Hushwire

setup · Cursor + Kimi K2.5

Use Cursor with Kimi K2.5, privately

Kimi K2.5 from Moonshot AI is one of the strongest open coding models. Hushwire serves it as hush-kimi on a zero-retention endpoint, so Cursor can use it without your code being stored, and without anything going to OpenAI or Anthropic.

1. Get a key

Pick a plan (from $19 a month) and create a key on the API keys page. Keys start with hw_.

2. Configure Cursor

  1. Cursor Settings → Models → turn on "OpenAI API Key" and paste the key.
  2. Turn on "Override OpenAI Base URL" and paste the URL.
  3. Add a custom model with the model id below.
Cursor settings
# Powered by Hushwire private inference (zero retention): hushwire.vibiz.app
OpenAI API Key:          hw_your_key_here
Override OpenAI Base URL: https://hushwire.vibiz.app/v1
Custom model:            hush-kimi

3. Test the connection

This sends one tiny request. You should see a reply containing “hushwire ok”.

terminal
curl -s https://hushwire.vibiz.app/v1/messages \
  -H "x-api-key: hw_your_key_here" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"hush-kimi","max_tokens":32,"messages":[{"role":"user","content":"Reply with: hushwire ok"}]}'

About Kimi K2.5 on Hushwire

Alias hush-kimi, context 262,144 tokens, quota weight 4×. Strong agentic coder with long tool chains. Request bodies are never written to a database or log; the only thing kept is a metadata row (token counts, model, latency) for 30 days, visible in your dashboard.

Compare models

AliasModelContextGood for
hush-coderQwen3 Coder 30B A3B (Qwen)262kFast, cheap default for edits, refactors and tool loops.
hush-coder-largeQwen3 Coder Next (Qwen)262kThe largest Qwen coder. Best for multi-file agent work.
hush-kimiKimi K2.5 (Moonshot AI)262kStrong agentic coder with long tool chains.
hush-glmGLM 4.7 (Z.ai)200kReliable tool calling, good at planning and front-end work.
hush-deepseekDeepSeek V4 Flash (DeepSeek)128kVery cheap general coder, good for quick questions.

Other guides for Cursor: Qwen3 Coder, Qwen3 Coder Next, GLM 4.7, DeepSeek V4 Flash.

Try it before you configure anything

The playground runs the same open coding models. No sign-up, 20 messages a day, and nothing you type is stored.