`EMBEDDING_MODEL` env var is silently ignored — should be `OLLAMA_MODEL`

[NikolaRHristov]

Description

The docker documentation/examples reference EMBEDDING_MODEL as the env var to override the embedding model, but the application never reads it. The actual env var the code checks is OLLAMA_MODEL.

Root Cause

In lib/elixir_nexus/embedding_model.ex, model resolution is:

def model_name do
  System.get_env("OLLAMA_MODEL") || Application.get_env(:elixir_nexus, :ollama_model, @default_model)
end

Passing -e EMBEDDING_MODEL=nomic-embed-text in docker run is silently ignored, causing the app to always fall back to the hardcoded default embeddinggemma:300m.

Steps to Reproduce

docker run -d --name code_nexus -p 4100:4100 -p 3002:3002 \
  -e QDRANT_URL=http://host.docker.internal:6333 \
  -e MCP_HTTP_PORT=3002 \
  -e OLLAMA_URL=http://host.docker.internal:11434 \
  -e EMBEDDING_MODEL=nomic-embed-text \
  -e WORKSPACE_HOST="$HOME/Developer/Application/CodeEditorLand/Land" \
  -v "$HOME/Developer/Application/CodeEditorLand/Land:/workspace:ro" \
  iksnerd/code-nexus:latest

Despite passing EMBEDDING_MODEL=nomic-embed-text, the app uses embeddinggemma:300m.

Fix

Either:

• Rename the env var check in embedding_model.ex from OLLAMA_MODEL to EMBEDDING_MODEL for clarity, or
• Update all documentation and examples to use OLLAMA_MODEL=nomic-embed-text consistently

The correct working command is:

docker run -d --name code_nexus -p 4100:4100 -p 3002:3002 \
  -e QDRANT_URL=http://host.docker.internal:6333 \
  -e MCP_HTTP_PORT=3002 \
  -e OLLAMA_URL=http://host.docker.internal:11434 \
  -e OLLAMA_MODEL=nomic-embed-text \
  -e WORKSPACE_HOST="$HOME/Developer/Application/CodeEditorLand/Land" \
  -v "$HOME/Developer/Application/CodeEditorLand/Land:/workspace:ro" \
  iksnerd/code-nexus:latest

[NikolaRHristov]

Alternative Fix: Update docs to reference OLLAMA_MODEL

The simpler path (no code change needed) is to update the documentation to consistently use OLLAMA_MODEL everywhere EMBEDDING_MODEL is currently referenced.

Specifically, the README or any docker usage examples should replace:

-e EMBEDDING_MODEL=nomic-embed-text

with:

-e OLLAMA_MODEL=nomic-embed-text

CLAUDE.md already uses the correct name (OLLAMA_MODEL=nomic-embed-text), so it's just a matter of making sure all other docs, examples, and docker-compose snippets are in sync with that.

[iksnerd]

Fixed in v1.13.0: embedding_model.ex now checks OLLAMA_MODEL first, then EMBEDDING_MODEL as an alias, then the app config default. Both env vars work.