NewSphere Bot is in development

Meet Sphere Bot. AI that finishes the task, and speaks isiZulu, Sesotho, isiXhosa, Afrikaans, Setswana.

Not a chatbot that talks about the work. Sphere Bot has a computer of its own, with its own desktop and browser, where tasks run start to finish. Close the app and it carries on; come back and it is done. It remembers what you told it, and does it all out loud, without touching the keyboard.

Month-end

Finance

0:25

Sphere Bot, mid-task. Search the agents, switch between them, or type it a message.Explore Sphere Bot

AsyncRouter · Coming soon

One endpoint. Every model, every provider.

Your application calls one endpoint with one key. AsyncRouter picks a healthy provider for the model you asked for, fails over when one degrades, and shows you what each call cost.

  • Route by price, latency, or availability
  • Automatic failover when a provider degrades
  • Per-call cost and latency, visible in one place

GPT-6 Astra · Claude Opus 5 · Grok 4.6 · Gemini 3 Pro · GPT-5.6 Luna · Llama 3.3 70B · Mistral Small 3 · Qwen2.5 72B · DeepSeek V3 · Qwen3 32B · Gemma 3 27B

Optimise for
HostLatencyThroughput$/1MUptime
openairouted420ms48 t/s$50.0099.90%
azure-openai460ms44 t/s$50.0099.95%
aws-bedrock510ms40 t/s$52.0099.80%
Request
curl https://api.asyncrouter.co.za/v1/chat/completions \
 -H "Authorization: Bearer $ASYNC_API_KEY" \
 -d '{"model":"gpt-6-astra","route":"price"}'

Illustrative figures. Providers are the routing targets.

Try it: pick a model, change what to optimise for, watch where the call routes.

Providers of every model

  • OpenAI
  • AWS
  • Anthropic
  • Microsoft Azure
  • xAI
  • Meta
  • Google Cloud
  • Mistral AI
  • DeepSeek
  • Qwen
  • Nvidia
  • Hugging Face
  • Perplexity
  • Alibaba Cloud
  • Moonshot AI
  • Databricks
  • Replicate

Provider marks belong to their respective owners

Want early access?

Tell us which product matters to you. We will come back when there is something to try.