ChickenBenny commented on issue #144:
URL: https://github.com/apache/flink-agents/issues/144#issuecomment-6825873215
Thanks @wenjin272! Yeah, a retrieval tool makes more sense to me too. I'd
like to work on it.
Here's roughly what I'm thinking:
```python
ResourceDescriptor(
clazz=ResourceName.Tool.VECTOR_STORE_SEARCH,
vector_store="knowledge_base",
max_results=3,
name="search_flink_docs",
description="Search the Flink documentation.",
)
```
- The LLM only passes `query`. `max_results` is set in config so the results
don't get too long.
- It returns each doc's `content` and `metadata`, so the model can tell
where the info came from.
- If nothing is found, it returns something like "No relevant documents
found." instead of an empty string.
- If the vector store name is wrong or the query fails, it's reported as a
tool error.
- I'd leave filters and scores out for now and add them later if needed.
- I'll keep the current RAG example and add a new one with `ReActAgent` +
this tool.
A couple of questions:
1. There's no `ResourceName.Tool` yet. Is it OK to add one (plus the YAML
alias)?
2. Some vector stores are Java-only (OpenSearch, Milvus, pgvector, etc.).
Should I start with Java or Python?
If this sounds good, I'll start with a small PR.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
