SWIRL Community 4.5 Update

I wrote about SWIRL here last summer. Time for an update.

SWIRL Community 4.5 is out. It is the open-source, Apache-2.0 build of SWIRL: federated search and RAG across your apps, running on your own machine, no vector database required. Three things in this release are worth your attention.

1. Point RAG at any LLM

The big one. Community now lets you configure AI Providers and send your RAG queries to any model you want: OpenAI, Anthropic, Azure OpenAI, or a fully local model behind Ollama or vLLM. Pick the provider, set the model, done.

ai_provider:
  name: anthropic      # openai | azure | ollama | vllm
  model: your-model
  api_key: ${LLM_KEY}

This matters for two reasons. RAG quality is now your choice, not ours, so you can run a stronger model when accuracy counts. And you can keep the whole pipeline local, with retrieval and generation both inside your network. This was an Enterprise feature; it is in Community as of 4.5.

2. Ask RAG a question directly

RAG now accepts a natural-language question as its input. Before, you ran a search and generated over the results; now you can hand SWIRL a plain question and it does retrieval and generation in one step. Question in, cited answer out, which means a lot less glue code to wire SWIRL into a chat box or an app.

3. Updated Galaxy

Community now includes the new Galaxy 5 UI, including a search history widget + dashboard, cleaned up admin interface, built-in activity analytics and more.

That is the release. Free, self-hosted, and now model-agnostic.