Two local endpoints: an Anthropic-compatible one (so tools like Claude Code, via ANTHROPIC_BASE_URL, can run against Memo) and an OpenAI-compatible sibling (GET /v1/models, POST /v1/chat/completions), both pointed at Memo's local model or any configured provider/key.
Model selection via a type/model-id format (local/qwen2.5, openai/gpt-4o, ...). Full agentic tool calling for openai/custom/local/groq/openrouter/grok/opencode-zen/opencode-go providers.
"Require API Key" now enforced on both gateways for any non-loopback caller (v4.5.0 security fix) — previously, with remote access on and the key requirement left off, the OpenAI-compatible pair could be reached by anything else able to reach the port, no credential at all.
