Claude Code

Claude Code speaks the Anthropic Messages API, which the gateway relays at POST /v1/messages. It is configured entirely through environment variables — no config file edit is required.

The base URL here is https://api.lulutokens.ai with no /v1 suffix. The Anthropic client appends /v1/messages itself; adding /v1 yourself produces requests to /v1/v1/messages, which the gateway answers with 404.

Configure

Set the environment variables

~/.zshrc

export ANTHROPIC_BASE_URL="https://api.lulutokens.ai"
export ANTHROPIC_AUTH_TOKEN="sk-..."

ANTHROPIC_AUTH_TOKEN — not ANTHROPIC_API_KEY — is the variable for a third-party gateway: it makes the client send Authorization: Bearer <token>, which is what LuluTokens expects.

Choose the models

Claude Code uses two models: a main one for reasoning and a small, fast one for background work such as summarising and file classification. Point both at IDs that exist on the gateway:

export ANTHROPIC_MODEL="claude-sonnet-4-5"
export ANTHROPIC_SMALL_FAST_MODEL="claude-haiku-4-5"

Run it

claude

Check /status inside Claude Code — it shows the base URL and model in use, so you can confirm the traffic is going to LuluTokens.

Keeping it project-local

Exporting the variables globally sends all your Claude Code usage through the gateway. To scope it to one project, put them in that project's settings instead:

.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.lulutokens.ai",
    "ANTHROPIC_AUTH_TOKEN": "sk-...",
    "ANTHROPIC_MODEL": "claude-sonnet-4-5"
  }
}

.claude/settings.json is committed by default. Put the key in .claude/settings.local.json (git-ignored) instead, or keep the key in the environment and only set the URL and model here.

Cost

Claude Code replays the full conversation on every turn, so a long session costs far more than the same number of questions asked separately. Two things help:

Use /clear between unrelated tasks rather than letting one session grow.
Keep ANTHROPIC_SMALL_FAST_MODEL pointed at a genuinely cheap model — it is called constantly in the background.

Per-request cost is visible in the console's usage logs.

Troubleshooting

Symptom	Fix
`404` on every request	`ANTHROPIC_BASE_URL` includes `/v1` — remove it
`401`	Using `ANTHROPIC_API_KEY` instead of `ANTHROPIC_AUTH_TOKEN`
Still hitting Anthropic directly	A `settings.json` `env` block is overriding your shell exports
Works, but the model is wrong	The model variable name differs in your version — see the note above