Public repo for shared endpoint config used by llm-api-proxy and other consumers.

Keyed by model type: embedding, llm, rerank, visual. Each section has base_url, optional model, optional base_urls[] for fallback. Consumers ask for a type (e.g. embedding) and get the endpoint; which machine serves it is irrelevant. Tokens for model-type lookup: ~/.config/model-type-keys.toml (sections = type or type-1, type-2, …; key = model_key). See workspace Cursor rule model-type-endpoints-lookup.

Local lookup module: scripts/model_type_endpoints_lookup.py (stdlib only: tomllib or tomli). Use in this repo or copy into others; no dependency on another workspace repo.

• get_endpoint(model_type) → { "base_url", "model" } (from hardcoded URL; section order: type, type-1, type-2, …).
• get_model_key(model_type) → API key from ~/.config/model-type-keys.toml, same section order.
• Optional: get_working_endpoint(model_type, ...) (probe base_urls and cache), invalidate_working(model_type=None).

python3 scripts/model_type_endpoints_lookup.py get_endpoint llm
python3 scripts/model_type_endpoints_lookup.py get_model_key llm

Stores dynamic ngrok base URLs per machine (e.g. LM Studio exposed via ngrok).
Updated by llm-api-proxy from config/remote-machine-endpoints.toml via:

cd llm-api-proxy && make publish-remote-endpoints

Consumers can use the raw file URL (e.g. GitHub raw or GitLab) or download it:

• llm-api-proxy: make fetch-remote-endpoints downloads this file into config/remote-machine-endpoints.toml, or set REMOTE_MACHINE_ENDPOINTS_URL to the public URL to load it directly.
• Any Makefile/task: Download the TOML from the public URL and reuse (e.g. curl -sL $(REMOTE_MACHINE_ENDPOINTS_URL) -o config/remote-machine-endpoints.toml).