Using Downsizing with Warp

Why there's no ds link warp

Warp added two relevant features: "Bring Your Own API Key" (BYOK), which lets you use your own Anthropic, OpenAI, or Google keys, and a "custom inference endpoint" option that can route through any OpenAI-compatible endpoint (a router like OpenRouter or LiteLLM, or your own gateway). Both are documented at docs.warp.dev/agent-platform/inference/custom-inference-endpoint and docs.warp.dev/agent-platform/inference/bring-your-own-api-key.

The catch is that both are configured entirely inside the Warp desktop app's Settings panel (search for "inference endpoint" or "API keys") — there is nosettings.json, TOML file, or environment variable Warp reads this from. Keys and endpoint URLs are stored in your OS keychain, not in a plaintext file Downsizing's CLI could write to on your behalf. Since ds linkonly works by writing and managing config files, there's nothing for it to manage here.

Warp also requires that a custom endpoint be reachable at a publicURL, since requests are proxied through Warp's own servers before reaching your endpoint — Downsizing's Inference API URLs are public, so that part is not a blocker, it's purely the lack of a file-based config surface.

Manual setup (do this in Warp's Settings)

You can still point Warp at Downsizing yourself, in about a minute:

1. In Warp, open Settings and search for inference endpoint.
2. Set the base URL to your Downsizing Inference endpoint, e.g. https://api.downsizing.dev/YOUR_INFERENCE_ID/openai/v1 (find your Inference ID on the Inference API page).
3. Enter your Downsizing API key (dws_...) as the API key for that endpoint.
4. Set the model identifier to a Downsizing model name, e.g. downsizing-claude-opus-4-8.
5. Save, then select that model from Warp's model picker.

This uses the OpenAI-compatible surface of the Downsizing Inference API (Chat Completions is what Warp's custom endpoint feature expects), so requests still land on Downsizing and show up in your savings dashboard.

What you can do instead

If you'd rather have this managed automatically, check out our other supported integrations — tools like Claude Code, Codex, Droid, and Cline all read their custom endpoint from a config file, so ds link can manage them end-to-end.

If Warp later exposes its inference-endpoint setting through a config file, CLI flag, or environment variable, we'll wire up ds link warpimmediately. If you know of such a mechanism we've missed, please let us know.