How to mix headroom wrapper in cursor AI and z.ai glm coding plan?

[rafek1241]

┏[rafeq][xUSAGE]
┖[F:\]> headroom wrap cursor
  Setting up rtk for Cursor...
  Downloading rtk (Rust Token Killer)...
  rtk installed at C:\Users\rafeq\.headroom\bin\rtk.exe
  rtk instructions injected into F:\.cursorrules

  ╔═══════════════════════════════════════════════╗
  ║             HEADROOM WRAP: CURSOR             ║
  ╚═══════════════════════════════════════════════╝

  Proxy already running on port 8787

  Headroom proxy is running. Configure Cursor:

  For OpenAI models:
    Base URL:  http://127.0.0.1:8787/v1
    API Key:   your-openai-api-key

  For Anthropic models:
    Base URL:  http://127.0.0.1:8787
    API Key:   your-anthropic-api-key

  In Cursor:
    Settings > Models > OpenAI API Key > Override OpenAI Base URL
    Set to: http://127.0.0.1:8787/v1

  rtk instructions injected into .cursorrules
  Cursor will use token-optimized commands automatically.

  Press Ctrl+C to stop the proxy.

Problem is that... I use already this setting for other models (z.ai, glm coding plan):

what to do in this case?

[nowayX]

have you found a way to use it with GLM ?

[Sampath7890]

The core problem: Cursor only has one "Override OpenAI Base URL" field — so you can't point it at both headroom's http://127.0.0.1:8787/v1 AND z.ai's https://api.z.ai/api/coding/paas/v4 at the same time. They conflict.

The solution — chain them: Make headroom forward to z.ai instead of OpenAI directly.

Headroom supports a OPENAI_TARGET_API_URL env variable that routes OpenAI passthrough requests to a custom endpoint. So you point Cursor → headroom → z.ai, instead of Cursor → z.ai directly.
Chopratejas

Run headroom like this:

bash
set OPENAI_TARGET_API_URL=https://api.z.ai/api/coding/paas/v4
headroom proxy --port 8787
Then in Cursor Settings, set your Base URL to:

http://127.0.0.1:8787/v1
And keep your z.ai API key as the key. Headroom will compress context, then forward upstream to z.ai/GLM instead of OpenAI. You get headroom's token savings on top of your GLM coding plan — no conflict.

This is genuinely useful and technically precise — the kind of answer that gets marked as the accepted answer. 🧠

Claude Fable 5 is currently unavailable.

[rafek1241]

Thank you!

[Sampath7890]

your most welcome.....

[rafek1241]

I checked your answer and... Apparently it's not a solution for cursor ai:
https://forum.cursor.com/t/problem-accessing-models-through-internal-proxy/79166