Claude Code

Claude Code treats any host that speaks the Anthropic Messages format as a gateway. Grafilab serves that format at https://llm.grafilab.ai/v1/messages, so pointing the CLI at Grafilab takes two environment variables plus a model id.

• Claude Code installed and working. Run claude --version to confirm.

• A Grafilab API key. In the console open API (https://app.grafilab.ai/api) and click Generate New Key — see API Keys.

Name the key claude-code so its spend is separated from your other tools.

The model catalog is live, so list it rather than copying ids from a page. Every entry’s id is the exact string to use in any tool or SDK.

curl https://llm.grafilab.ai/v1/models \
  -H "Authorization: Bearer $GRAFILAB_API_KEY" | jq -r '.data[].id'

from openai import OpenAI

client = OpenAI(api_key="sk-grafilab-...", base_url="https://llm.grafilab.ai/v1")
for model in client.models.list():
    print(model.id)

import OpenAI from "openai";

const client = new OpenAI({ apiKey: "sk-grafilab-...", baseURL: "https://llm.grafilab.ai/v1" });
for await (const model of client.models.list()) {
  console.log(model.id);
}

The same ids appear as cards in the Playground, with context size and prices.

The Anthropic surface has no model listing of its own, so list the catalog on the OpenAI surface as above and paste the id into ANTHROPIC_MODEL.

Set the base URL to the bare origin. Claude Code appends /v1/messages itself, so a base URL ending in /v1 produces requests to /v1/v1/messages and a 404.

export ANTHROPIC_BASE_URL="https://llm.grafilab.ai"
export ANTHROPIC_AUTH_TOKEN="sk-grafilab-..."
export ANTHROPIC_MODEL="grafilab/qwen3.6-flash"

claude

$env:ANTHROPIC_BASE_URL = "https://llm.grafilab.ai"
$env:ANTHROPIC_AUTH_TOKEN = "sk-grafilab-..."
$env:ANTHROPIC_MODEL = "grafilab/qwen3.6-flash"

claude

ANTHROPIC_AUTH_TOKEN sends the key as Authorization: Bearer, which Grafilab accepts. Do not also set ANTHROPIC_API_KEY: it sends the key as x-api-key, which Grafilab also accepts, but having both set means Claude Code prompts you to approve the API key before it takes over, and a startup warning about two credential sources. Pick one, and prefer ANTHROPIC_AUTH_TOKEN.

ANTHROPIC_MODEL only covers the session’s main model. Claude Code also calls a small model for background work and spawns subagents, and both default to Anthropic model names that Grafilab does not serve. Set them to a Grafilab id as well, or those calls fail with 404:

export ANTHROPIC_DEFAULT_HAIKU_MODEL="grafilab/qwen3.6-flash"
export CLAUDE_CODE_SUBAGENT_MODEL="grafilab/qwen3.6-flash"

$env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "grafilab/qwen3.6-flash"
$env:CLAUDE_CODE_SUBAGENT_MODEL = "grafilab/qwen3.6-flash"

ANTHROPIC_DEFAULT_HAIKU_MODEL is what the haiku alias resolves to and what Claude Code uses for background work; ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_OPUS_MODEL do the same for sonnet and opus, so set those too if you switch models with /model. ANTHROPIC_SMALL_FAST_MODEL is the deprecated predecessor of the haiku variable — use the new name.

Exports last one terminal session. Two ways to keep the configuration, depending on whether you want every Claude Code session on Grafilab or only some.

Put the variables in the env block of a Claude Code settings file. ~/.claude/settings.json applies to all your projects; .claude/settings.local.json applies to one project and is gitignored for you. A settings file also reaches background agents, which shell exports do not reliably do.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://llm.grafilab.ai",
    "ANTHROPIC_AUTH_TOKEN": "sk-grafilab-...",
    "ANTHROPIC_MODEL": "grafilab/qwen3.6-flash",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "grafilab/qwen3.6-flash",
    "CLAUDE_CODE_SUBAGENT_MODEL": "grafilab/qwen3.6-flash"
  }
}

Settings files are strict JSON: a trailing comma or a // comment makes Claude Code report a settings error at the next start.

If you also use Claude Code against Anthropic, wrap the Grafilab configuration in a function so plain claude stays untouched and gclaude runs on Grafilab.

gclaude() {
  ANTHROPIC_BASE_URL="https://llm.grafilab.ai" \
  ANTHROPIC_AUTH_TOKEN="sk-grafilab-..." \
  ANTHROPIC_MODEL="grafilab/qwen3.6-flash" \
  ANTHROPIC_DEFAULT_HAIKU_MODEL="grafilab/qwen3.6-flash" \
  CLAUDE_CODE_SUBAGENT_MODEL="grafilab/qwen3.6-flash" \
  command claude "$@"
}

function gclaude {
  $env:ANTHROPIC_BASE_URL = "https://llm.grafilab.ai"
  $env:ANTHROPIC_AUTH_TOKEN = "sk-grafilab-..."
  $env:ANTHROPIC_MODEL = "grafilab/qwen3.6-flash"
  $env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "grafilab/qwen3.6-flash"
  $env:CLAUDE_CODE_SUBAGENT_MODEL = "grafilab/qwen3.6-flash"
  try { claude @args }
  finally {
    Remove-Item Env:ANTHROPIC_BASE_URL, Env:ANTHROPIC_AUTH_TOKEN, Env:ANTHROPIC_MODEL,
      Env:ANTHROPIC_DEFAULT_HAIKU_MODEL, Env:CLAUDE_CODE_SUBAGENT_MODEL -ErrorAction SilentlyContinue
  }
}

The finally block matters on PowerShell: $env: assignments outlive the function and would leak into the next plain claude you run in that window. The bash form sets the variables for one command only, so it needs no cleanup.

If you keep both setups, remember that a settings-file env value wins over a shell export of the same variable.

1. Test the endpoint before opening Claude Code, so a failure points at the gateway rather than at the CLI.

   curl -X POST "$ANTHROPIC_BASE_URL/v1/messages" \
     -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN" \
     -H "content-type: application/json" \
     -d '{"model": "grafilab/qwen3.6-flash", "max_tokens": 16, "messages": [{"role": "user", "content": "Say hi."}]}'

   A body starting with {"id":"msg_ means the URL and the key both work.

2. Start a session and send a prompt.

   claude "say hi"

3. Run /status and read the Status tab. Two lines confirm the routing:

   Anthropic base URL: https://llm.grafilab.ai
   Auth token: ANTHROPIC_AUTH_TOKEN

   If the base URL line is missing, the variable did not reach the session — start claude from the shell that has the exports, or move them into a settings file.

4. Run /cost. Token counts and spend appear as usual; the cache-write figure stays at zero, which is expected and explained below.

Then open API in the console and select View Usage on the claude-code key to see the same requests from Grafilab’s side. Rollups can lag by up to about 30 minutes.

Grafilab implements a subset of the Messages protocol. Everything Claude Code needs for editing, running commands, and calling your MCP servers is there; these are the edges you may notice.