RubyLLM::GitLab

Access Claude and GPT models through GitLab Duo via the RubyLLM interface. One provider, automatic routing.

Requirements

• Ruby >= 3.1
• RubyLLM >= 1.0
• GitLab Premium/Ultimate with Duo Enterprise
• Personal Access Token with ai_features scope

Installation

gem 'ruby_llm-gitlab'

Configuration

gitlab.com

RubyLLM.configure do |config|
  config.gitlab_api_key = ENV['GITLAB_TOKEN']
end

Self-managed

RubyLLM.configure do |config|
  config.gitlab_api_key = ENV['GITLAB_TOKEN']
  config.gitlab_instance_url = 'https://gitlab.example.com'
  config.gitlab_gateway_url = 'https://cloud.gitlab.com'  # default
end

Usage

Chat

chat = RubyLLM.chat(model: 'duo-chat-opus-4-6', provider: :gitlab)
chat.ask "What's the best way to learn Ruby?"

Streaming

chat = RubyLLM.chat(model: 'duo-chat-sonnet-4-6', provider: :gitlab)
chat.ask "Tell me a story" do |chunk|
  print chunk.content
end

Tools

class Weather < RubyLLM::Tool
  description "Get current weather"
  param :city, type: :string

  def execute(city:)
    { temperature: 22, condition: "sunny" }
  end
end

chat = RubyLLM.chat(model: 'duo-chat-opus-4-6', provider: :gitlab)
chat.with_tool(Weather).ask "What's the weather in Berlin?"

Agents

class CodeReviewer < RubyLLM::Agent
  model 'duo-chat-opus-4-6'
  provider :gitlab
  instructions "You are a senior Ruby developer. Review code for bugs and style."
end

CodeReviewer.new.ask "Review this: def foo(x) x+1 end"

Rails

class Chat < ApplicationRecord
  acts_as_chat model: 'duo-chat-sonnet-4-6', provider: :gitlab
end

chat = Chat.create!
chat.ask "Explain Active Record callbacks"

Models

GitLab Model ID	Routes To	Backend
duo-chat-opus-4-6	claude-opus-4-6	Anthropic
duo-chat-sonnet-4-6	claude-sonnet-4-6	Anthropic
duo-chat-opus-4-5	claude-opus-4-5-20251101	Anthropic
duo-chat-sonnet-4-5	claude-sonnet-4-5-20250929	Anthropic
duo-chat-haiku-4-5	claude-haiku-4-5-20251001	Anthropic
duo-chat-gpt-5-1	gpt-5.1-2025-11-13	OpenAI
duo-chat-gpt-5-2	gpt-5.2-2025-12-11	OpenAI
duo-chat-gpt-5-mini	gpt-5-mini-2025-08-07	OpenAI

How It Works

User → RubyLLM.chat(provider: :gitlab)
         ↓
       GitLab Provider (router)
         ├── Claude model? → AnthropicDelegate → GitLab AI Gateway → Anthropic API
         └── GPT model?    → OpenAIDelegate   → GitLab AI Gateway → OpenAI API

1. Token exchange — Your PAT is exchanged for a short-lived gateway token via GitLab's Direct Access API
2. Routing — The provider inspects the model ID and delegates to the correct sub-provider
3. Proxy — Requests go through GitLab's AI Gateway (cloud.gitlab.com) which proxies to the upstream API
4. Model mapping — GitLab model IDs (e.g. duo-chat-opus-4-6) are transparently mapped to upstream IDs (e.g. claude-opus-4-6)

Limitations

• Chat completions only (no embeddings, images, audio, moderation)
• No Codex/Responses API models
• Token cached for 25 minutes; auto-refreshes on expiry
• Requires GitLab Duo Enterprise license

Troubleshooting

ConfigurationError: Missing configuration for GitLab: gitlab_api_key

Set your token:

RubyLLM.configure { |c| c.gitlab_api_key = ENV['GITLAB_TOKEN'] }

ConfigurationError: Failed to obtain GitLab direct access token

• Verify your PAT has ai_features scope
• Check your GitLab instance URL is correct
• Ensure Duo Enterprise is enabled for your group/project

401 errors during chat

The gem automatically retries once with a fresh token. If it persists:

• Your PAT may have expired
• The Duo license may have been revoked

License

MIT