Skip to content

quick-reference.md

Latest commit

 

History

History
515 lines (390 loc) · 15.1 KB

quick-reference.md

File metadata and controls

515 lines (390 loc) · 15.1 KB
Raw

Provider Setup Quick Reference

Concise setup instructions for configuring LLM providers in LLxprt Code. For complete documentation, see the full provider guide. For specific model names, context windows, pricing, and recommended runtime tuning, see Provider Models and Limits.

How to configure a provider

LLxprt Code supports two configuration methods. Most providers have a built-in alias for quick setup; providers without an alias use the OpenAI-compatible endpoint approach.

Using a built-in alias

/provider anthropic
/key sk-your-api-key
/model claude-opus-5

Using an OpenAI-compatible endpoint

For providers without an alias, or custom endpoints:

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

If the endpoint needs a model-specific reasoning request shape, see Reasoning Wire Formats.

Built-in provider aliases

LLxprt Code ships with these aliases. Use /provider <alias> to switch. The alias name is what you type after /provider. Default model names change over time and are kept in the dated Provider Models and Limits page; the table below lists only the stable alias and authentication details.

Provider Alias Auth method
Anthropic (API key) anthropic API key (ANTHROPIC_API_KEY)
Claude Code (subscription) claudecode OAuth
Google Gemini gemini API key (GEMINI_API_KEY)
OpenAI (API) openai API key (OPENAI_API_KEY)
OpenAI Codex (ChatGPT sub) codex OAuth
Qwen (DashScope) qwen API key (DASHSCOPE_API_KEY)
Kimi kimi API key
xAI (Grok) xai API key (XAI_API_KEY)
DeepSeek deepseek API key (DEEPSEEK_API_KEY)
Z.AI zai API key (ZAI_API_KEY)
Synthetic synthetic API key
Chutes AI chutes-ai API key (CHUTES_API_KEY)
Makora makora API key (MAKORA_API_KEY)
Fireworks fireworks API key (FIREWORKS_API_KEY)
OpenRouter openrouter API key (OPENROUTER_API_KEY)
Cerebras Code cerebras-code API key (CEREBRAS_API_KEY)
Mistral mistral API key (MISTRAL_API_KEY)
LiteLLM (gateway) litellm API key (LITELLM_API_KEY)
Ollama Cloud (hosted) ollama-cloud API key (OLLAMA_API_KEY)
LM Studio (local) lm-studio None required
llama.cpp (local) llama-cpp None required

For the current default model name each alias uses, see Default models by alias. For context windows and pricing, see Provider Models and Limits.

Subscription and OAuth providers

Two providers support OAuth for authentication:

  • Claude Code (claudecode) — Claude.ai subscription OAuth
  • Codex (codex) — ChatGPT Plus/Pro subscription OAuth
# Enable OAuth for a subscription provider
/auth claudecode enable
/provider claudecode
/model claude-opus-5

# Or for Codex
/auth codex enable
/provider codex
/model gpt-6-astra

gpt-6-astra uses an 872,000-token OAuth context limit and supports low, medium, high, xhigh, and max reasoning effort. gpt-5.6-sol remains the unconditional alias default; gpt-6-astra is an explicit model selection for accounts with Astra access (staged Trusted Access rollout).

OAuth is lazy — authentication happens when you first use the provider, not when you enable it. Check OAuth status with /auth, and log out with /auth <provider> logout.

Note: Anthropic API keys and Claude.ai subscription OAuth are separate identities. anthropic is API-key access (no OAuth). claudecode is subscription OAuth. /auth anthropic does not perform OAuth; it redirects subscription users to /auth claudecode and API-key users to /provider anthropic plus /key or /keyfile.

Authentication methods

API keys

Set a key directly or load it from a file:

# Set key directly (session-only)
/key sk-your-api-key

# Load from a file
/keyfile ~/.keys/your-provider.key

Environment variables

Set keys in your shell environment. LLxprt Code auto-detects the standard environment variable for each provider's alias:

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

The environment variable name for each alias is listed in the alias table above.

Keyfiles

A keyfile loads an API key from a file on disk, which is more secure than passing the key inline:

/keyfile ~/.kimi_key

Saving configuration as profiles

Save your provider setup for reuse across sessions:

# 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 setup examples

Anthropic (Claude)

Using an Anthropic API key:

/provider anthropic
/key sk-ant-your-key
/model claude-opus-5

Or Claude Code OAuth (Claude.ai subscription):

/auth claudecode enable
/provider claudecode
/model claude-opus-5

Google Gemini

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

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

OpenAI (API key)

/provider openai
/keyfile ~/.openai_key
/model gpt-5.5

OpenAI transport selection (Responses vs. Chat Completions)

OpenAI models can use the newer Responses API or the classic Chat Completions API. LLxprt Code picks one automatically:

  • GPT-5.6 and later (including gpt-6-astra, bare model IDs like gpt-5.6, and durable-tier IDs like gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna) use the Responses transport when pointed at the official api.openai.com endpoint. Chat Completions is not available for these on that endpoint.
  • Custom OpenAI-compatible endpoints (proxies, gateways, self-hosted servers, or any non-api.openai.com base URL) default to Chat Completions, even for GPT-5.6+.

This matters when you point the openai alias at a custom endpoint: a Responses-only model may fail if the endpoint does not implement the Responses API. To force a specific transport, set one of these:

# Force Responses (useful on a custom endpoint that supports it)
/set responses-mode responses

# Force Chat Completions
/set responses-mode chat

You can also set it permanently in a profile via the apiMode or responsesMode provider setting, or globally via the responses-mode ephemeral setting. When forced to responses, a custom endpoint uses Responses for models that support it; when forced to chat, Chat Completions is used unless the model requires Responses on the official OpenAI endpoint (GPT-5.6+), where Chat is unavailable and the override is ignored.

OpenAI Codex (ChatGPT Plus/Pro)

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

WebSocket transport

The Codex provider sends Authorization, ChatGPT-Account-ID, originator: codex_cli_rs, and any configured custom headers during the WebSocket handshake. It also sends session-id, thread-id, and x-client-request-id, each set to the session's runtime ID, plus OpenAI-Beta: responses_websockets=2026-02-06. The identity headers use the hyphenated forms used by the current Codex client.

The WebSocket handshake times out after 15 seconds. After the connection is established, the stream has a five-minute idle timeout that resets on each valid text frame. If it expires, LLxprt closes the socket. When no output has been streamed, LLxprt serves that request over HTTP/SSE instead; after repeated consecutive pre-output WebSocket failures it stays on HTTP for the session. When output has already been streamed, LLxprt reports a stream interruption without replaying partial output.

Qwen

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

Note: Qwen is API-key-only. The 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.

Kimi

Kimi's kimi alias defaults to the subscription path (kimi-for-coding). For the pay-per-token Moonshot API, point the alias at the raw endpoint:

# Subscription (kimi-for-coding)
/provider kimi
/keyfile ~/.kimi_key
/model kimi-for-coding

# Pay-per-token (kimi-k3 on the Moonshot API)
/provider kimi
/baseurl https://api.moonshot.ai/v1
/keyfile ~/.moonshot_key
/model kimi-k3

See Provider Models and Limits for Kimi K3 context windows, pricing, and multimodal support details.

Kimi provider Files policy

Kimi PDF uploads are disabled by default. Choose an explicit session or workspace scope before enabling them:

/set provider-files session
/set provider-files-retention-ms 86400000
/set provider-files-delete delete

Use workspace instead of session when the same credential and base URL may reuse file IDs across sessions in the current workspace. IDs are isolated by provider, base URL, credential identity, and scope. The persisted workspace scope is a credential-keyed identifier, not the local directory path. Moving or renaming the directory produces a different scope and does not reuse its prior file IDs. Valid IDs are written to the media reference in session history so replay can reuse them without uploading the same bytes again. LLxprt keeps dynamic IDs in message content and does not change the system instruction.

The retention duration controls how long LLxprt may reuse an ID. With delete, expiry, cache eviction, and session cleanup request best-effort remote deletion. A failed deletion remains tracked for retry. With retain, LLxprt drops its local entry without calling the delete endpoint, and Moonshot's account retention rules continue to apply. Setting provider-files back to off prevents new uploads but does not by itself delete existing remote files.

Kimi Files storage is incompatible with zero-data-retention semantics while a remote file remains stored. Set provider-files-zdr require to reject this mode, or leave allow-retention only when provider storage is acceptable. Memory pressure never enables Files mode.

Experimental Kimi video requires both kimi.experimental-video=true and an explicit provider-files scope. It follows the same retention, deletion, and ZDR settings. Anthropic Files references are currently unsupported in this adapter, so Anthropic remains inline even when provider-files is enabled.

Synthetic (Hugging Face models)

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

Chutes AI

/provider chutes-ai
/key your-chutes-key
/model zai-org/GLM-5-TEE

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

xAI (Grok)

/provider xai
/key your-xai-key
/model grok-4

OpenRouter

/provider openrouter
/key your-openrouter-key
/model nvidia/nemotron-nano-9b-v2

Fireworks

/provider fireworks
/key your-fireworks-key
/model fireworks/minimax-m3

Cerebras Code

/provider cerebras-code
/key your-cerebras-key
/model qwen-3-coder-480b

Note: The /provider qwen alias is for Qwen's own DashScope service, not for Cerebras.

Mistral

/provider mistral
/key your-mistral-key
/model mistral-large-latest

AI gateways and 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
/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

Environment variable: export LITELLM_API_KEY=sk-...

Local models

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

LM Studio

/provider lm-studio
/model your-local-model

Or without the alias:

/provider openai
/baseurl http://127.0.0.1:1234/v1/
/model your-local-model

llama.cpp

/provider llama-cpp
/model your-model

Or without the alias:

/provider openai
/baseurl http://localhost:8080/v1/
/model your-model

Ollama (local)

Ollama exposes an OpenAI-compatible endpoint. There is no separate local ollama alias — the 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

Provider commands reference

Command Description
/provider List all providers or switch to one
/model List available models or switch models
/baseurl Set a custom API endpoint (for OpenAI-compatible providers)
/key Set the API key for the current session
/keyfile Load an API key from a file
/auth Manage OAuth authentication
/profile Save, load, and manage configuration profiles
/set Set model parameters or ephemeral settings

Next steps

  1. Configure your provider using the examples above
  2. Save as profile for easy reuse: /profile save my-config
  3. Adjust model parameters: /set modelparam temperature 0.7
  4. Check model limits and pricing: Provider Models and Limits
  5. Learn about profiles: Settings and Profiles

See the complete CLI provider documentation for advanced configuration.