mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-10 08:28:25 +00:00
- one table (litellm/router_endpoint_table.py) names each factory-built endpoint, its host op, call type and dispatch; both cores bind and dispatch from it, and an unknown call type is refused when bound - the endpoints that resolved to None without a call now dispatch like their siblings; sync create/get/delete/cancel_interaction answer synchronously - sync search picks among the router's search tools like asearch - ~80 endpoints bound to litellm functions (and agenerate_content's overwritten binding, avector_store_create) now go through host.invoke - routing decisions key on the endpoint's call type, never the host function's __name__ - arerank no longer sends the builtin input; responses get/cancel/delete/list with a plain id and no model call the handler directly - prompt-management groups work without a litellm_call_id or logging object |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_vector_store_access_control.py | ||
| test_vector_store_endpoints.py | ||
| test_vector_store_rbac.py | ||
| test_vector_store_tenant_guard.py | ||
| test_vector_store_utils.py | ||