loadout
Chooses the model tier, skills and MCP servers for each coding-agent session with two Jev requests, then launches Claude Code or Codex.
Overview
loadout picks the model, skills and MCP servers for each Claude Code or Codex session from a library of any size, for about a tenth of a cent and a second per session according to the README. Local retrieval with keywords and embeddings narrows a library (1,262 skills in its tests) to 200 candidates. Then two Jev requests run: a wide one ranks the candidates and asks whether the request needs a specialised skill at all, and a close one grades the top eight.
Code then loads the best fits (at most three), offers other candidates with fit of at least 0.7 as optional, injects a skill's full instructions when you name it or Jev finds a clear winner, or stays quiet when general coding is enough. It also picks a model tier with probabilities. loadout launches the agent with that selection using official CLI flags and never proxies API traffic. Its benchmark used realistic requests written by models that never saw the library, judged by Claude Opus against the whole catalogue, with the rule tuned on a third of requests and reported on the rest.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this request need a specialised skill at all? | — |
| Score | How well does each candidate skill fit the request (0-1)? | — |
| Choice | Which model tier should run this session? | small · medium · large |
How it works
- User describes the task
- Local retrieval narrows the library to 200 candidates
- Wide Jev request ranks candidates and checks need
- Close Jev request grades the top 8
- Load, offer or inject skills and launch the agent with chosen tier
Strengths
- Scales to large skill libraries
- Low cost per session
- Never proxies agent traffic
- Held-out evaluation
Caveats
- Judged by another model, not humans
- Requires a Jev key
- Library-specific tuning
Getting started
loadout route "<your task>" to see the selection, then launch Claude Code or Codex through loadout.At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- CLI launcher for Claude Code and Codex
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Pick the model, skills and MCP servers for each coding-agent session from a…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide?
Whether a specialised skill is needed, how well each candidate fits, and which model tier to use.
Does it sit between me and the model?
No, it launches the agent with official CLI flags and never proxies API traffic.
What does it cost?
About $0.0011 per routing decision on the README's benchmark.
Related projects
Orchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.
Agent skillbetaPythonHooks for Claude Code, Codex and pi that use Jev to suggest a skill each turn, plus a rule gate for subagent model routing.
HookbetaJavaScriptClaude Code plugin that asks Jev before each turn for reasoning effort, subagent model, strategy and skill, with a decision log.
Agent skillbetaTypeScriptExperimental plugin wiring Jev into coding-agent harnesses to pick skills, tools, models and browser moves, starting with OpenCode.
Agent skillprototypeTypeScriptPer-turn router: one ~350 ms Jev call picks model tier, effort, tools and skill, with a hard deadline and regex fallback.
Agent skillbetaTypeScriptFourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.
ProjectbetaPython