Add Oracle AI Vector Search integration page#443
Conversation
|
Thank you, @fede-kamel! I'll have a look once the integration itself is merged |
|
Hi @kacperlukawski — the integration PR has been merged: deepset-ai/haystack-core-integrations#3096 Full e2e validation passed locally against Oracle 23ai (Docker):
This documentation PR is ready for review whenever you get a chance. Thanks! |
|
Hi @kacperlukawski — gentle nudge on this one. The code integration (deepset-ai/haystack-core-integrations#3096) has been merged for 3 weeks and we'd like to announce the Oracle AI Vector Search integration publicly, but we'd prefer the listing page be live first so users have somewhere to land. Is there anything blocking the review, or changes you'd like me to make? Happy to iterate. cc @davidsbatista — since you merged the code PR, wanted to loop you in so you're aware of the announcement dependency. |
| repo: https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/oracle | ||
| type: Document Store | ||
| report_issue: https://github.com/deepset-ai/haystack-core-integrations/issues | ||
| logo: /logos/oracle.png |
There was a problem hiding this comment.
The logo is missing. Could you please add it in this PR?
There was a problem hiding this comment.
Done — added the Oracle wordmark at /logos/oracle.png in the same commit. Thanks for the review!
| ```bash | ||
| pip install oracle-haystack | ||
| ``` | ||
|
|
||
| The only runtime dependency beyond Haystack itself is [python-oracledb](https://python-oracledb.readthedocs.io/) (thin mode — no Oracle Client libraries required). |
There was a problem hiding this comment.
It sounds a bit like installing python-oracledb is still required, even though we installed oracle-haystack. Isn't it going to be installed as a sub-dependency of the oracle-haystack package? If it's installed automatically, I would suggest removing the reference to python-oracledb.
There was a problem hiding this comment.
Good catch — python-oracledb is pulled in transitively by oracle-haystack, so users do not need to install it separately. Reworded that line to focus on the genuinely useful part for Oracle users (no separate Oracle Client / Instant Client install needed, since the driver defaults to thin mode).
| For faster approximate nearest-neighbor search on large collections, create an HNSW index after writing documents: | ||
|
|
||
| ```python | ||
| document_store.create_hnsw_index() | ||
| ``` |
There was a problem hiding this comment.
Do we need to call it once, or after each document ingestion?
There was a problem hiding this comment.
Once. Oracle maintains the HNSW index incrementally as new documents are written, so there is no need to rebuild it after each ingestion. The call is idempotent too (CREATE VECTOR INDEX IF NOT EXISTS). I clarified this in the doc and also pointed out the create_index=True constructor shortcut for users who want the index created automatically on initialization.
|
I'm sorry, @fede-kamel I missed the notification, but finally did the first review |
|
Thanks for the review @kacperlukawski! Pushed a362199 addressing all three comments:
Ready for another look. |
|
@kacperlukawski one more — I also added a white-on-transparent variant at |
|
Hi @kacperlukawski — just a quick check-in. The review feedback was addressed in a362199 (logo, install wording, HNSW clarification) and the white-on-transparent logo variant was added in 12d1083. Let me know if anything else is needed — happy to iterate. Thanks! |
|
@kacperlukawski — friendly bump on this. All three review comments from Apr 29 have been addressed in the latest push:
Ready for re-review whenever you have a moment. Thanks! cc @davidsbatista (who reviewed and merged the code-integration PR |
Add documentation for the oracle-haystack integration providing OracleDocumentStore and OracleEmbeddingRetriever backed by Oracle Database 23ai/26ai native VECTOR type. Related: deepset-ai/haystack-core-integrations#3096
- Add Oracle wordmark logo at /logos/oracle.png (referenced in front-matter) - Reword install note so it no longer reads as a separate dependency - Clarify that create_hnsw_index() is a one-time call (Oracle maintains the index incrementally) and document the create_index=True shortcut
For dark-themed listings if the site supports a logo variant per theme.
fede-kamel
force-pushed
the
feat/oracle-integration
branch
from
12d1083 to
4131fbc
Compare
kacperlukawski
left a comment
kacperlukawski
left a comment
There was a problem hiding this comment.
@fede-kamel thanks for contributing!
2 participants
Summary
oracle-haystack— Oracle AI Vector Search Document Store for HaystackThe underlying code integration is being reviewed in deepset-ai/haystack-core-integrations#3096.
cc @davidsbatista — who is reviewing the code integration PR.
Checklist
draft-integration.mdtemplate