Architecture 

• Domains
  • Assistant domain
  • Connection domain
  • Indexing domain
  • Backend domain
• Pipeline architecture
  • Core classes
  • Processor responsibilities
  • LLM processors
  • Streaming execution
  • Prompt context builders
  • Pipeline tracing
• Indexing architecture
  • Core classes
  • Indexing stages
  • Adapters and connectors
  • JSON and JSONL handling
• JSON and JSONL indexing
  • Configuration fields
  • Dot notation
  • Text fields
  • Metadata fields
  • Metadata mapping
  • Complete example
  • Source identity
  • Operational advice
• Connection architecture
  • AI connector interface
  • Vector store connector interface
  • Connection records
• Shared JSON controller
  • Architecture
  • Controller action
  • Assistant response
  • Security
  • Extending the controller
• Retrieval architecture
  • Data flow
  • Retrieved documents
  • Answer context builder
• Hashing and source identity
  • Source hash
  • Content hash
  • Storage scope
  • Vector document identity
• Data flow
  • Request lifecycle
  • Indexing lifecycle
• Architecture decisions
  • ADR-001: Registry-based extension architecture
  • ADR-002: Separate AI and vector store connections
  • ADR-003: Pipeline instead of fixed chat workflow
  • ADR-004: Prompts in records, not processors
  • ADR-005: SourceHash and ContentHash separation
  • ADR-006: Collection and storage are indexing scope
  • ADR-007: Metadata-driven retrieval