Provider Quick Reference

This guide provides concise setup instructions for common LLM providers. For complete documentation, see the full provider guide.

Provider Configuration Methods

LLxprt Code supports two main ways to configure providers:

1. Using Built-in Aliases

Many popular providers have built-in aliases for quick setup:

# Use the alias (recommended for supported providers)
/provider anthropic
/provider gemini
/provider qwen
/provider synthetic

# Then set your key and model
/key sk-your-api-key
/model your-model-name

2. Using OpenAI-Compatible Endpoint

For providers without aliases, use the OpenAI protocol:

/provider openai
/baseurl https://provider-api-url/v1/
/key your-api-key
/model model-name

Model Geometry and Budgeting (all providers)

When you set a model, configure both context-limit (ephemeral) and max_tokens (model param):

context-limit: The total tokens allowed for the entire request (prompt + output)
max_tokens: The maximum tokens reserved for the model's response (output only)
Effective prompt budget = context-limit − max_tokens − safety-margin

Important constraint: You cannot set context-limit + max_tokens to exceed the model's actual limit. For example:

If a model supports 200k total context, you CANNOT set context-limit=200000 AND max_tokens=100000
The system needs room for both your prompt AND the response within the limit

Safety margin: 256–2048 tokens (recommend 1024) to avoid last-second overflows from tool wrappers, system prompt, and LLXPRT.md.

Tip: If you see "would exceed the token context window" errors, lower max_tokens first or reduce LLXPRT.md size.

Examples:

Large coding session: context-limit 121000, max_tokens 10000 → prompt budget ≈ 110k (minus safety).
Writing mode: context-limit 190000, max_tokens 8000 → prompt budget ≈ 181k (minus safety).

Auth-variant note: Context windows often differ between API-key access and OAuth/subscription access for the same model. The numbers below name the variant where it matters. When in doubt, start lower and increase until you hit a provider limit error.

Reasoning tips: Interleaved-thinking models (e.g. MiniMax, Kimi) rely on prior reasoning tokens, so keep recent reasoning in context (/set reasoning.stripFromContext none). When you need to manage a large window, trim older reasoning while surfacing recent thinking blocks (/set reasoning.stripFromContext allButLast or all).

Subscription & OAuth Providers

OpenAI Codex (ChatGPT Plus/Pro OAuth)

Use your ChatGPT Plus or Pro subscription directly — no API key needed:

/auth codex enable
/provider codex
/model gpt-5.6-sol

This uses OAuth to authenticate with your ChatGPT subscription.

Model geometry & recommended settings (Codex)

Context: 262,144 tokens (Codex OAuth)
gpt-5.x reasoning models do NOT support temperature — use /set reasoning.effort instead
Reasoning effort: minimal, low, medium, high, xhigh, max

/set context-limit 262144
/set modelparam max_tokens 8192
/set reasoning.effort high

Common models: gpt-5.6-sol (default), gpt-5.6-terra, gpt-5.6-luna, gpt-5.5

Anthropic (Claude)

Using Alias (Recommended)

/provider anthropic
/key sk-ant-your-key
/model claude-opus-4-8

Or OAuth (Claude Pro/Max)

/auth anthropic enable

Note: OAuth is lazy — authentication happens when you first use the provider.

Model geometry & recommended settings (Anthropic)

Common models: claude-opus-4-8, claude-sonnet-5, claude-sonnet-4-6, claude-haiku-4-5

Guidance:

Default context-limit 200000 (Opus). Sonnet may support a larger window depending on your Anthropic plan; the very large (1M-class) windows are plan/credit-gated rather than always-on. Check Anthropic's documentation for your current limits.
If you enable thinking, increase max_tokens as needed and keep ≥1k tokens of safety.

/set context-limit 200000
/set modelparam max_tokens 4096
/set reasoning.effort high

Profile JSON:

{
  "version": 1,
  "provider": "anthropic",
  "model": "claude-opus-4-8",
  "modelParams": { "max_tokens": 4096 },
  "ephemeralSettings": { "context-limit": 200000 }
}

Environment variable: export ANTHROPIC_API_KEY=sk-ant-...

Google Gemini

Using Alias

/provider gemini
/key your-gemini-key
/model gemini-2.5-pro

Or API Key

/key save gemini your-gemini-key

Note: Set the key before your first request to that provider.

Important (Gemini "Login with Google" removed): Google has removed the free consumer "Login with Google" flow for the Gemini CLI entirely. Use a Gemini API key (GEMINI_API_KEY) or Vertex AI credentials instead. See Google Cloud auth.

Model geometry & recommended settings (Gemini)

Common models: gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite (stable). Preview models gemini-3-pro-preview and gemini-3-flash-preview are also selectable.

Guidance:

Context-limit up to 1048576 (API key) for Gemini 2.5 models; lower if you see provider limit errors.
Max output tokens: up to 65536

/set context-limit 1048576
/set modelparam max_tokens 4096

Profile JSON:

{
  "version": 1,
  "provider": "gemini",
  "model": "gemini-2.5-pro",
  "modelParams": { "temperature": 0.7, "max_tokens": 4096 },
  "ephemeralSettings": { "context-limit": 1048576 }
}

Environment variable: export GEMINI_API_KEY=...

Qwen

Using Alias with API Key

/provider qwen
/key your-dashscope-key
/model qwen3-coder-plus

Qwen is now API-key-only. Qwen's free OAuth tier ended 2026-04-15 and the OAuth provider has been removed. Use a DashScope API key (DASHSCOPE_API_KEY) or an OpenRouter API key. See authentication for details.

Model geometry & recommended settings (Qwen)

Common models: qwen3-coder-plus, qwen3-coder

Guidance:

Start with context-limit 200000; lower if you hit provider limits.
This alias is for Qwen's own service. It is not used for Cerebras.

/set context-limit 200000
/set modelparam max_tokens 4096

Profile JSON:

{
  "version": 1,
  "provider": "qwen",
  "model": "qwen3-coder-plus",
  "modelParams": { "temperature": 0.7, "max_tokens": 4096 },
  "ephemeralSettings": { "context-limit": 200000 }
}

API-Key Providers (with aliases)

OpenAI (API Key)

/provider openai
/keyfile ~/.openai_key
/model gpt-5.6-sol

Model geometry & recommended settings (OpenAI)

Common models: gpt-5.6 (Sol alias), gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5

GPT-5.6+ models automatically use the Responses API on canonical OpenAI (api.openai.com); custom OpenAI-compatible base URLs stay on Chat Completions.

Guidance:

Context: up to 1,048,576 tokens on the OpenAI API (vs. 262,144 via Codex OAuth — see the Codex section above)
gpt-5.x reasoning models do NOT support temperature — use /set reasoning.effort instead
Reasoning effort: minimal, low, medium, high, xhigh, max. max is model/provider-specific; project minimal maps to wire none for GPT-5.6 Responses.

/set context-limit 400000 # adjust to your model's actual window (check provider docs)
/set modelparam max_tokens 8192
/set reasoning.effort high

Kimi (Moonshot AI)

Kimi ships the K3 frontier model with deep reasoning, multi-step tool orchestration, and native multimodal vision. There are two ways to reach it:

kimi-for-coding — the subscription-served model (Kimi Code subscription). Use the kimi provider alias, which points at the /coding/v1 endpoint. Thinking is always on.
kimi-k3 — the pay-per-token model on the raw Moonshot API (https://api.moonshot.ai/v1). Same K3 capabilities, billed per token.

Note: The kimi alias's defaultModel is kimi-for-coding (the subscription path). To use the pay-per-token kimi-k3 on the Moonshot API, point the alias at the raw endpoint with /baseurl https://api.moonshot.ai/v1 and /model kimi-k3.

Using the subscription (kimi-for-coding)

/provider kimi
/keyfile ~/.kimi_key
/model kimi-for-coding

Using the Moonshot API pay-per-token (kimi-k3)

/provider kimi
/baseurl https://api.moonshot.ai/v1
/keyfile ~/.moonshot_key
/model kimi-k3

Model geometry & recommended settings (Kimi K3)

Context: 1,000,000 tokens (1M)
Max output: 131,072 tokens (default; up to 1,048,576 max)
Architecture: Frontier MoE with always-on thinking
Strengths: Deep reasoning, 200-300 sequential tool calls, native vision (images and video)
Reasoning effort: low / high / max (default max). There is no medium for K3. Thinking is always on and cannot be disabled.
Vision: Native, but requires base64 or ms://<file-id> inputs — public image URLs are not accepted.

/set context-limit 1000000
/set modelparam max_tokens 131072
/set reasoning.effort max
/set reasoning.enabled true
/set reasoning.includeInResponse true

Profile JSON (pay-per-token kimi-k3 on the Moonshot API):

{
  "version": 1,
  "provider": "kimi",
  "model": "kimi-k3",
  "modelParams": { "max_tokens": 131072 },
  "ephemeralSettings": {
    "context-limit": 1000000,
    "base-url": "https://api.moonshot.ai/v1",
    "reasoning.effort": "max",
    "reasoning.enabled": true,
    "reasoning.includeInResponse": true
  }
}

Pricing (Kimi K3)

Subscription (kimi-for-coding) — flat-rate Kimi Code subscription plan; usage is covered by the subscription.
Pay-per-token (kimi-k3 on the Moonshot API) — $0.30 / 1M cached input tokens, $3.00 / 1M non-cached input tokens, $15.00 / 1M output tokens (flat).

Kimi K3 via Synthetic/Chutes

Kimi K3 is also available through third-party providers:

# Via Synthetic
/provider synthetic
/keyfile ~/.synthetic_key
/model hf:moonshotai/Kimi-K3

# Via Chutes
/provider chutes-ai
/keyfile ~/.chutes_key
/model moonshotai/Kimi-K3

Synthetic (Hugging Face Models)

/provider synthetic
/key your-synthetic-key
/model hf:zai-org/GLM-4.7

Model geometry & recommended settings (Synthetic)

Popular models: hf:zai-org/GLM-4.7, hf:moonshotai/Kimi-K3

Guidance:

Context varies by model/runtime. Start with context-limit 200000 and adjust.

/set context-limit 200000
/set modelparam max_tokens 4096

Profile JSON:

{
  "version": 1,
  "provider": "synthetic",
  "model": "hf:zai-org/GLM-4.7",
  "modelParams": { "temperature": 0.7, "max_tokens": 4096 },
  "ephemeralSettings": { "context-limit": 200000 }
}

Chutes AI

/provider chutes-ai    # Has built-in alias
# OR
/provider openai
/baseurl https://api.chutes.ai/v1/
/key your-chutes-key
/model zai-org/GLM-5-TEE

Model geometry & recommended settings (Chutes AI)

/set context-limit 200000
/set modelparam max_tokens 4096

DeepSeek

/provider deepseek
/key your-deepseek-key
/model deepseek-v4-flash

Z.AI

/provider zai
/key your-zai-key
/model glm-5

Makora

/provider makora
/key your-makora-key
/model nvidia/Kimi-K2.6-NVFP4

Models Requiring Custom BaseURL

These providers use the OpenAI-compatible endpoint approach (most also have built-in aliases shown above).

xAI (Grok)

/provider xai          # Has built-in alias
# OR
/provider openai
/baseurl https://api.x.ai/v1/
/key your-xai-key
/model grok-4

Profile JSON:

{
  "version": 1,
  "provider": "openai",
  "model": "grok-4",
  "modelParams": { "max_tokens": 4096, "temperature": 0.7 },
  "ephemeralSettings": {
    "context-limit": 200000,
    "base-url": "https://api.x.ai/v1"
  }
}

OpenRouter

/provider openrouter   # Has built-in alias
# OR
/provider openai
/baseurl https://openrouter.ai/api/v1/
/key your-openrouter-key
/model nvidia/nemotron-nano-9b-v2

Profile JSON:

{
  "version": 1,
  "provider": "openai",
  "model": "nvidia/nemotron-nano-9b-v2",
  "modelParams": { "max_tokens": 4096, "temperature": 0.7 },
  "ephemeralSettings": {
    "context-limit": 200000,
    "base-url": "https://openrouter.ai/api/v1"
  }
}

Fireworks

/provider fireworks    # Has built-in alias
# OR
/provider openai
/baseurl https://api.fireworks.ai/inference/v1/
/key your-fireworks-key
/model fireworks/minimax-m3

Cerebras Code

/provider cerebras-code   # Has built-in alias
# OR
/provider openai
/baseurl https://api.cerebras.ai/v1/
/key your-cerebras-key
/model qwen-3-coder-480b
# Recommended runtime tuning:
/set context-limit 131000
/set modelparam max_tokens 10000

Notes:

The Cerebras endpoint may limit context below a model's full window; budget room for completions.
Effective prompt budget = context-limit − max_tokens − safety.
The /provider qwen alias is for Qwen's own service, not for Cerebras.

AI Gateways / Proxies

LiteLLM

LiteLLM is an open-source AI gateway that provides a unified OpenAI-compatible interface to 100+ LLM providers (Azure OpenAI, AWS Bedrock, Vertex AI, Groq, Together, and more).

/provider litellm      # Has built-in alias
/key your-litellm-key
/model anthropic/claude-sonnet-4-20250514

Or without the alias:

/provider openai
/baseurl http://127.0.0.1:4000/v1/
/key your-litellm-key
/model gpt-4o

Model geometry & recommended settings (LiteLLM)

Context and output limits depend on the underlying model routed through the proxy. Start with conservative defaults and adjust:

/set context-limit 128000
/set modelparam max_tokens 4096

Profile JSON:

{
  "version": 1,
  "provider": "litellm",
  "model": "anthropic/claude-sonnet-4-20250514",
  "modelParams": { "max_tokens": 4096 },
  "ephemeralSettings": { "context-limit": 200000 }
}

Environment variable: export LITELLM_API_KEY=sk-...

Local Models

For complete local-model guidance, see Using Local Models.

LM Studio

/provider lm-studio    # Has built-in alias
# OR
/provider openai
/baseurl http://127.0.0.1:1234/v1/
/model your-local-model

llama.cpp

/provider llama-cpp    # Has built-in alias
# OR
/provider openai
/baseurl http://localhost:8080/v1/
/model your-model

Model geometry & recommended settings (Local)

Context depends on your local runtime and model build. Start small and increase:

/set context-limit 32000
/set modelparam max_tokens 2048

Ollama

Ollama exposes an OpenAI-compatible endpoint. Use the openai provider with a local base URL (there is no separate local ollama alias; the built-in ollama-cloud alias is for the hosted ollama.com service):

/provider openai
/baseurl http://localhost:11434/v1/
/key dummy-key        # Ollama may require a non-empty key
/model qwen2.5-coder

For the hosted Ollama Cloud service:

/provider ollama-cloud
/key your-ollama-key
/model kimi-k2.6

Authentication Methods

API Keys

Set directly with /key or load from file:

# Set key directly
/key sk-your-api-key

# Load from file (more secure)
/keyfile ~/.keys/your-provider.key

OAuth

Two providers support OAuth for authentication: Anthropic and Codex (ChatGPT).

# Enable OAuth provider (lazy authentication - happens on first use)
/auth anthropic enable
/auth codex enable

# Check OAuth status
/auth

# Logout from provider
/auth provider-name logout

Environment Variables

Set keys in your shell environment (auto-detected):

export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GEMINI_API_KEY="..."

Saving Configuration as Profiles

Save your provider setup for reuse:

# After configuring your provider
/profile save my-setup

# Load later
/profile load my-setup

# Use at startup
llxprt --profile-load my-setup

See Settings and Profiles for complete profile management

Provider Commands Reference

/provider - List all providers or switch to one
/model - List available models or switch models
/baseurl - Set custom API endpoint (for OpenAI-compatible providers)
/key - Set API key for current session
/keyfile - Load key from file
/auth - OAuth authentication
/profile save - Save current provider configuration

Next Steps

Configure your provider using the examples above
Save as profile for easy reuse: /profile save my-config
Adjust model parameters like temperature: /set modelparam temperature 0.7
Learn about profiles: Settings and Profiles Guide

See complete CLI provider documentation for advanced configuration

Uh oh!

FilesExpand file tree

quick-reference.md

Latest commit

History

quick-reference.md

File metadata and controls

Provider Quick Reference

Provider Configuration Methods

1. Using Built-in Aliases

2. Using OpenAI-Compatible Endpoint

Model Geometry and Budgeting (all providers)

Subscription & OAuth Providers

OpenAI Codex (ChatGPT Plus/Pro OAuth)

Model geometry & recommended settings (Codex)

Anthropic (Claude)

Using Alias (Recommended)

Or OAuth (Claude Pro/Max)

Model geometry & recommended settings (Anthropic)

Google Gemini

Using Alias

Or API Key

Model geometry & recommended settings (Gemini)

Qwen

Using Alias with API Key

Model geometry & recommended settings (Qwen)

API-Key Providers (with aliases)

OpenAI (API Key)

Model geometry & recommended settings (OpenAI)

Kimi (Moonshot AI)

Using the subscription (kimi-for-coding)

Using the Moonshot API pay-per-token (kimi-k3)

Model geometry & recommended settings (Kimi K3)

Pricing (Kimi K3)

Kimi K3 via Synthetic/Chutes

Synthetic (Hugging Face Models)

Model geometry & recommended settings (Synthetic)

Chutes AI

Model geometry & recommended settings (Chutes AI)

DeepSeek

Z.AI

Makora

Models Requiring Custom BaseURL

xAI (Grok)

OpenRouter

Fireworks

Cerebras Code

AI Gateways / Proxies

LiteLLM

Model geometry & recommended settings (LiteLLM)

Local Models

LM Studio

llama.cpp

Model geometry & recommended settings (Local)

Ollama

Authentication Methods

API Keys

OAuth

Environment Variables

Saving Configuration as Profiles

Provider Commands Reference

Next Steps