Indexed repositories
What’s indexed
The corpus includes:- Issue resolutions — closed issues with confirmed fixes
- Pull requests — merged PRs with linked issues
- Discussions — community Q&A with accepted answers
- Release notes — version-specific changes and known issues
How indexing works
1
Ingestion
Issues, PRs, discussions, and release notes are collected from each repository via the GitHub API.
2
Chunking
Documents are split into semantically meaningful chunks — preserving context around error messages, stack traces, and configuration snippets.
3
Embedding
Each chunk is embedded using a transformer model that captures the semantic meaning of Kubernetes errors, configuration patterns, and troubleshooting procedures.
4
Storage
Embeddings are stored in a vector database optimized for high-recall semantic search.
5
Retrieval
On query, semantic search retrieves the top-k matching chunks, which are passed as context to the LLM for grounded response generation.
Why embeddings matter
Traditional keyword search fails for Kubernetes troubleshooting. An error likefailed to create pod sandbox has dozens of root causes — CNI misconfiguration, disk pressure, container runtime issues, and more.
Semantic embeddings capture the meaning behind error messages and stack traces, not just the keywords. This means Oracle can match a user’s error against resolutions that describe the same underlying problem in different terms.
Corpus updates
The corpus is updated regularly as new issues are resolved across the indexed repositories. New resolutions, PRs, and discussions are ingested, chunked, embedded, and added to the vector database on a recurring schedule.Enterprise customers can request custom corpus additions — internal runbooks, private repositories, or domain-specific documentation indexed alongside the public corpus.