-driven set of typed questions once per new pattern, and routes every line to page, ticket or log.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
tpateeq/tocsinRust
tpateeq/tocsinRustA Claude Code and Codex skill with a CLI that answers narrow agent judgments - is this file relevant, which team owns this request, how severe is this - with Jev's calibrated probabilities instead of another LLM call. Zero dependencies, MIT.
Jev Agents & MCP#community#catalog-import#madewithjev
Community
heyman333/jev-routerJavaScript
heyman333/jev-routerJavaScriptModeration for communities and apps, powered by Jev: every message gets a probability for spam, scam, harassment, nsfw, doxxing and more, plus rules in plain English. Thresholds and actions are yours, and flag-only is the default.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
ohernandezdev/jevmodPython
ohernandezdev/jevmodPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.