OpenGauss AI Provider Configuration and Setup Guide

OpenGauss supports a wide range of AI inference providers out of the box. You can use a single provider for everything, or route different tasks (main conversation, vision, compression) to different providers. Provider selection happens at startup based on which credentials are available, and you can switch models and providers at any time inside a running session.

Quick start

Run the interactive setup wizard to configure your provider on first use:

gauss setup

To change providers or models after initial setup, use the interactive model picker:

gauss model

Or set the model and provider directly in config:

gauss config set model anthropic/claude-sonnet-4
gauss config set model.provider openrouter

Supported providers

OpenRouter

OpenRouter is the recommended starting point — a single API key gives you access to 100+ models from Anthropic, Google, OpenAI, Meta, Mistral, and many others. Pricing is pay-per-token with no subscriptions.

Setup
config.yaml
Provider routing

Get an API key at openrouter.ai/keys

Add it to your .env:

gauss config set OPENROUTER_API_KEY sk-or-v1-...

Set the model in config.yaml:
```
model: "anthropic/claude-opus-4.6"
```

model: "anthropic/claude-opus-4.6"

# Optional: force OpenRouter even if other credentials exist
# model:
#   provider: "openrouter"

OpenRouter lets you control how requests are routed across underlying providers:

provider_routing:
  sort: "throughput"         # price | throughput | latency
  only: ["anthropic"]        # whitelist provider slugs
  ignore: ["deepinfra"]      # blacklist provider slugs
  data_collection: "deny"    # exclude providers that store data

Anthropic (direct)

Use the Anthropic API directly for Claude models without going through OpenRouter. Supports two authentication methods.

API key
OAuth login (claude.ai subscription)

gauss config set ANTHROPIC_API_KEY sk-ant-...

model: "anthropic/claude-opus-4.6"

Authenticate with your existing Claude subscription — no API key needed:

gauss login

OpenGauss will open a browser window for OAuth device authorization. The token is stored in ~/.gauss/auth.json and refreshed automatically.

OpenAI Codex

Access OpenAI’s Codex models. Supports both API key authentication and ChatGPT subscription OAuth.

API key
ChatGPT subscription (OAuth)

gauss config set OPENAI_API_KEY sk-...

model: "openai/gpt-4o"

If you have a ChatGPT Plus or Pro subscription, you can authenticate without an API key:

gauss model
# Select: OpenAI Codex → Login with browser

The session token is cached in ~/.gauss/auth.json.

Nous Portal

Nous Research’s inference platform, with access to Hermes and other research models. Authentication uses OAuth device code flow — no API key is required.

OAuth login
config.yaml
Base URL override

gauss login
# Select: Nous Portal

OpenGauss handles the device code flow, opens the browser, and stores the refresh token automatically. The token is saved in ~/.gauss/auth.json and refreshed transparently.

model: "nousresearch/hermes-3-llama-3.1-70b"
# model:
#   provider: "nous"

After logging in, provider: "nous" forces Nous Portal even when other credentials are present.

# Override Nous Portal base URL (advanced)
gauss config set NOUS_BASE_URL https://portal.nousresearch.com

Z.AI / GLM (Zhipu AI)

Access ZhipuAI’s GLM models through the Z.AI platform.

Setup
Custom endpoint

gauss config set GLM_API_KEY ...

model: "zhipuai/glm-4-plus"
# model:
#   provider: "zai"

Get your key at z.ai or open.bigmodel.cn.

# Override the default base URL
gauss config set GLM_BASE_URL https://api.z.ai/api/paas/v4

Kimi / Moonshot AI

Access Moonshot AI’s Kimi models, including the Kimi Code coding-focused API.

Setup
Endpoint variants

gauss config set KIMI_API_KEY sk-kimi-...

model: "moonshot/kimi-k2"
# model:
#   provider: "kimi-coding"

Get your key at platform.kimi.ai.

# Kimi Code API (default for sk-kimi- keys)
gauss config set KIMI_BASE_URL https://api.kimi.com/coding/v1

# Legacy Moonshot international
gauss config set KIMI_BASE_URL https://api.moonshot.ai/v1

# Moonshot China endpoint
gauss config set KIMI_BASE_URL https://api.moonshot.cn/v1

MiniMax

Access MiniMax models through the global or China endpoint.

Global endpoint
China endpoint

gauss config set MINIMAX_API_KEY ...

# model:
#   provider: "minimax"

Get your key at minimax.io.

gauss config set MINIMAX_CN_API_KEY ...

# model:
#   provider: "minimax-cn"

Get your key at minimaxi.com.

DeepSeek

Access DeepSeek models directly through the DeepSeek API.

Setup
Custom endpoint

gauss config set DEEPSEEK_API_KEY sk-...

model: "deepseek/deepseek-chat-v3"
# model:
#   provider: "deepseek"

Get your key at platform.deepseek.com.

# Override the default DeepSeek base URL
gauss config set DEEPSEEK_BASE_URL https://api.deepseek.com/v1

Custom OpenAI-compatible endpoint

Any server that implements the OpenAI Chat Completions API can be used as a provider — local vLLM, LM Studio, Ollama, LiteLLM, or a third-party proxy.

.env setup
config.yaml
Auxiliary models on custom endpoint

OPENAI_BASE_URL=http://localhost:8000/v1
OPENAI_API_KEY=not-needed   # required by some servers, ignored by others

model: "meta-llama/Llama-3.1-70B-Instruct"
# The model name must match what your server expects.

Route side tasks to your custom endpoint too:

auxiliary:
  vision:
    provider: "main"
    model: "llava-v1.6-34b"
    base_url: "http://localhost:8000/v1"
  compression:
    provider: "main"
    model: ""

Set provider: "main" in the auxiliary section to route side tasks (vision, compression, web extraction) through the same custom endpoint.

Multi-provider auxiliary models

OpenGauss uses lightweight “auxiliary” models for side tasks that don’t need to go through your main provider. You can route each task independently:

auxiliary:
  # Image analysis and browser screenshot processing
  vision:
    provider: "openrouter"
    model: "google/gemini-2.5-flash"

  # Web page summarization
  web_extract:
    provider: "auto"
    model: ""

  # Context compression summaries
  compression:
    provider: "openrouter"
    model: "google/gemini-3-flash-preview"

  # Session search and recall
  session_search:
    provider: "auto"
    model: ""

Provider options for auxiliary tasks:

Value	Description
`auto`	Best available: OpenRouter → Nous Portal → main endpoint (default)
`openrouter`	Force OpenRouter (requires `OPENROUTER_API_KEY`)
`nous`	Force Nous Portal (requires `gauss login`)
`codex`	Force OpenAI Codex OAuth
`main`	Use the custom endpoint from `OPENAI_BASE_URL`

Switching providers at runtime

Inside a running gauss session you can switch the active model without restarting:

/model          # Interactive model picker
/provider       # Show the currently active provider and model

The /model command shows a menu of configured providers and their available models. The selection persists for the current session; to make it permanent, save it to config.yaml:

gauss config set model anthropic/claude-sonnet-4

Provider priority chain

When model.provider is set to auto (the default), OpenGauss picks the first available provider in this order:

Active OAuth session in ~/.gauss/auth.json (Nous Portal, Anthropic, or OpenAI Codex — whichever was last logged in with gauss login)
OpenRouter (if OPENAI_API_KEY or OPENROUTER_API_KEY is set)
API-key providers checked in registry order:
- Z.AI / GLM (if GLM_API_KEY, ZAI_API_KEY, or Z_AI_API_KEY is set)
- Kimi / Moonshot (if KIMI_API_KEY is set)
- MiniMax global (if MINIMAX_API_KEY is set)
- MiniMax China (if MINIMAX_CN_API_KEY is set)
- Anthropic (if ANTHROPIC_API_KEY or ANTHROPIC_TOKEN is set)
- DeepSeek (if DEEPSEEK_API_KEY is set)
Fallback: OpenRouter (even without a key, so startup errors are consistent)

To lock in a specific provider and skip auto-detection, set model.provider explicitly in config.yaml or pass --provider on the command line.

Get Started

Core Concepts

Configuration

Guides

OpenGauss AI Provider Configuration and Setup Guide

Quick start

Supported providers

OpenRouter

Anthropic (direct)

OpenAI Codex

Nous Portal

Z.AI / GLM (Zhipu AI)

Kimi / Moonshot AI

MiniMax

DeepSeek

Custom OpenAI-compatible endpoint

Multi-provider auxiliary models

Switching providers at runtime

Provider priority chain

Build docs developers (and LLMs) love

Get Started

Core Concepts

Configuration

Guides

Documentation Index

​Quick start

​Supported providers

​OpenRouter

​Anthropic (direct)

​OpenAI Codex

​Nous Portal

​Z.AI / GLM (Zhipu AI)

​Kimi / Moonshot AI

​MiniMax

​DeepSeek

​Custom OpenAI-compatible endpoint

​Multi-provider auxiliary models

​Switching providers at runtime

​Provider priority chain

Build docs developers (and LLMs) love

Quick start

Supported providers

OpenRouter

Anthropic (direct)

OpenAI Codex

Nous Portal

Z.AI / GLM (Zhipu AI)

Kimi / Moonshot AI

MiniMax

DeepSeek

Custom OpenAI-compatible endpoint

Multi-provider auxiliary models

Switching providers at runtime

Provider priority chain