Add azure_anthropic backend for Anthropic models on Azure AI Foundry

[timothyjlaurent]

Summary

Add support for an azure_anthropic backend that enables using Anthropic models (Claude) hosted on Azure AI Foundry.

Motivation

Azure AI Foundry / Azure AI Model Catalog now offers Anthropic models as a managed service. Organizations using Azure often need to route API calls through their Azure endpoints for compliance, billing, and network policy reasons. The existing anthropic backend only targets api.anthropic.com, and the existing azure_openai backend only works with OpenAI models on Azure.

Proposed Implementation

The Anthropic Python SDK already supports a base_url parameter, so the implementation is straightforward — a thin wrapper around the existing AnthropicClient that reads Azure-specific config:

New file: rlm/rlm/clients/azure_anthropic.py

• Mirrors the structure of azure_openai.py and anthropic.py
• Reads env vars: AZURE_ANTHROPIC_API_KEY, AZURE_ANTHROPIC_ENDPOINT, AZURE_ANTHROPIC_API_VERSION
• Passes base_url=azure_endpoint to anthropic.Anthropic()
• Full usage tracking, sync/async support, system prompt extraction

Changes to existing files:

• rlm/rlm/clients/__init__.py — add azure_anthropic branch to get_client()
• rlm/rlm/core/types.py — add "azure_anthropic" to ClientBackend Literal

Usage:

export AZURE_ANTHROPIC_API_KEY="your-key"
export AZURE_ANTHROPIC_ENDPOINT="https://your-resource.services.ai.azure.com/v1"

rlm ask . -q "Summarize this repo" --backend azure_anthropic --model claude-opus-4-6

Or via config:

backend: azure_anthropic
model: claude-opus-4-6
backend_kwargs:
  azure_endpoint: https://your-resource.services.ai.azure.com/v1

Reference Implementation

I have a working implementation at https://github.com/rawwerks/rlm-cli — happy to open a PR if this approach looks good.

[timothyjlaurent]

Moved to upstream: alexzhang13/rlm#100 / alexzhang13/rlm#101