Repository navigation
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: af95d9e34f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| "source": [ | ||
| "### Metadata rules\n", | ||
| "\n", | ||
| "Metadata is a flat dict of `str`, `int`, `float`, or `bool` values, with keys that start with a letter and contain only letters, digits, and underscores. Its serialized size is at most 8 KB, and PageIndex Cloud rejects metadata that breaks a rule with `PageIndexAPIError`. See the [metadata documentation](https://docs.pageindex.ai/sdk/documents#metadata) for details.\n" |
There was a problem hiding this comment.
When a reader supplies float("nan"), float("inf"), or float("-inf"), the value satisfies the stated float rule but PageIndex Cloud rejects the submission. Add the finite-number restriction to this rules summary so users do not receive an unexpected PageIndexAPIError after following the notebook's guidance.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in ab24ee2: the rules now say numbers must be finite, so no NaN or Infinity.
| "source": [ | ||
| "messages = [{\"role\": \"user\", \"content\": \"What revenue did the finance team's 2025 earnings release report?\"}]\n", | ||
| "\n", | ||
| "for event in client.chat(messages, stream=True, protocol=\"responses\"):\n", |
There was a problem hiding this comment.
Both chat calls search the user's entire cloud library because neither doc_id nor a dedicated folder_id is passed. If the account already contains documents with matching tags—or if the notebook is rerun and creates suffixed duplicates—the answers can include unrelated files or become ambiguous instead of reproducing the displayed two-document result; scope the calls to doc_ids.values() or upload into an isolated folder.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Partly fixed in ab24ee2. Step 02 now reuses a document already in the library (get_document_id()), so a rerun no longer uploads suffixed duplicates with the same tags. The chat calls stay unscoped on purpose: the point of the notebook is that the agent finds documents across the whole library by their tags. If the library holds other documents with matching tags, the agent finding them is the intended behavior.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ab24ee27b3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
If a reader previously uploaded any of these example PDFs without tags, get_document_id() still finds it and this branch skips the only call that attaches metadata. The read-back step then returns missing or different tags, and the tag-based questions cannot reproduce the demonstrated results, especially the research-team query whose ownership exists only in metadata. Check that the existing document's metadata matches DOCS before reusing it; otherwise explain how to replace it with a tagged upload, since the notebook states metadata cannot be updated.
Useful? React with 👍 / 👎.
…nd filters by tag
This branch has not been deployed
1 participant
Adds
cookbook/pageindex-metadata.ipynb, following Metadata in the SDK docs. It declares a folder's metadata fields (type,team,year) withcreate_metadata_schema(), uploads four PDFs fromexamples/documents/into the folder with matching tags, reads them back, and then asks two questions only the tags can answer:get_metadata_schema, thensearch_documents(mode="metadata")withgroup_by: ["team"]and$count.search_documents(mode="metadata")withmetadata_filter: {"year": {"$gte": 2019}}.The layout follows the other cloud chat notebooks and ends with the optional usage and cost estimate. A rerun reuses the folder and the documents.
Needs a PageIndex API key and an OpenAI key. A run creates one folder, uploads four small PDFs and takes about 5 minutes.
Merge after the SDK release that ships #618 (
create_metadata_schema). Until then,pip install -U pageindexdoes not have it.