How zylon-ai/private-gpt works
PrivateGPT is an API layer for building private AI applications on OpenAI-compatible inference servers; it does not host models itself. The main workflow is a client request into the API, then chat orchestration and model/tool execution, with retrieval grounded by ingested files and artifacts. Separate branches provide embeddings, async ingestion, database/tabular tools, and MCP/custom-tool support. The README documents these capabilities; sampled source and the SOURCE INDEX support only some internal links, so undocumented wiring is omitted.
Main entry points: async_chat_engine.py Chat engine · ingest_service.py Ingestion service · ingestion_scheduler.py Ingestion scheduler · semantic_search_service.py Semantic search
Components 25
API and requests
- Messages API
private_gpt/server/chat/chat_router.py - Async chat API
private_gpt/server/chat_async/chat_async_router.py - Ingestion API
private_gpt/server/ingest/ingest_router.py - Embeddings API
private_gpt/server/embeddings/embeddings_router.py
Chat and tools
- Chat engine
private_gpt/components/engines/chat/async_chat_engine.py - Chat configuration
private_gpt/components/chat/models/chat_config_models.py - Database tool processor
private_gpt/components/tools/processors/database_query_processor.py - Database tool builder
private_gpt/components/tools/builders/database_query_builder.py - Database query engine
private_gpt/components/tabular/database_query_generator.py - Web search
private_gpt/components/web/web_search/web_search_service.py - Code execution
private_gpt/components/code_execution/code_execution_component.py - MCP service
private_gpt/server/mcp/mcp_service.py
Knowledge and ingestion
- Ingestion scheduler
private_gpt/components/ingestion/ingestion_scheduler.py - Ingestion service
private_gpt/server/ingest/ingest_service.py - Document parsing
private_gpt/components/ingest/parse_component.py - Ingest processing
private_gpt/components/ingest/ingest_component.py - Artifact index
private_gpt/artifact_index/base_artifact_index.py - Semantic search
private_gpt/server/primitives/semantic_search_service.py
Models and services
- Embedding service
private_gpt/components/embedding/embedding_component.py - LLM component
private_gpt/components/llm/llm_component.py - Node store
private_gpt/components/node_store/node_store_component.py - Background worker
private_gpt/celery/celery.py
Other components
- Application client
- OpenAI-compatible server
- External tools / MCP servers
Connections 27
- Application client → Messages API: sends messages
- Application client → Async chat API: requests async
- Messages API → Chat engine: runs chat
- Async chat API → Chat engine: runs chat
- Chat engine → LLM component: invokes model
private_gpt/components/engines/chat/async_chat_engine.py - LLM component → OpenAI-compatible server: calls inference
- Chat engine → Database tool processor: resolves tools
- Database tool processor → Database tool builder: builds tool
private_gpt/components/tools/processors/database_query_processor.py - Database tool builder → Database query engine: queries database
private_gpt/components/tools/builders/database_query_builder.py - Chat engine → Web search: offers search
- Chat engine → Code execution: offers execution
- Chat engine → MCP service: supports MCP
- MCP service → External tools / MCP servers: connects servers
- Application client → Ingestion API: uploads content
- Ingestion API → Ingestion scheduler: schedules ingestion
private_gpt/server/ingest/ingest_router.py - Ingestion scheduler → Background worker: dispatches tasks
- Ingestion scheduler → Ingestion service: processes request
- Ingestion service → Document parsing: parses documents
private_gpt/server/ingest/ingest_service.py - Ingestion service → Ingest processing: ingests documents
private_gpt/server/ingest/ingest_service.py - Ingestion service → Artifact index: populates index
private_gpt/server/ingest/ingest_service.py - Artifact index → Node store: reads nodes
private_gpt/artifact_index/base_artifact_index.py - Application client → Semantic search: requests retrieval
- Semantic search → Embedding service: creates embeddings
private_gpt/server/primitives/semantic_search_service.py - Semantic search → Artifact index: searches index
private_gpt/server/primitives/semantic_search_service.py - Application client → Embeddings API: requests embeddings
- Embeddings API → Embedding service: creates embeddings
- Application client → External tools / MCP servers: uses API