Skip to main content
Claude Code uses the Anthropic Messages API. Set ANTHROPIC_BASE_URL to the AIOHub gateway root; Claude Code appends /v1/messages itself.

​
Prerequisites

  • Claude Code installed with Node.js 18+
  • An AIOHub sk- API key
  • A group assigned to the API key that can access Claude-compatible models
  • A positive account balance

​
Temporary setup

Set the variables in the shell where you start Claude Code: 
export ANTHROPIC_BASE_URL="https://api.aiohub.org"
export ANTHROPIC_AUTH_TOKEN="sk-your-api-key"
export ANTHROPIC_MODEL="claude-sonnet-4-20250514"
claude
Do not add /v1 to ANTHROPIC_BASE_URL. Set ANTHROPIC_AUTH_TOKEN to the full sk- API key value, without a manual Bearer prefix; Claude Code sends it with Authorization Bearer authentication.

​
Persistent setup

Add the variables to your shell profile: 
echo 'export ANTHROPIC_BASE_URL="https://api.aiohub.org"' >> ~/.zshrc
echo 'export ANTHROPIC_AUTH_TOKEN="sk-your-api-key"' >> ~/.zshrc
echo 'export ANTHROPIC_MODEL="claude-sonnet-4-20250514"' >> ~/.zshrc
source ~/.zshrc
You can also store the same values in Claude Code settings: 
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.aiohub.org",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-api-key",
    "ANTHROPIC_MODEL": "claude-sonnet-4-20250514"
  }
}
User settings live at ~/.claude/settings.json. Project-local private settings can live at .claude/settings.local.json; do not commit real API keys.

​
Optional: 1-hour prompt caching

Claude API prompt caching uses a 5-minute TTL by default. If you often resume long Claude Code sessions, work with large repository context, or reuse the same long prompt prefix after more than 5 minutes, add ENABLE_PROMPT_CACHING_1H to the env block in ~/.claude/settings.json to request a 1-hour TTL: 
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.aiohub.org",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-api-key",
    "ANTHROPIC_MODEL": "claude-sonnet-4-20250514",
    "ENABLE_PROMPT_CACHING_1H": "1"
  }
}
Use the 1-hour TTL for sessions that repeatedly reuse long context. In Anthropic pricing, 1-hour cache writes are charged at 2x the base input token price, while cache reads are cheaper; if you typically resume within 5 minutes, the default cache is enough.

​
Optional: disable attribution header

Claude Code prepends a dynamic attribution header (containing the client version and prompt fingerprint) to the system prompt on every request. When routing through a gateway, this changing prefix invalidates the KV cache and slows down inference. Set CLAUDE_CODE_ATTRIBUTION_HEADER to "0" in ~/.claude/settings.json to omit this dynamic block and improve prompt cache hit rates: 
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.aiohub.org",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-api-key",
    "ANTHROPIC_MODEL": "claude-sonnet-4-20250514",
    "CLAUDE_CODE_ATTRIBUTION_HEADER": "0"
  }
}
This only affects caching on the gateway side. Anthropic’s own prompt caching is unaffected. If you connect directly to the Anthropic API without a gateway, you do not need this setting.

​
Command alias

You can create a shell alias when you want a separate command for AIOHub-backed Claude Code: 
alias mycc="ANTHROPIC_BASE_URL=https://api.aiohub.org ANTHROPIC_AUTH_TOKEN=sk-your-api-key ANTHROPIC_MODEL=claude-sonnet-4-20250514 claude"
Run mycc to start Claude Code with the AIOHub environment.

​
Verify

Start Claude Code and send a short prompt. You can first confirm that the command is available: 
claude --version
To verify the environment without printing the API key value: 
test "$ANTHROPIC_BASE_URL" = "https://api.aiohub.org" && echo "base url ok"
test -n "$ANTHROPIC_AUTH_TOKEN" && echo "api key configured"
test -n "$ANTHROPIC_MODEL" && echo "model configured"

​
Session commands

Inside a Claude Code session, these commands help locate client state:
CommandUse
/statusView current session status
/contextInspect context token usage
/clearClear the current conversation and start fresh
/modelView or change the selected model

​
Protocol notes

Claude Code sends Anthropic Messages requests, normally POST /v1/messages. The Base URL must therefore be https://api.aiohub.org, not https://api.aiohub.org/v1. When testing manually with curl, include the full path and the Anthropic version header: 
curl https://api.aiohub.org/v1/messages \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 128,
    "messages": [{"role": "user", "content": "Hello from AIOHub"}]
  }'

​
Troubleshooting

Check that the shell you launched Claude Code from has ANTHROPIC_AUTH_TOKEN. If ANTHROPIC_API_KEY is also set, Claude Code may use API-key mode; AIOHub setup should use ANTHROPIC_AUTH_TOKEN.If you store configuration in settings.json, you can add apiKeyHelper:
{
  "apiKeyHelper": "echo 'sk-your-api-key'",
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.aiohub.org",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-api-key",
    "ANTHROPIC_MODEL": "claude-sonnet-4-20250514"
  }
}
Confirm ANTHROPIC_BASE_URL does not include /v1. If it is https://api.aiohub.org/v1, Claude Code can end up calling /v1/v1/messages.
Clear the current conversation with /clear, then retry with a fresh prompt. Also confirm that the selected model exists in the group assigned to your API key. If the error is related to Claude Code experimental features, start Claude Code with CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 and retry.
Use /context to inspect context usage. If needed, clear the conversation, start a new session, or reduce long history before automatic compaction.
Open wallet management to view balance and add funds.

​
Official references