Covibe sits between your apps, agents and upstream model accounts, including OpenAI, Gemini and Grok. Issue keys, cap spend, route by policy and keep a clean log of every request.
import OpenAI from "openai"; const client = new OpenAI({ apiKey: process.env.COVIBE_KEY, baseURL: "https://covibe.me/v1" }); await client.chat.completions.create({ model: "coding-prod", messages: [{ role: "user", content: "Fix this test" }] });
Shared keys, agent runs and provider limits make it hard to know who used what, where it went and what it cost.
Covibe maps each model alias to an allowed provider pool after checking budget, health and fallback rules.
Send routine traffic to cheaper pools and reserve premium routes for keys that are allowed to use them.
Switch providers on 429/5xx and keep the reason, quota check and request log attached.
Each downstream key has an owner, allowed pools, spend cap, rate limit, expiry and revoke path.
See the key, model alias, selected pool, cost estimate and policy result for each request.
Start self-hosted for private testing. Move to managed ops when the gateway becomes part of daily work.
Keep keys, provider accounts and logs under your control. Expose HTTPS, not raw upstream ports.
Give each team its own keys, limits and allowed provider pools. Revoke access without touching upstream accounts.
Hosted gateway operations, updates, dashboard access and basic alerts.