[FEATURE] Separate Embedding Base URL Configuration

[pmffromspace]

Problem

Currently SurfSense appears to use a single OLLAMA_BASE_URL for both:

• chat/completion models
• embedding models

This makes it impossible to:

• run embeddings on a separate Ollama instance
• use a dedicated embedding server
• split workloads across GPUs/machines
• combine Ollama chat with another embedding backend

At the moment only the embedding model itself seems configurable:

OLLAMA_BASE_URL=http://localhost:11434
EMBEDDING_MODEL=nomic-embed-text

but not a separate embedding endpoint/base URL.

---

Requested Feature

Add support for a dedicated embedding base URL, for example:

EMBEDDING_BASE_URL=http://localhost:11435

or provider-specific variants such as:

OLLAMA_EMBEDDING_BASE_URL=

---

Example Use Cases

Separate Ollama instances

OLLAMA_BASE_URL=http://ollama-chat:11434
EMBEDDING_BASE_URL=http://ollama-embed:11434

Dedicated embedding server

OLLAMA_BASE_URL=http://ollama:11434
EMBEDDING_BASE_URL=http://tei:8080

---

Why This Matters

Embedding generation is often:

• GPU intensive
• latency sensitive
• deployed separately in production RAG systems

Separating inference and embedding backends is a common architecture pattern and would improve scalability and deployment flexibility.