[Feature] Support .epub file indexing

[Adrien-Tertras]

Feature Description

I would like GAIA to support: Support .epub file indexing

Use Case

Describe how you would use this feature:

Current Workaround

Is there a workaround you are currently using?

---

Environment:

• GAIA Version: 0.18.1
• Platform: Win32
• User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) GAIAAgentUI/0.18.1 Chrome/144.0.7559.236 Electron/40.8.5 Safari/537.36
• Interface: Agent UI (Web)

Submitted from GAIA Agent UI

[github-actions]

Thanks for the request! EPUB indexing isn't supported today — the RAG ingest path dispatches on file extension and currently handles PDF, plain text/markdown (.txt, .md, .rst, .log), CSV, JSON, Excel, and a wide range of source-code formats, but has no EPUB branch (src/gaia/rag/sdk.py:1228).

Adding it should be a fairly contained change:

1. Add an _extract_text_from_epub method alongside the existing extractors. EPUB is just a ZIP of XHTML, so ebooklib + BeautifulSoup (html.parser) is the typical combo — walk book.get_items_of_type(ITEM_DOCUMENT), strip tags, join per-chapter text. Preserve chapter titles as headings so the downstream semantic chunker has clean boundaries.
2. Wire .epub into the dispatch in _extract_text_from_file (around src/gaia/rag/sdk.py:1228) and surface chapter count in the returned metadata, mirroring how PDFs report num_pages.
3. Add ebooklib to the rag extras in setup.py (it's pure-Python, no native deps).
4. Tests: drop a small fixture EPUB under tests/fixtures/ and add a unit test mirroring the existing PDF extraction test. Update the supported-formats list in docs/sdk/sdks/rag.mdx.

Could you share a bit about the use case (technical manuals, books, public-domain corpora, …)? That would help prioritise — and let us know if image-heavy EPUBs need to route through VLM the way scanned PDFs do today.

PRs welcome — see CONTRIBUTING.md and docs/reference/dev.mdx.