Skip to main content

​
What is Claude Code

Claude Code is Anthropic’s agentic coding tool for reading codebases, editing files, running commands, and integrating with development tools. This CLI setup uses Capriole AI by configuring Claude Code’s API endpoint, auth token, and model environment variables.

​
Prerequisites

Before you begin, make sure you have:

​
Installation

npm install -g @anthropic-ai/claude-code

​
Configuration

Create or update ~/.claude/settings.json for persistent setup: 
{
  "model": "claude-latest",
  "effortLevel": "high",
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.caprioletech.com",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_CAPRIOLE_AI_API_KEY",
    "ANTHROPIC_MODEL": "claude-latest",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-latest",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-latest",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-latest",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-latest",
    "CLAUDE_CODE_SUBAGENT_MODEL": "claude-latest"
  }
}
For a one-time shell setup, export the same values before launching Claude Code: 
export ANTHROPIC_BASE_URL="https://api.caprioletech.com"
export ANTHROPIC_AUTH_TOKEN="YOUR_CAPRIOLE_AI_API_KEY"

export ANTHROPIC_MODEL="claude-latest"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-latest"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-latest"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-latest"
export ANTHROPIC_SMALL_FAST_MODEL="claude-latest"
export CLAUDE_CODE_SUBAGENT_MODEL="claude-latest"

​
Run Claude Code

After saving ~/.claude/settings.json or exporting the variables, start Claude Code from your project directory: 
claude
For a one-time non-interactive run without a saved settings file, pass the model explicitly: 
claude --model claude-latest -p "Explain this repository."

​
Supported models

Claude Code supports Claude latest, Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6 through Capriole AI. Use claude-latest for new setups so Capriole AI can route Claude Code to our recommended flagship Claude model without changing your config.
  • claude-latest
  • claude-opus-4-8
  • claude-opus-4-7
  • claude-opus-4-6
  • claude-sonnet-4-6

​
Troubleshooting

​
Why does Claude Code return an authentication error?

Confirm that ANTHROPIC_AUTH_TOKEN is exported in the same shell before launching claude. Use ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY, for this setup.

​
Why do requests fail with endpoint errors?

Confirm that ANTHROPIC_BASE_URL is https://api.caprioletech.com without /v1; Claude Code sends requests to /v1/messages and /v1/messages/count_tokens.

​
Why doesn’t Claude Code use Claude latest?

Confirm that model, ANTHROPIC_MODEL, ANTHROPIC_SMALL_FAST_MODEL, and the default model environment variables are set before launch, then check /model or /status for an active session or settings override.

​
Why doesn’t Claude Code use the configured effort level?

Set "effortLevel": "high" in ~/.claude/settings.json, then restart Claude Code. Claude Code settings accept low, medium, high, and xhigh; max is not accepted in effortLevel. For the absolute maximum effort in a session, use /effort max, launch with claude --effort max, or set CLAUDE_CODE_EFFORT_LEVEL=max before launching claude.

​
Why can’t Claude Code select Opus 4.8?

Update Claude Code with claude update, then run claude --version.

​
Why does token counting or startup model discovery fail?

Keep the model environment variables in place and confirm that the model ID is claude-latest or a supported concrete Claude model ID. Capriole AI proxies /v1/messages/count_tokens to the selected upstream, so token-count preflight depends on that upstream supporting Anthropic token counting.

​
Notes

  • ~/.claude/settings.json applies to your user account across projects. Use YOUR_CAPRIOLE_AI_API_KEY as a placeholder and replace it with your actual Capriole AI API key.
  • ANTHROPIC_BASE_URL changes where Claude Code sends requests. The model environment variables control which model ID is sent.
  • ANTHROPIC_SMALL_FAST_MODEL should match claude-latest so Claude Code helper requests use the same Capriole AI latest alias as the main conversation.
  • This setup uses high as the documented default effort level. On Opus 4.8, xhigh is also a persistent effortLevel option, while max is a session-level override and can spend more thinking tokens.
  • Official references: Claude Code setup, environment variables, model configuration, and Claude effort.