Launch Posts

September 20, 2026 ยท View on GitHub

These drafts describe the current v0.1.0 release. Update the release number if they are reused for a later version.

Facebook

I just released JEV Control Plane, an open source decision and routing layer for AI coding agents, including Codex and Claude Code.

The idea is simple. You should not have to memorize which model and reasoning level is best for every kind of task. Describe the work normally, and JEV Control Plane evaluates the request before recommending how your coding agent should handle it.

The biggest advantage is that the routing decision does not consume the coding agent's reasoning tokens. JEV does not generate text or a chain of thought. It reads the task and returns typed decisions with probabilities, leaving the coding model's reasoning budget available for the actual work.

The current release includes packaged automatic hooks for Codex. Claude Code can use the same local decision engine and JSON router contract through its command and hook workflows. A dedicated Claude Code adapter is also planned to make setup and automatic runtime application even easier.

It considers the complexity of the work, the project scope, repository identity, production risk, reversibility, and whether the task belongs in a new thread or an isolated worktree.

It currently includes:

  1. Automatic model and reasoning decisions for coding agent tasks
  2. Different runtime profiles for quick decisions, everyday builds, complex development, and critical reviews
  3. Project manifests for goals, scope, exclusions, environment, and production policy
  4. Safety rules for production, billing, authentication, permissions, database changes, secrets, integrations, and other consequential work
  5. Thread and worktree recommendations
  6. Repository mismatch protection and external write blocking
  7. A local fallback when the JEV router is unavailable
  8. Secret redaction, private authenticated routing, and privacy conscious audit logs
  9. Automated tests across macOS and Ubuntu
  10. Full open source documentation under the Apache License 2.0

JEV still reads input tokens, and the selected development model uses the normal allowance or billing for its host once the actual work begins. The important part is that the routing decision itself does not spend the coding agent's reasoning tokens.

This is the first public release, and more updates are coming. I am planning more routing profiles, easier setup, richer visibility into decisions, and broader host support.

Repository: https://github.com/JxWayne890/jev-control-plane

Release: https://github.com/JxWayne890/jev-control-plane/releases/tag/v0.1.0

If you try it, I would genuinely like to hear what worked, what did not, and what you want it to route next.

LinkedIn

I have released JEV Control Plane, an open source decision and routing layer for AI coding agents, including Codex and Claude Code.

Developers should not need to memorize an entire model lineup before assigning a task. JEV Control Plane evaluates the request, project context, and operational risk, then recommends an appropriate runtime profile before work begins.

The routing decision does not consume the coding agent's reasoning tokens. JEV is a System One decision model that returns typed decisions with probabilities instead of generating text or an autoregressive chain of thought. This preserves the coding model's reasoning budget for implementation, debugging, review, and verification.

The current release includes packaged automatic hooks for Codex. Claude Code can use the same local decision engine and JSON router contract through its command and hook workflows, with a dedicated adapter planned for a more direct installation experience.

The first release includes:

  1. Automatic model and reasoning decisions for coding agents
  2. Runtime profiles for rapid decisions, balanced implementation, complex development, and critical review
  3. Scope classification based on a project manifest
  4. Repository, branch, environment, and worktree awareness
  5. Deterministic safety floors for production, billing, authentication, permissions, migrations, secrets, APIs, and integrations
  6. Confirmation requirements and external write controls for consequential actions
  7. A labeled local fallback when the JEV service is unavailable
  8. Credential redaction and privacy conscious audit logs
  9. Configurable model mappings and a deployable private router example
  10. Automated testing on macOS and Ubuntu

JEV still reads input tokens, and the configured router may have its own provider cost. The selected development model also uses the normal allowance or billing for its host once work begins. The value is that model selection and routing happen without spending the coding agent's reasoning tokens, with consistent project and safety controls applied every time.

JEV Control Plane is available now under the Apache License 2.0. More routing profiles, easier setup, richer observability, and broader host support are planned.

Repository: https://github.com/JxWayne890/jev-control-plane

Release: https://github.com/JxWayne890/jev-control-plane/releases/tag/v0.1.0

Feedback and contributions are welcome.

#Codex #ClaudeCode #OpenSource #AIEngineering #DeveloperTools #AgenticAI