RagCode update: lightweight models, easier model switching, and smarter indexing

[doITmagic]

Hi everyone 👋

We’ve shipped a set of improvements to make RagCode easier to run and easier to tune, even on smaller machines.

What’s new

1) Works well with small local models

• RagCode can run with lightweight Ollama models, including sub-1B options.
• This is useful for laptops, low-resource environments, and fast local workflows.

2) Model switching is now easier

You can change models from:

• your IDE MCP server config (environment values), or
• your global RagCode config.

On startup, RagCode now resolves the active model configuration consistently.

3) Automatic model download on startup

• If a required model is missing in Ollama, RagCode now tries to download it automatically.
• Startup fails only if the download itself fails (for example: wrong model name, network issue, etc.).

4) Safer handling of embedding model differences

• RagCode checks embedding vector dimensions and handles migration/reindex when needed.
• This helps avoid index issues when switching embedding models.

5) Smarter indexing behavior

• Better incremental reindexing when files change.
• Better watcher/debounce flow for automatic updates.
• Non-code languages without analyzers (like YAML/Markdown in code-index paths) no longer trigger noisy indexing errors.

6) Skills support remains available

• You can list and install workspace skills to guide behavior for specific workflows.
• This helps teams keep conventions consistent with less manual setup.

Practical recommendation

Small models are supported and useful, but for day-to-day code quality we still recommend using a larger coding model (for example, 4B+) if your hardware allows it.

A practical setup is:

• smaller model for lightweight/dev loops,
• larger model for deeper code understanding and complex tasks.