Claude Code

Use Claude Code with Gab AI through the Anthropic-compatible /v1/messages endpoint.

Prerequisites

• A Gab AI Plus subscription
• A Gab AI API key
• Claude Code installed locally

Configuration

Claude Code expects Anthropic's Messages API and appends /v1/messages itself. Set the base URL to the root domain, not https://gab.ai/v1. Use ANTHROPIC_MODEL=auto for the agent/main model only, and set a cheap explicit ANTHROPIC_SMALL_FAST_MODEL (arya or the admin quick default) for compact and title calls. GET /v1/agent-setup?client=claude-code returns the env vars to copy.

export ANTHROPIC_BASE_URL="https://gab.ai"
export ANTHROPIC_AUTH_TOKEN="YOUR_GAB_API_KEY"
export ANTHROPIC_MODEL="auto"
export ANTHROPIC_SMALL_FAST_MODEL="arya"

claude

Base URL

Use https://gab.ai for Claude Code. If you set https://gab.ai/v1, Claude Code may call /v1/v1/messages.

Supported Features

• Anthropic /v1/messages request and response shape
• Streaming server-sent events, including tool_use input_json_delta
• Tool use with tool_use and tool_result blocks
• Prompt caching via cache_control (auto-marked on system + last tool when omitted)
• Extended thinking / reasoning_effort
• Model aliases (anthropic/claude-sonnet-4-5, date-stamped Claude ids)
• /v1/messages/count_tokens compatibility
• stop_sequences, temperature, top_p, top_k, and max_tokens
• Image blocks from URL, base64, or Gab AI uploaded file_id

Smoke Test

You can validate the endpoint directly with cURL before launching Claude Code:

curl https://gab.ai/v1/messages \\
  -H "x-api-key: YOUR_GAB_API_KEY" \\
  -H "anthropic-version: 2023-06-01" \\
  -H "content-type: application/json" \\
  -d '{
    "model": "claude-sonnet-4-5",
    "max_tokens": 128,
    "messages": [
      {
        "role": "user",
        "content": [{ "type": "text", "text": "Say hello from Claude Code." }]
      }
    ]
  }'

Troubleshooting

• 401 authentication_error: confirm ANTHROPIC_AUTH_TOKEN is your Gab AI API key.
• 404 or duplicated /v1 path: set ANTHROPIC_BASE_URL to https://gab.ai, not https://gab.ai/v1.
• unsupported tool use: choose a model whose /v1/models capability function_calling is true.
• cache_read_input_tokens stays 0: send cache_control or rely on auto-marking of system + last tool; the prefix must be identical across turns.
• insufficient credits: API access consumes Gab AI credits and requires Plus. Cache hits are billed cheaper.

Last updated: 2026-09-08