GitHub Docs

Tutorial: Setting Up Your First AI Provider

Setting Up Your First AI Provider

Connect VibeCody to an AI model so you can start chatting, generating code, and running the agent loop.

Prerequisites: VibeCody installed and on your PATH. See the Quickstart if you need to install it first.

Choose a Provider

VibeCody supports 23 AI providers. Here are the most common starting points:

Provider Type Cost API Key Required Best For
Ollama Local Free No Privacy, offline use, experimentation
Claude Cloud Paid Yes Best coding quality, extended thinking
OpenAI Cloud Paid Yes GPT-4o, broad ecosystem compatibility
Gemini Cloud Free tier Yes Large context windows, free tier
Groq Cloud Free tier Yes Ultra-fast inference
DeepSeek Cloud Paid Yes Cost-effective coding model

Recommendation: Start with Ollama if you want free and local. Start with Claude if you want the best coding results.

Option 1: Ollama (Free, Local)

Ollama runs models entirely on your machine. No API key, no network access, no data leaves your laptop.

Step 1: Install Ollama

# macOS / Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Or visit https://ollama.ai for the desktop installer

Step 2: Pull a Model

ollama pull qwen3-coder:480b-cloud

This downloads the default model VibeCLI expects. Other good options:

ollama pull codellama:13b        # Smaller, faster
ollama pull deepseek-coder:33b   # Strong coding model
ollama pull llama3:70b           # General purpose

Step 3: Start Ollama

ollama serve

Leave this running in a separate terminal (or use the Ollama desktop app, which runs the server automatically).

Step 4: Launch VibeCLI

vibecli

That is it. Ollama is the default provider, so no flags are needed. You should see:

VibeCLI v0.3.3 — AI coding assistant
Provider: ollama (qwen3-coder:480b-cloud)

vibecli>

Step 5: Verify

vibecli> Write a hello world in Rust

You should see streamed output with a working Rust program.

Using a Different Ollama Model

vibecli --model codellama:13b

Or set it permanently in ~/.vibecli/config.toml:

[ollama]
enabled = true
model = "codellama:13b"

Option 2: Claude (Cloud, API Key)

Anthropic’s Claude models are among the strongest for code generation, reasoning, and agentic tasks.

Step 1: Get an API Key

  1. Go to console.anthropic.com
  2. Sign up or log in
  3. Navigate to API Keys and create a new key
  4. Copy the key (starts with sk-ant-)

Step 2: Set the Environment Variable

export ANTHROPIC_API_KEY="sk-ant-your-key-here"

Add this to your ~/.bashrc, ~/.zshrc, or ~/.profile to persist across sessions.

Step 3: Launch with Claude

vibecli --provider claude

Expected output:

VibeCLI v0.3.3 — AI coding assistant
Provider: claude (claude-sonnet-4-6)

vibecli>

Step 4: Verify

vibecli> Explain the borrow checker in one paragraph

You should see a response from Claude.

Using a Different Claude Model

vibecli --provider claude --model claude-sonnet-4-6

Enable Extended Thinking

For complex tasks, Claude supports extended thinking mode. Add to ~/.vibecli/config.toml:

[claude]
enabled = true
thinking_budget_tokens = 10000

Option 3: OpenAI (Cloud, API Key)

Step 1: Get an API Key

  1. Go to platform.openai.com
  2. Navigate to API Keys and create a new key
  3. Copy the key (starts with sk-)

Step 2: Set the Environment Variable

export OPENAI_API_KEY="sk-your-key-here"

Step 3: Launch with OpenAI

vibecli --provider openai

Expected output:

VibeCLI v0.3.3 — AI coding assistant
Provider: openai (gpt-4o)

vibecli>

Step 4: Verify

vibecli> What model are you?

Switch Between Providers

You do not have to commit to one provider. Switch at any time:

# Use Claude for a coding task
vibecli --provider claude --exec "refactor auth.rs to use async"

# Use Ollama for private exploration
vibecli --provider ollama

# Use OpenAI for a quick question
vibecli --provider openai

Inside the REPL, you can also compare providers head-to-head:

vibecli> /arena compare claude openai "Write a binary search in Rust"

This shows both responses side by side with hidden identities so you can vote on quality without bias.

Store Provider Config Permanently

Instead of passing flags every time, edit ~/.vibecli/config.toml:

[claude]
enabled = true
# api_key is read from ANTHROPIC_API_KEY env var

[openai]
enabled = true
# api_key is read from OPENAI_API_KEY env var

[ollama]
enabled = true
api_url = "http://localhost:11434"
model = "qwen3-coder:480b-cloud"

Then just use the --provider flag to switch:

vibecli --provider claude

See the full Configuration Guide for all provider options.

Troubleshooting

“Connection refused” (Ollama)

Ollama server is not running. Start it:

ollama serve

“401 Unauthorized” (Cloud providers)

Your API key is missing or invalid. Check:

# Is the variable set?
echo $ANTHROPIC_API_KEY

# Is it the right key? (should start with sk-ant- for Claude)

“Model not found” (Ollama)

You need to pull the model first:

ollama list                          # See what you have
ollama pull qwen3-coder:480b-cloud   # Pull the default

“Rate limited” (Cloud providers)

You have hit the provider’s rate limit. Wait a minute and retry, or switch to a different provider. For sustained use, check your plan’s rate limits on the provider’s dashboard.

Slow responses with Ollama

Local inference speed depends on your hardware. Options:

  • Use a smaller model: ollama pull codellama:7b
  • Ensure you have enough RAM (13B models need ~8GB, 70B needs ~40GB)
  • On macOS with Apple Silicon, Ollama uses the GPU automatically

Next Steps