Managing three or four AI subscriptions gets old fast. ModelReins is a new open-source control plane that lets you dispatch tasks to Claude, OpenAI Codex, Ollama, or any API-compatible provider from a single dashboard, using your own API keys instead of stacking monthly plans.
The pitch is straightforward: instead of switching between tabs and copying prompts around, you define "workers" that connect to different AI backends, and ModelReins handles the routing. It scores task complexity automatically and picks the cheapest provider that can handle the job. If your primary model hits a rate limit, it fails over to the next one in line - Opus to Sonnet to a local Ollama instance, for example.
How It Actually Works
The server runs on Python with SQLite. You connect workers via an npm package (npx modelreins-worker), point them at your API keys, and start dispatching from the web dashboard. Jobs queue up, run in parallel where possible, and stream results back in real time.
The security model is the most interesting piece. ModelReins calls it "zero-knowledge architecture" - your API keys and prompts never pass through their central servers. Workers talk directly to providers using your credentials. For teams worried about data residency, that matters.
Other controls include job chaining (run task B after task A finishes), approval gates (require human sign-off before proceeding), budget caps, and a killswitch that halts all workers instantly.
Pricing and Licensing
The free tier gives you 2 workers with unlimited jobs. Pro at $29/month bumps that to 10 workers and adds chain templates and analytics. Teams pay $79/month for 50 workers and multi-user access with role-based permissions.
There is also a fully self-hosted option under the Business Source License 1.1 - free for non-competing use, with commercial licenses available.
Who This Is For
This is clearly aimed at dev teams and power users who are already juggling multiple AI providers and want centralized cost tracking and failover. If you are using one AI tool for everything, ModelReins adds complexity you do not need. But if you are running Claude for writing, Codex for code, and a local model for sensitive data, having one place to manage routing and spending is genuinely useful.
The project is early - it just launched on GitHub under mediagato/modelreins. The self-hosted angle and zero-knowledge design set it apart from other multi-model tools, but adoption will depend on whether the routing logic actually saves money versus just picking the right model yourself.