OpenAI can now pin a single API request to a processing region

Swap the base URL on a Global-project key to route a call through the US, EU or UAE. Plus a prompt-cache dashboard, and Bedrock AgentCore search gains filters.

Nowline AUG 24 7:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Region pinning without region-specific projects

    Point the base URL at a prefixed domain — us.api.openai.com, eu.api.openai.com or ae.api.openai.com — on a single Global-geography project key, and that one request is processed in that region, no separate per-region project required. The US, EU/EEA-plus-Switzerland and UAE cover storage and processing; Australia, Canada, Japan, India, Singapore, South Korea and the UK are storage-only. Non-US routing needs abuse-monitoring approval and a Modified Retention amendment.

  • See whether your prompt cache is actually hitting

    OpenAI also shipped a prompt-caching dashboard that charts cache hit rate over time and cache reads per write, with token breakdowns you can filter by model and service tier. Now you can confirm your shared prefixes are being reused before spending a day refactoring prompts to chase the discount.

  • Elsewhere: Bedrock AgentCore web search learns to filter

    AWS added per-request domain include/exclude lists (up to 100 domains each) and published-date windows to Web Search on Bedrock AgentCore, and expanded it to Europe (Ireland) and Asia Pacific (Tokyo) alongside US East. An agent can now constrain its web grounding to trusted sources or a recent time range with no admin reconfiguration.

  • Build this weekend: an AI co-op partner for your game

    A developer wired a low-latency voice loop — speech-to-text, a fast LLM, then TTS — into Skyrim so an AI companion talks and reacts in real time while they play; it's on the Hacker News front page today. The pattern (read game state, stream a quick model, speak back in under a second) transfers to any moddable game you can hook.