#general
Aug 5August 6, 2026
Hi everyone! 👋 I'm Jabir from Nigeria :flag-ng:. I'm a Software Engineering student, frontend developer with over 2 years of experience , and open-source contributor currently learning backend development as I work towards becoming a full-stack developer.
I'm passionate about AI, developer tools, and building products that solve real-world problems. I discovered Archestra through Algora and became interested in the project after exploring the platform, contributor docs, and previous hackathon apps.
I'm excited to be part of the community, learn from everyone, and hopefully contribute to Archestra. Looking forward to connecting with you all! 🚀
We are considering replacing the RAG layer with Onyx, which has more connectors and a more advanced pipeline. Would that be possible?
Read-only live mirror of Archestra.AI Slack
👋Join the discussion withAI enthusiasts!We are considering replacing the RAG layer with Onyx, which has more connectors and a more advanced pipeline. Would that be possible?
4 replies
hey! in fact, we've made it very easy for folks to contribute new connectors 🙂 archestra.ai/docs/platform-adding-knowledge-connectors
I'd be happy to review any PRs and get those merged/released
when you say a "more advanced pipeline", what did you mean exactly?
is there generally any missing RAG functionality which you were hoping for?
Personally I've wrapper RAG in a standalone MCP (on my company's preferred stack) - and then just connected as MCP
Perhaps similar flow may work
@user Thanks, that makes sense. I agree connectors themselves seem straightforward to add in Archestra.
When I said “more advanced pipeline”, I wasn’t referring only to missing connectors. I meant the full
enterprise/legal RAG stack behind retrieval:
- larger connector catalog and mature connector sync/permission-sync behavior;
- OpenSearch/BM25 + vector hybrid search at scale;
- query expansion and reranking;
- contextual retrieval, e.g. document-level summary + chunk-level context added during indexing;
- multipass indexing / multiple chunk granularities;
- document sets / vault-style scoped search;
- image/OCR/vision processing for PDFs and document images;
- stronger citation mapping, so answers can be tied back to exact source chunks;
- context expansion around retrieved chunks;
- observability and controls around indexing/search quality.
For our use case, Archestra would remain the agentic runtime, chat, tools and UI. We are evaluating whether a
headless RAG engine, potentially Onyx-backed, should sit behind Archestra as a retrieval tool.
The fastest path may be exactly what Pavel described: expose the RAG engine as a standalone MCP server and
connect it to Archestra. Longer term, we may also want a native integration/provider so connector management,
document sets, permissions and search results can be surfaced directly inside our UI.
Does Archestra’s built-in Knowledge pipeline currently support things like contextual RAG, multipass indexing,
OpenSearch-style hybrid search, document-level permission sync, image/OCR/vision indexing and citation-level
source mapping? If not, is the intended extension point MCP/custom tools, or would you prefer these
capabilities to be contributed into the native Knowledge system?
👀1