In the future, there will be many models, and providers.
How will you choose the best one?And how will you measure the usage?
1Router is the single, OpenAI-compatible control plane that answers both. Every model, every provider, on one bill — and you can run the whole thing on your own hardware.
The next five years look nothing like today.
Three quiet cracks are already forming under the AI stack everyone is standing on.
The provider graph keeps growing
OpenAI, Anthropic, Google, xAI, Meta, Mistral, DeepSeek, Qwen, plus a long tail of open-source labs and private endpoints. Each ships its own SDK, its own pricing, its own rate limits, and its own authentication. Your integration surface area grows linearly with the number of vendors you adopt — and so does the maintenance burden.
Prices move independently, every week
Provider pricing is now a moving target — per-million-token rates, batch discounts, prompt-cache pricing, regional surcharges, fine-tuning markups. Without a single source of truth you can't tell which model is actually cheaper for your workload this month, much less audit last quarter's bill.
Nobody can tell you what happens to your data
Logging, retention, and training-on-prompts policies differ by provider, by SKU, and sometimes by region. The same 'send a chat completion' call has wildly different downstream consequences depending on which upstream you pick. Today, picking is a guess.
1Router is the one place every model meets.
A single API. One bill. One source of truth you can deploy on your own hardware.
One interface
Every model — hosted and self-hosted — speaks a single OpenAI-compatible API. Your code doesn't change when you change providers.
Pay-as-you-go, forever
No subscriptions, no monthly minimum. Self-host for free; use our cloud for a transparent 5% markup on provider token costs.
Self-hostable to the metal
The entire control plane is source-available. Deploy on Kubernetes, Docker, or your laptop. Your API keys and usage data never leave your network.
Open by default
FSL-1.1 today, Apache 2.0 two years after every release — irrevocably. Use it, fork it, ship it.
A whole category of work disappears.
You're not buying a model. You're buying back the time and infrastructure that vendor sprawl took from you.
Switching is one line of code
Change the model id, not the SDK. Move a workload from GPT-4o to Claude Sonnet to a self-hosted Llama without rewriting a single call.
Side-by-side, on real numbers
Pricing, context length, capabilities, modality, free endpoints and downloadable weights — laid out so an actual decision can be made in thirty seconds.
One usage ledger
Tokens in, tokens out, cost per call, cost per feature — surfaced in seconds. Stop reconciling per-provider CSVs at the end of every month.
Provable data sovereignty
Every upstream you pick publishes a data-handling attestation back to 1Router. You can see what happens to your prompts before you send them.
Where we are on the journey.
1Router ships in public, every step of the way. Here's the surface area — then the engine underneath it.
- 01Today
The catalog is live
Browse, filter, and compare every supported model. Provider detail pages, capability filtering, free-endpoint callouts, and downloadable-weight badges are all shipped.
- 02Now
Multi-model chat and rankings
Use the chat surface to drive any model directly. Rankings surface what's actually getting used across 1Router — not what's most marketed.
- 03Next
OpenAI-compatible proxy
Replace your existing OpenAI/ Anthropic base URL with 1Router. Routing, failover and load balancing happen behind a stable, drop-in interface.
- 04Soon
Data-handling attestations
Per-provider no-log, retain-prompts, train-on-inputs and eval-share attestation flags flow from the shared catalog into the model pages and the proxy itself.
- 05Later
Provider pool, auth, metering
Federated credentials with rotation, API-key auth, per-key rate limits and spend caps, Postgres-backed usage metering — all self-hostable.
Pick the model. We'll handle the rest.
One API, every provider, total sovereignty. Start in the browser, or take the whole control plane home.
Open-source · FSL-1.1, irrevocable Apache 2.0 in two years.