FastMetal now works with Cursor and DeepSeek Harness

Cursor and DeepSeek Harness (dsh), DeepSeek's open-source coding agent, can now run on FastMetal models. Both need only a FastMetal API key and a base URL.

Cursor

Cursor sends requests shaped like OpenAI's Responses API to the chat completions path, which the standard /v1 endpoint rejects. FastMetal now has an endpoint built for those requests. In Cursor, open Settings, then Models, turn on OpenAI API Key and paste your FastMetal key, then turn on Override OpenAI Base URL and enter:

https://api.fastmetal.ai/cursor

Add a model with Add Custom Model, for example anthropic-claude-sonnet-5-5, and it shows up in the chat and agent model pickers. We tested the Claude models with Cursor's request format, including streamed answers and tool calls.

Two limits come from Cursor, not from FastMetal. Your own API key only works on the Premium, Super and Ultra plans. Tab completion, Auto, Cloud Agents and Background Agents always use Cursor's own models.

DeepSeek Harness

Start dsh with npx @deepseek-ai/dsh web, open Settings, then Models, and add a Custom model API with base URL https://api.fastmetal.ai/v1 and the OpenAI Chat Completions protocol. Then deepseek-v4-pro or deepseek-v4.1-flash can drive the agent. In our test, DeepSeek V4 Pro wrote a test, watched it fail, fixed both bugs and reran the test to green. The full run is in our DeepSeek Harness walkthrough.

To turn on thinking, set the model's Reasoning levels so that high sends high.

Billing

Usage from Cursor and dsh comes out of the same prepaid balance as any other API call and appears on the same usage page. Agents call the model many times per task, so expect more tokens than a plain chat. Per-token rates are on the pricing page.

The step-by-step setup is in the Cursor guide and the DeepSeek Harness guide. Create a key on the API keys page.