By @spyker
What is your idea about?
The last Budget round took the SEAL ecosystem from an idea to a production-ready search solution. This round takes it to the next level: a chunk-level, incremental and cost-aware vectorization pipeline, more quality in the output related to WCAG, a index bridge for existing ke_search projects, and a small MCP interface exposing TYPO3 search results to AI agents.
What must be achieve by 30th of November 2026?
EXT:index 3.0: Consolidation and Migration Paths
- Generic configuration-driven indexer: any TCA table becomes indexable through settings.yaml — field mapping, tag derivation, geo fields — without writing PHP. Replaces the extension-by-extension approach with a scalable one.
- ke_search bridge with a genuine migration path: existing ke_search plugins keep working while flexible and powerful indexing moves to EXT:index (separated extensions)
- EXT:sf_event_mgt shipped as Extender configuration.
EXT:seal 2.0: Result Quality and Accessibility
- Keyword highlighting base on SEAL project
- WCAG 2.2 conformance of the search plugin: ARIA live regions for result counts, focus management on facet changes, keyboard-operable filters, screen-reader-appropriate markup for the AI summary. Addresses BFSG and EAA obligations.
EXT:seal_ai 2.0: Vectorization Pipeline and Hybrid Retrieval
- Baseline alignment with current Symfony AI: store lifecycle including StoreInterface::clear(), StoreFactory and DSN configuration, FailoverPlatform for production resilience, current platform bridges.
- EXZ:aim integration as model/provider for AI functions as alternative for Symfony AI to get more transparency in the TYPO3 backend related to the AI integration.
- Chunking strategies: configurable semantic chunking with adjustable overlap, chunk-level embedding, and parent-document retrieval — matches happen at chunk level, results are presented at page level. Chunk size and strategy configurable.
- Content-hash embedding cache: a hash per chunk over normalized content plus model, dimension and chunking configuration. Only changed chunks are re-embedded; incremental re-indexing replaces full rebuilds. Includes token and cost reporting so the saving is visible.
- read-only MCP server exposing search and retrieve tools, with token-based authentication.
What is the potential impact of your idea for the overall goal?
- Speed up search integrations for TYPO3
- Direct integration into AI Agents for RAG support
- Increased adoption options related to EXT:index & EXT:seal
How does your Idea align with the TYPO3 Association Strategy
- 5.2 Improving Governance → related to the EXT:aim integration
- 5.4 Strengthening the Innovation Process and Product Ownersahip
Budget for this idea?
5.000 Euro
My Name
Tim Lochmüller