Jev Codex Bridge

TypeSafe Jev selects a model and reasoning effort for an optional Codex session, using your existing Codex login. Normal Codex keeps its native OpenAI provider and voice route.

Install and run

Requires Node.js 24+, npm, Git, Codex and a TypeSafe API key.

npm install --global git+https://github.com/ansidium/jev-codex-bridge.git --ignore-scripts

Create ~/.jev-router.env:

TYPESAFE_API_KEY=your-key
JEV_CODEX_INCLUDE_MODELS=gpt-6.1-terra,gpt-6.1-sol,gpt-6-astra

jev-bridge install
jev-codex
jev-codex resume --last
jev-codex exec "fix the failing test"

Default installation starts no service and does not edit config.toml. The proxy runs only while jev-codex is open. Jev Router enables routing; selecting a concrete model bypasses it. Use $jev-explain for the last decision.

Behavior

• Classifies fresh user instructions once; tools keep the selected pair.
• Uses the account's model catalog and reasoning levels, including GPT-6.1 Terra.
• Preserves the current eligible pair when TypeSafe is unavailable.
• Cancels classification with the request and pauses repeated auth, credit or rate-limit failures for 60 seconds.

TypeSafe receives routing text and model metadata. OpenAI receives the full Codex request. Routing decisions are stored locally.

Optional persistent gateway

On a new installation, jev-bridge install --desktop adds a Windows service and ~/.codex/jev.config.toml. Select the profile with codex --profile jev. The main configuration remains unchanged. Keep native OpenAI sessions for voice. On macOS and Linux add --no-service and run jev-bridge serve under a supervisor.

See configuration, routing and operations for options, updates and removal.

Development

npm ci --ignore-scripts
npm run check
npm test

Tests run without API keys. Based on Jev Router. MIT / Attribution