LiteLLM

Request flow

ChatBox
    ↓ local master key
http://127.0.0.1:4000/v1/chat/completions
    ↓ LiteLLM model mapping
https://www.mofangai001.cn/v1/chat/completions
    ↓ 魔方AI API key
MiniMax-M2.7

LiteLLM exposes the client alias mofang-minimax-m2-7, while the actual model ID sent to 魔方AI is MiniMax-M2.7. These values serve different purposes and must not be mixed.

01. Prepare and install LiteLLM

Before you begin, prepare:

• 魔方AI API Base: https://www.mofangai001.cn/v1.
• An active 魔方AI token and the exact model IDs allowed by that token.
• A directory for the configuration, such as C:\litellm-mofang on Windows.
• Optional: the latest ChatBox for client verification.

LiteLLM 1.84.0 and newer require Python 3.10 or later. The official install script and uv can provision a compatible Python environment automatically.

macOS, Linux, or Windows WSL

LiteLLM marks the following one-command installer as the recommended local and beginner setup:

curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm/main/scripts/install.sh | sh

The script installs litellm[proxy] and may open the general setup wizard. This guide uses a dedicated config.yaml for 魔方AI, so you can exit the wizard and continue below.

Windows PowerShell

Native PowerShell normally does not include sh. Install uv, then use LiteLLM's documented uv tool method:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
uv tool install "litellm[proxy]"
litellm --version

If uv or litellm is not found after installation, close and reopen the terminal.

02. Configure 魔方AI and LiteLLM keys

Create .env in the configuration directory:

MOFANG_API_BASE=https://www.mofangai001.cn/v1
MOFANG_API_KEY=sk-your-mofang-key
LITELLM_MASTER_KEY=sk-your-litellm-master-key

• MOFANG_API_KEY is the upstream token from 魔方AI Token Management.
• LITELLM_MASTER_KEY is a separate password used by clients connecting to local LiteLLM and must start with sk-.
• Do not reuse the same value for both keys, and never put real values in config.yaml, screenshots, or Git.

Add .env to the project's .gitignore:

.env

03. Configure the model mapping

Create config.yaml in the same directory:

model_list:
  - model_name: mofang-minimax-m2-7
    litellm_params:
      model: openai/MiniMax-M2.7
      api_base: os.environ/MOFANG_API_BASE
      api_key: os.environ/MOFANG_API_KEY

litellm_settings:
  drop_params: true

general_settings:
  master_key: os.environ/LITELLM_MASTER_KEY

Setting	Purpose	Example
model_name	Alias used by clients calling LiteLLM	mofang-minimax-m2-7
model	LiteLLM provider plus the exact upstream model ID	openai/MiniMax-M2.7
api_base	魔方AI's OpenAI-compatible Base URL	https://www.mofangai001.cn/v1
api_key	魔方AI upstream token	Read from .env

魔方AI exposes an OpenAI-compatible endpoint, so the LiteLLM provider prefix is openai/. MiniMax-M2.7, mofang-minimax-m2-7, and the model name entered in ChatBox must each exactly match their corresponding settings, including case.

To add more models, append entries under model_list:

  - model_name: mofang-your-model
    litellm_params:
      model: openai/your-exact-mofang-model-id
      api_base: os.environ/MOFANG_API_BASE
      api_key: os.environ/MOFANG_API_KEY

04. Start LiteLLM locally

Start LiteLLM from the directory that contains .env and config.yaml:

cd C:\litellm-mofang
litellm --config .\config.yaml --port 4000

On macOS, Linux, or WSL:

cd ~/litellm-mofang
litellm --config ./config.yaml --port 4000

Keep the terminal running after LiteLLM reports that it is listening on http://0.0.0.0:4000. Local clients should use the explicit loopback address http://127.0.0.1:4000.

05. Advanced option: run LiteLLM with Docker

Use Docker when you need a background service, server deployment, or container isolation. Skip this section for a simple local setup.

Install and start Docker Desktop first; Windows users should prefer its WSL 2 backend. Create docker-compose.yml in the configuration directory:

services:
  litellm:
    image: docker.litellm.ai/berriai/litellm:latest
    container_name: litellm-mofang
    command:
      - "--config"
      - "/app/config.yaml"
      - "--port"
      - "4000"
    ports:
      - "4000:4000"
    volumes:
      - ./config.yaml:/app/config.yaml:ro
    env_file:
      - .env
    restart: unless-stopped

Start and inspect the container:

cd C:\litellm-mofang
docker compose pull
docker compose up -d --force-recreate
docker compose ps
docker compose logs --tail 50 litellm

Verify that the master key is present without printing its value:

docker compose exec litellm sh -c 'test -n "$LITELLM_MASTER_KEY" && echo MASTER_KEY_OK || echo MASTER_KEY_MISSING'

latest is acceptable for evaluation. Production deployments should pin a tested release tag or image digest for reproducible upgrades and rollbacks.

Failed to fetch remote model cost map means LiteLLM could not retrieve the remote cost map and fell back to its local copy. If model calls still work, the warning normally does not block basic chat. If calls also fail, check the network, proxy, and TLS certificate configuration.

06. Optional: verify 魔方AI directly

If LiteLLM reports an upstream error, bypass LiteLLM temporarily to verify the 魔方AI Base URL, token, and exact model ID:

$baseUrl = "https://www.mofangai001.cn/v1"
$mofangKey = "sk-your-mofang-key"
$directBody = @{
    model = "MiniMax-M2.7"
    messages = @(
        @{ role = "user"; content = "Reply only with: 魔方AI direct connection succeeded" }
    )
    stream = $false
} | ConvertTo-Json -Depth 10

$directResult = Invoke-RestMethod `
    -Uri "$baseUrl/chat/completions" `
    -Method Post `
    -Headers @{ Authorization = "Bearer $mofangKey" } `
    -ContentType "application/json; charset=utf-8" `
    -Body $directBody

$directResult.choices[0].message.content

Clear sensitive terminal variables or close the terminal after testing. Do not publish command history or screenshots containing credentials.

07. Test the local LiteLLM endpoint

Keep LiteLLM running and open another PowerShell window:

$litellmKey = "sk-your-litellm-master-key"
$proxyBody = @{
    model = "mofang-minimax-m2-7"
    messages = @(
        @{ role = "user"; content = "Reply only with: LiteLLM connection succeeded" }
    )
    stream = $false
} | ConvertTo-Json -Depth 10

$proxyResult = Invoke-RestMethod `
    -Uri "http://127.0.0.1:4000/v1/chat/completions" `
    -Method Post `
    -Headers @{ Authorization = "Bearer $litellmKey" } `
    -ContentType "application/json; charset=utf-8" `
    -Body $proxyBody

$proxyResult.choices[0].message.content

This request must use the LiteLLM master key and client alias mofang-minimax-m2-7, not the 魔方AI key or the upstream model ID.

08. Connect ChatBox to LiteLLM

In ChatBox, open Settings → Model Provider → Add Custom Provider, choose OpenAI API Compatible, and enter:

ChatBox setting	Value
Name	LiteLLM
API Mode	OpenAI API Compatible
API Key	LITELLM_MASTER_KEY from .env
API Host	http://127.0.0.1:4000
API Path	/v1/chat/completions
Model	mofang-minimax-m2-7

Select Fetch, or add mofang-minimax-m2-7 manually. Save the provider, return to the chat screen, select the model, and send a test message.

ChatBox stores only the local LiteLLM master key. Do not enter the upstream 魔方AI token in ChatBox.

Troubleshooting

Security

• Create a dedicated 魔方AI token for LiteLLM so it can be revoked independently.
• .env, terminal history, and LiteLLM logs may contain credentials or request data. Do not upload or share them.
• Clients should receive only the LiteLLM master key, never the upstream 魔方AI token.
• If either key may have leaked, revoke it in the corresponding system, update .env, and restart LiteLLM.

Official resources

• LiteLLM Getting Started and recommended installation
• LiteLLM documentation
• LiteLLM repository
• uv installation documentation
• ChatBox provider configuration