Alloneia
← Docs

Use l3k with Claude Code

Anthropic's Claude Code CLI works against any backend that speaks the Anthropic Messages API. l3k speaks OpenAI Chat Completions, so we point Claude Code at a tiny local router that translates between the two — no code changes, no patched binary.

Why a local router?

Claude Code sends `POST /v1/messages` (Anthropic shape). l3k accepts `POST /v1/chat/completions` (OpenAI shape). The router accepts the first and forwards the second, including streaming, tool calls and image parts.

Option A — claude-code-router (recommended)

Single npm install, ~10 lines of config, ships a `ccr` command that boots the router and Claude Code together.

  1. 1. Install
    npm install -g @musistudio/claude-code-router
  2. 2. Configure

    Create ~/.claude-code-router/config.json

    {
  "Providers": [
    {
      "name": "l3k",
      "api_base_url": "https://alloneia.com/v1/chat/completions",
      "api_key": "sk-l3k-YOUR_KEY",
      "models": [
        "anthropic/claude-3.5-haiku",
        "anthropic/claude-3.5-haiku"
      ]
    }
  ],
  "Router": {
    "default":    "l3k,anthropic/claude-3.5-haiku",
    "background": "l3k,anthropic/claude-3.5-haiku",
    "think":      "l3k,anthropic/claude-3.5-haiku",
    "longContext":"l3k,anthropic/claude-3.5-haiku"
  }
}

    The two model slugs above are pulled live from your active catalog. Replace with any model id from /docs.

  3. 3. Run
    # Inside any project
ccr code

# ccr boots the local router and launches Claude Code pointed at it.
# All requests now go: Claude Code → ccr → l3k → upstream provider.

Option B — LiteLLM proxy

If you already run LiteLLM for other tools, register l3k as an OpenAI-compatible upstream and point Claude Code at LiteLLM's Anthropic endpoint:

# litellm config.yaml
model_list:
  - model_name: anthropic/claude-3-5-sonnet
    litellm_params:
      model: openai/anthropic/claude-3.5-haiku
      api_base: https://alloneia.com/v1
      api_key: sk-l3k-YOUR_KEY

# Run the proxy
litellm --config config.yaml --port 4000

# Point Claude Code at it
export ANTHROPIC_BASE_URL=http://localhost:4000
export ANTHROPIC_AUTH_TOKEN=any-non-empty-string
claude

LiteLLM exposes `/v1/messages` automatically when the registered model name starts with `anthropic/…`.

Verify it's working

After your first prompt:

  • Tokens land on /usage in the dashboard within a few seconds. If they don't, the router isn't reaching l3k — check the error shape against /docs/errors.
  • Each request shows the API key prefix on /keys (last used timestamp updates).
  • `curl https://alloneia.com/v1/models -H 'Authorization: Bearer sk-l3k-…'` returns the catalog this key is allowed to use.

Troubleshooting

  • 401 invalid_api_key — Wrong key, key revoked, or key expired. Issue a fresh one from the dashboard.
  • 404 model_not_found — The slug in your config isn't in the live catalog. Hit `/v1/models` or check /docs.
  • 402 insufficient_balance — Top up at https://www.alloneia.com/billing. Claude Code makes large parallel requests — keep at least a few dollars of headroom.
  • 429 rate_limit_exceeded — Per-key RPM limit hit. Raise it from the dashboard or split work across two keys.