October 2026 Summaries

Ollama Cloud has been integrated with Fly.io Sprites, persistent Linux virtual-machine sandboxes that support coding agents such as Claude Code, Codex, Gemini CLI, and Cursor. Through the Ollama Connector, agents running inside authorized Sprites can access cloud models including kimi-k3, glm-5.3, and deepseek-v4.1-flash without receiving the user’s Ollama API key, which Fly.io stores encrypted and injects through a gateway after validating the calling Sprite. Access is blocked by default and can be granted based on Sprite labels, name prefixes, or all Sprites, while requests are limited to the Sprite environment. Users configure the connector in the Fly.io dashboard, test connectivity, and can call Ollama through OpenAI-compatible SDKs or native Ollama API routes using a gateway URL and connection ID. Fly.io also provides built-in agent guidance and highlights uses such as multi-model fuzz testing, in which smaller models generate malformed inputs and larger models evaluate resulting failures.