@ganziliang/zhizh-pi-model-setup

Interactive model setup extension for Pi and Zhizhengroup LLM Gateway

Packages

Package details

extensionskill

Install @ganziliang/zhizh-pi-model-setup from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:@ganziliang/zhizh-pi-model-setup
Package
@ganziliang/zhizh-pi-model-setup
Version
0.2.0
Published
Sep 15, 2026
Downloads
760/mo · 9/wk
Author
ganziliang
License
MIT
Types
extension, skill
Size
19 KB
Dependencies
0 dependencies · 1 peer
Pi manifest JSON
{
  "extensions": [
    "./extensions"
  ],
  "skills": [
    "./skills"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

Zhizh Pi Model Setup

Pi extension for configuring the Zhizhengroup LLM Gateway without an existing model.

Install for team members

Install the published npm package:

pi install npm:@ganziliang/zhizh-pi-model-setup

Start Pi:

pi

Run the setup command inside Pi:

/model-setup

Obtain apiId and API Key

Before running /model-setup, open the following gateway page:

https://llm-gateway.zhizhengroup.com/admin-next/api-stats

Use the page to obtain your apiId and API Key. The apiId is available in the browser address bar after the page loads, for example:

https://llm-gateway.zhizhengroup.com/admin-next/api-stats?apiId=56b07442-eb3d-43af-8c23-c13686fabdd0

You can paste either the complete URL or only the value after apiId= into the setup wizard. Keep the API Key private and do not commit it to source control.

Setup steps

The wizard asks for the following values:

  1. apiId or the complete api-stats URL.
  2. The gateway API key.
  3. The default provider.
  4. The default model.

You can paste either the ID:

56b07442-eb3d-43af-8c23-c13686fabdd0

or the complete URL copied from the browser address bar:

https://llm-gateway.zhizhengroup.com/admin-next/api-stats?apiId=56b07442-eb3d-43af-8c23-c13686fabdd0

The extension queries the gateway, creates the available models, and updates the following local Pi files:

~/.pi/agent/models.json
~/.pi/agent/settings.json
~/.pi/agent/company-gateway.json

On Windows, ~/.pi/agent normally means:

C:\Users\<your-windows-user>\.pi\agent

The extension manages these providers:

company-gpt
company-anthropic

The generated company-gpt configuration marks GPT-5.6 models (gpt-5.6-luna, gpt-5.6-terra, and gpt-5.6-sol, including future suffix variants) as accepting both text and image input. This fallback is applied when the gateway stats response omits image metadata.

Preconfigured model tuning

Some gateway metadata lags behind the upstream providers. The extension corrects it while generating the configuration, so a fresh install is already tuned:

DeepSeek

DeepSeek retired the legacy names and now serves deepseek-v4-flash, deepseek-v4-flash-vision-exp, and deepseek-v4-pro with the DeepSeek-V4.1-Flash model. The official model name is deepseek-flash.

The gateway still reports the old contextWindow (128000), reasoning: false, and no image input for these models, and it does not list deepseek-flash at all. The extension therefore applies the official DeepSeek-V4.1-Flash specification to the whole family:

Field Value
contextWindow 1048576 (official context length: 1M)
maxTokens 393216 (official maximum output: 384K)
reasoning true (thinking mode is the DeepSeek default)
input ["text", "image"] (V4.1-Flash is multimodal)

deepseek-flash is added automatically right after deepseek-v4-flash, but only when the account already provides another DeepSeek model. The retired alias deepseek-v4-flash-vision-exp is skipped because it resolves to the exact same model.

Thinking level

The extension also sets the per-model thinking level in settings.json:

{
  "modelThinkingLevels": {
    "company-anthropic/deepseek-flash": "medium"
  }
}

Existing entries in modelThinkingLevels are preserved; only the keys listed above are added or corrected.

Because these defaults live in the extension, re-running /model-setup keeps the tuned values instead of overwriting them with stale gateway metadata.

After setup completes, run:

/reload

or restart Pi.

Update

Update the installed package when a new version is released:

pi update npm:@ganziliang/zhizh-pi-model-setup

Remove

pi remove npm:@ganziliang/zhizh-pi-model-setup

Removing the package does not automatically remove the generated providers from models.json. Edit or remove company-gpt and company-anthropic if they are no longer needed.

Local development

The local path installation is only for the package maintainer and is not needed by normal users:

pi install "C:/Users/ganziliang/Desktop/ai_work/pi_agent/company-model-setup"

Publish a new version

Maintainers can publish a new version from the package directory:

npm version patch
npm publish --access public