Question 1

Do I need a Jev account to use the main command, or only the router?

Accepted Answer

Only the router, as documented. The main command routes on its own, while the intent-gated router requires Jev to classify the request and check that the selected route preserves the requested outcome. For it you pick a transport explicitly: a hosted gateway key, or the direct API key. The TypeSafe MCP plugin is not required, and an explicit transport never silently switches to the other.

Question 2

What stops an agent from quietly widening the task?

Accepted Answer

The gates demand evidence, and for the router, a separate judgment. The router restates the requested outcome and requires a receipt for that exact proposed intent before dispatch, recording alignment judgments locally without raw request text and keeping the judge model separate from the executing agent's. Elsewhere, claiming the code looks correct does not satisfy an exit gate that wants test output, and asserting confidence does not satisfy a gate that wants an exit code.

Question 3

Will the hooks actually fire on Codex and Reasonix?

Accepted Answer

Not everywhere, and the README quantifies it. Codex gets 29 native plus 30 adapter-backed registrations out of 70, leaving 11 unsupported, and its adapter cannot intercept writes through unified execution, unmatched MCP tools, or web search, while two lifecycle events see less telemetry than under Claude Code. Reasonix has no agent or custom-command surface at all, so the command arrives as a skill with only four events exposed and 10 allowlisted registrations mirrored.

Question 4

Is it safe to run the install script over my existing agent configs?

Accepted Answer

Plan for it rather than assume. The script installs into the Claude, Codex, Factory, Hermes, and Reasonix homes when those commands are on PATH, symlinks by default so a git pull propagates changes, and by default moves old unowned copies to the trash instead of leaving them. It offers dry-run, target, configure, no-takeover, uninstall, and rollback flags, and documents its layout rules separately.

Question 5

Which Jev transports does the /d router support, and how is one chosen?

Accepted Answer

JEV_TRANSPORT can be vercel, direct or auto. Auto, the default, prefers Vercel when AI_GATEWAY_API_KEY is set and otherwise uses the direct API when only TYPESAFE_API_KEY is set; an explicit transport never silently switches.

Question 6

What gets recorded when the router judges intent alignment?

Accepted Answer

Intent-alignment judgments go into learning.db without raw request text, and receipts keep the Jev judge model separate from the executing agent's model and effort. View proposed-intent difference rates by agent profile, judge model and transport with scripts/jev-intent-stats.py.

Question 7

How do I install only part of the toolkit, or preview the install first?

Accepted Answer

Run ./install.sh --configure, or copy .local.example/profile.yaml to .local/profile.yaml and edit it; without a profile the full toolkit installs. Use ./install.sh --dry-run to print the plan and --target to limit it to one runtime.

Vexjoy Agent

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev community