jev-codex-router vs Roo Code

Roo Code and jev-codex-router operate at completely different layers of the AI coding stack. Roo Code is a forthcoming VS Code extension that plans to orchestrate multiple specialized AI agents through a natural language interface: writing functions, reviewing and optimizing existing code, editing files, controlling the integrated terminal, and using Git integration for version-aware assistance. It emphasizes local processing for privacy and customizable model deployment, with an autonomous mode marked coming soon. Its listed capabilities describe an end-user coding environment—not routing infrastructure. jev-codex-router, by contrast, is not a coding assistant itself. It is an open-source routing layer that sits between Codex and the backend, making a model-plus-thinking-effort decision per model call rather than once per session. Its distinct mechanics include a four-question choice per request (Astra policy, capability tier, effort depth, route lease), a native tier ladder Luna → Terra → Sol → Astra, mandatory Astra policy for architecture and risk review, routing applied to continuations after tool calls, Responses-in/Responses-out SSE relayed verbatim, two independent projections (bounded decision state for Jev, full canonical replay for the model), fail-open behavior, a sentinel-file kill switch, quota fallback on observed native quota exhaustion, and a local decision log per routed turn at ~/.codex/codex-router/jev-router-live.jsonl that is never published. The overlap is eligibility, not capability.

The two products are not priced for the same buyer and cannot be meaningfully compared on cost. Roo Code lists contact pricing only—no public tiers, no free plan, no SLA, and no downloadable extension. Because the product is pre-launch, there is no way to know today what you would pay, what limits apply, or what support you would receive; the only conversion path on the site is an email signup and contact form. jev-codex-router is free and MIT-licensed, distributed as a self-contained monorepo on GitHub at 0xNatoshi/jev-codex-router. Its cost is operational rather than financial: you install from a GitHub README and install.sh, run a local server component, and maintain the tooling yourself. It explicitly targets the downstream cost of Codex quota burn by selecting the least expensive configuration with sufficient capability per call, but it does not publish a live measured quota-savings number before adoption. Neither product offers standard SaaS billing, transparent plan tiers, vendor SLAs, or compliance certifications, so a procurement-minded buyer will find no apples-to-apples budget line here.