Proxy Provider

chimera.providers.proxy implements a Provider that forwards every LLM call to an HTTP relay instead of calling a model API directly. This is useful for centralised key management, cost tracking, or running agents in air-gapped environments.

ProxyProvider

Constructor param	Type	Description
proxy_url	str	Base URL of the relay (trailing slash stripped)
auth_token	str | None	Optional Bearer token for proxy authentication
model	str	Model identifier forwarded in the request body

The provider calls POST {proxy_url}/api/complete for non-streaming completions. The request body is standard Anthropic-compatible JSON (model, messages, tools, temperature, max_tokens).

Expected response schema

{
  "content": "...",
  "tool_calls": [{"id": "...", "name": "...", "arguments": {}}],
  "usage": {"input_tokens": 0, "output_tokens": 0}
}

Self-registration

The module calls register_provider("proxy", _proxy_factory) at import time. The factory requires base_url; api_key is forwarded as the Bearer token.

Usage

from chimera.providers.factory import create_provider

provider = create_provider(
    provider_type="proxy",
    base_url="http://my-relay:8080",
    api_key="secret",
    model="glm-4-flash",
)

# Or via environment / config — same keyword args

The context_window property returns 128000 by default; configure your relay to report the actual limit if needed.