A gateway to every major model, agents that build software, a serverless cloud for them to run on and the core infrastructure underneath, all on one balance.
Everything below is built to run together: the same login, the same credits and the same usage ledger, from the top of the stack to the metal.
Point the OpenAI SDK you already use at api.nicrron.ai and reach 65+ models from ten providers with one key and one bill.
const client = new OpenAI({ baseURL: "https://api.nicrron.ai/v1", apiKey: process.env.NICRRON_API_KEY, }); await client.chat.completions.create({ model: "nicrron/auto", messages, });
Ninoo reads your repository, edits files and runs commands from the terminal, showing every diff and command before it runs.
❯ fix the flaky retry test in billing ◆ grep(retry, tests/billing) ╰ 3 matches in tests/billing/retry.test.ts ◆ edit(tests/billing/retry.test.ts) ╰ + vi.useFakeTimers() ◆ bash(npx vitest run tests/billing) ╰ 12 passed (12) ◜ Weaving… 14s · 3.2k tokens · esc to interrupt
$ curl -fsSL https://ninoo.ai/install.sh | shThe same agent in a browser: say what you want, follow the build in a live preview and publish to your own ninoo.app address.
NiNI Labs studies the computing systems that serve AI workloads. NiNI Cloud, its first service, is serverless inference for agents: import a pipeline and let the runtime schedule, scale and optimize it.
Traditional halls were planned around general-purpose servers. Nicroo Core is designed around accelerators, the fabric between them and the cooling they need.
An agent shows the diff or the command and waits for a yes. Nothing touches disk on its own.
API calls, terminal sessions and builder turns draw from the same balance and land in the same ledger.
Routing, caching and failover decisions come from live latency and cost, not from a benchmark slide.
No. One Nicrron account covers the API, Ninoo Code and Ninoo Builder. Ninoo signs in through your browser and mints a key on that same account.
The CLI is free. You pay the per-token cost of the model calls it makes, at the same published rates as the gateway, with no markup.
The live catalog on nicrron.ai lists 65+ models across ten providers, covering chat, reasoning, embeddings, image, video and speech.
Chat prompts and completions are never stored by the gateway. Generated media is kept 30 days so you can re-download it.
Not yet. It is in development at NiNI Labs and taking waitlist signups by email. Pricing will be published at launch.
AI data centers built for AI: accelerator-first halls with high-bandwidth fabric and liquid cooling. It is in the design and partnership phase.