Ollama announced on Oct 13, 2026 that its cloud models can be accessed from Fly.io Sprites – isolated Linux VMs with persistent disks. A new Ollama Connector lets a Sprite call models such as kimi‑k3, glm‑5.3 or deepseek‑v4.1‑flash while Fly.io stores the API key encrypted and adds it at the gateway. Agents inside a Sprite cannot see the key, and access is blocked by default unless you label a Sprite to allow it. The setup works with any OpenAI‑compatible SDK by pointing to the gateway URL.
Why it matters
Developers can run AI agents in a sandboxed environment without risking their Ollama API keys.
