Ollama Cloud models now run inside Fly.io Sprites for safer AI agents

What a Sprite is and what the connector adds

According to Ollama, a Sprite is a sandbox from Fly.io. Each one is a full Linux virtual machine with its own disk that stays around between sessions. Claude Code, Codex and Gemini CLI are supported out of the box, so agents can run in the sandbox and execute commands securely.

With the new Ollama Connector, Sprites can use Ollama's cloud models. The announcement names kimi-k3, glm-5.3 and deepseek-v4.1-flash as examples.

How the key stays protected

The user adds an Ollama API key to their Fly.io organization. Fly.io encrypts it and never returns it through the API. When code inside a Sprite calls Ollama, the request goes through a Fly.io gateway. The gateway checks which Sprite is making the call and adds the key to the request. The Sprite itself only knows a gateway URL and a connection ID.

Ollama lists several consequences of this design:

• An agent in a Sprite cannot leak the key.
• New connectors block every Sprite by default.
• Users choose which Sprites get access: by label, by name prefix or all of them.
• Calls only work from inside a Sprite.

Getting started

Users need an Ollama API key and at least one Sprite. In the Fly.io dashboard, they open Sprites, then Connectors, and add an Ollama connector with the key. Under Access Configuration they choose which Sprites can use it and save. A Test reachability button next to each Sprite checks that it can reach Ollama.

From inside the Sprite, a request goes to a gateway URL of the form https://api.sprites.dev/v1/gateway/ollama/<connection_id>/v1/chat/completions. There is no Authorization header, because the gateway adds the key. The connector page shows the command with the user's connection ID filled in.

Ollama says any OpenAI SDK works. In Python, the user sets base_url to the gateway and uses any value for api_key. Ollama's native /api/* routes also work through the gateway.

Coding agents and a fuzzing example

Fly.io Sprites come with a skill that tells Claude Code, Cursor, Codex and Gemini CLI about Ollama. If the Sprite has access, the agent can use Ollama models without any setup.

Fly.io also has a guide that combines several models for fuzzing. A small model such as nemotron-3-nano:30b writes malformed inputs, a Sprite runs them against the user's program, and a larger model such as kimi-k3 examines the crashes. Switching models only requires changing the model field.

What it means for users

For developers who let coding agents run commands, the setup separates the agent from the credential. The sandbox can reach Ollama's cloud models, but the key stays with Fly.io's gateway and access is off until the user turns it on for specific Sprites. Ollama thanked the Fly.io team for building the integration with it, and points users to the Ollama Connector quickstart on Fly.io for setup details.