RAG Logs
RAG Logs provide complete visibility into how enterprise knowledge is ingested, processed, stored, and retrieved inside Karyam.
Unlike traditional AI systems where retrieval happens behind the scenes, Karyam exposes every stage of the RAG pipeline.
This allows teams to understand:
- Which files were processed
- Which embedding model was used
- How artifacts were generated
- When vectors were stored
- Which chunks were retrieved
- Which chunks were discarded
- What context was sent to the LLM
- Why retrieval failed
Where to Find RAG Logs
Section titled “Where to Find RAG Logs”RAG execution logs are available from:
AI Infra └── RAG RunsThe RAG Runs page provides complete visibility into both ingestion and retrieval operations across the workspace.
RAG Run Types
Section titled “RAG Run Types”Karyam tracks two categories of RAG operations.
Ingestion Runs
Section titled “Ingestion Runs”Responsible for transforming enterprise knowledge into searchable embeddings.
File Upload ↓Chunk Generation ↓Embedding Creation ↓Vector StorageRetrieval Runs
Section titled “Retrieval Runs”Responsible for retrieving relevant knowledge during execution.
User Query ↓Vector Search ↓Chunk Filtering ↓Context Building ↓LLM ResponseSupported Run Types
Section titled “Supported Run Types”| Type | Description |
|---|---|
| INGESTION | Knowledge ingestion and indexing |
| RETRIEVAL | Runtime knowledge retrieval |
Ingestion Events
Section titled “Ingestion Events”rag_ingestion_started
Section titled “rag_ingestion_started”Generated when an ingestion pipeline begins.
Example:
rag_ingestion_startedFile: vpn-policy.pdfType: INGESTIONStatus: RUNNINGrag_model_resolution
Section titled “rag_model_resolution”Generated when Karyam resolves the embedding model configured for the listener.
Example:
rag_model_resolutionModel: text-embedding-3-largeProvider: OpenAIStatus: SUCCESSrag_artifacts_generated
Section titled “rag_artifacts_generated”Generated after artifacts are successfully extracted from uploaded files.
Examples include:
- Text extraction
- PDF parsing
- Image OCR
- Metadata generation
Example:
rag_artifacts_generatedArtifacts: 14Status: SUCCESSrag_artifacts_skipped
Section titled “rag_artifacts_skipped”Generated when an artifact is intentionally skipped.
Examples include:
- Unsupported file types
- Sidecar files
- Duplicate processing
Example:
rag_artifacts_skippedReason: Sidecar description fileStatus: INFOrag_embedding_started
Section titled “rag_embedding_started”Generated when embedding generation begins.
Example:
rag_embedding_startedChunks: 42Status: RUNNINGrag_embedding_completed
Section titled “rag_embedding_completed”Generated when embedding generation completes successfully.
Example:
rag_embedding_completedEmbeddings Generated: 42Status: SUCCESSrag_vector_store_started
Section titled “rag_vector_store_started”Generated when vectors begin writing to the configured vector database.
Example:
rag_vector_store_startedVector DB: QdrantStatus: RUNNINGrag_vector_store_completed
Section titled “rag_vector_store_completed”Generated when vector storage completes.
Example:
rag_vector_store_completedVectors Stored: 42Status: SUCCESSrag_ingestion_completed
Section titled “rag_ingestion_completed”Generated when ingestion finishes successfully.
Example:
rag_ingestion_completedDuration: 5.3sStatus: SUCCESSRetrieval Events
Section titled “Retrieval Events”rag_retrieval_started
Section titled “rag_retrieval_started”Generated when retrieval begins for an agent or workflow request.
Example:
rag_retrieval_startedQuery:"How do I restore VPN access?"Status: RUNNINGrag_search_started
Section titled “rag_search_started”Generated when vector search begins.
Example:
rag_search_startedTopK: 10Status: RUNNINGrag_search_completed
Section titled “rag_search_completed”Generated after vector search completes.
Example:
rag_search_completedResults Found: 8Status: SUCCESSrag_chunk_retrieved
Section titled “rag_chunk_retrieved”Generated for every retrieved chunk selected for ranking.
Example:
rag_chunk_retrievedFile: vpn-policy.pdfScore: 0.92Status: SUCCESSrag_chunk_discarded
Section titled “rag_chunk_discarded”Generated when a retrieved chunk is discarded.
Examples include:
- Low similarity score
- Duplicate content
- Filtering rules
Example:
rag_chunk_discardedReason: Similarity below thresholdStatus: INFOrag_chunk_filtered
Section titled “rag_chunk_filtered”Generated after chunk filtering completes.
Example:
rag_chunk_filteredSelected Chunks: 5Discarded Chunks: 3Status: SUCCESSrag_context_built
Section titled “rag_context_built”Generated after the final retrieval context is prepared.
Example:
rag_context_builtChunks Included: 5Context Size: 6200 tokensStatus: SUCCESSrag_llm_started
Section titled “rag_llm_started”Generated before the final context is sent to the language model.
Example:
rag_llm_startedModel: Claude SonnetStatus: RUNNINGrag_llm_completed
Section titled “rag_llm_completed”Generated after the language model produces a response.
Example:
rag_llm_completedCompletion Tokens: 721Status: SUCCESSrag_retrieval_completed
Section titled “rag_retrieval_completed”Generated when retrieval execution completes.
Example:
rag_retrieval_completedDuration: 1.8sStatus: SUCCESSrag_error
Section titled “rag_error”Generated whenever ingestion or retrieval fails.
Examples include:
- Embedding failures
- Vector database errors
- Model failures
- Parsing failures
- Retrieval failures
Example:
rag_errorReason: Qdrant connection timeoutStatus: FAILEDStatus Values
Section titled “Status Values”Every RAG event contains a status.
| Status | Description |
|---|---|
| RUNNING | Operation currently executing |
| SUCCESS | Operation completed successfully |
| FAILED | Operation failed |
| INFO | Informational event |
Example Ingestion Timeline
Section titled “Example Ingestion Timeline”rag_ingestion_startedrag_model_resolutionrag_artifacts_generatedrag_embedding_startedrag_embedding_completedrag_vector_store_startedrag_vector_store_completedrag_ingestion_completedExample Retrieval Timeline
Section titled “Example Retrieval Timeline”rag_retrieval_startedrag_search_startedrag_search_completedrag_chunk_retrievedrag_chunk_retrievedrag_chunk_discardedrag_chunk_filteredrag_context_builtrag_llm_startedrag_llm_completedrag_retrieval_completedWhy RAG Logs Matter
Section titled “Why RAG Logs Matter”RAG Logs provide:
- Retrieval transparency
- Knowledge debugging
- Embedding visibility
- Vector database observability
- Production troubleshooting
This allows teams to understand exactly how enterprise knowledge influences AI decisions.
Next Steps
Section titled “Next Steps”Continue with:
- Execution Traces
- Token Usage
- Cost Analytics
- Debugging