sail-proxy-rs

An OpenAI Chat Completions proxy for Sail Research flex mode.

Sail's flex window only runs as a background job on the Responses API, which existing tooling can't use. This proxy accepts a normal /v1/chat/completions request, submits it to Sail as a flex background job, polls until it finishes, and streams the result back. Tool calls are supported.

Install

Prebuilt binaries for Linux (amd64, arm64) are published on GitHub Releases.

curl -fsSL https://github.com/alshdavid/sail-proxy-rs/releases/latest/download/sail-proxy-rs-linux-amd64.tar.gz | tar -xz
./sail-proxy-rs

Or build from source with cargo run.

Run

SAIL_API_KEY=... sail-proxy-rs

Variable	Default
SAIL_API_KEY	—	Required.
SAIL_API_BASE	https://api.sailresearch.com/v1	Sail API endpoint.
SAIL_PROXY_ADDRESS	0.0.0.0:9090	Address the proxy listens on.

Update

On startup the proxy checks GitHub Releases and prints a notice when a newer version is available. To update in place:

sail-proxy-rs --update

Use

Point any OpenAI-compatible client at http://localhost:9090/v1. For example, in opencode:

{
  "provider": {
    "sail": {
      "npm": "@ai-sdk/openai-compatible",
      "options": { "baseURL": "http://localhost:9090/v1" },
      "models": {
        "deepseek-flash": { "id": "deepseek-ai/DeepSeek-V4.1-Flash" }
      }
    }
  }
}

Limitations

• Only streaming requests ("stream": true) are proxied; non-streaming requests return a mock response.
• The response arrives in one go once the job finishes; keep-alives are sent while waiting.
• response_format: json_object and non-function tools are not supported by Sail.