I found a bug in ProviderConfig (memori/core/providers.py) where the same http_client parameter is internally used for both synchronous and asynchronous LLM client creation.

Expected Behavior

Memori should allow separate configuration for:

• sync_http_client
• async_http_client

And use them correctly in:

• create_client() → sync_http_client
• create_async_client() → async_http_client

Actual Behavior

Both functions use the same http_client value, which breaks custom LLM providers that rely on different sync/async HTTP clients.

Reproduction

1. Pass a custom synchronous client
2. Pass a custom asynchronous client
3. Memori uses the same client into both constructors

Proposed Fix

I have implemented the following changes in my fork:

• Rename http_client → sync_http_client
• Add new async_http_client field
• Update create_client() and create_async_client() accordingly
• Add backward compatibility for older configs

I will open a Pull Request referencing this Issue.