How zyren41/agi_saber works
AGI-assistant is a multimodal conversational agent exposed through a FastAPI service. The principal flow accepts a chat request, routes it to conversation, RAG, a single tool, or multi-step ReAct, then returns an answer and execution details; memory and snapshots support continuity and recovery. Separate document ingestion and retrieval branches support knowledge-base use. The graph preserves README-documented capabilities and source-indexed core modules; some connections are documented rather than verified in sampled implementation excerpts.
Main entry points: agent.py Unified agent · ragchunk.py Repositories · handler.py HTTP API · main.py Service assembly
Components 25
Service interface
- HTTP API
final/internal/handler/handler.py - Service assembly
final/main.py
Agent orchestration
- Unified agent
final/internal/agent/agent.py - Mode router
final/internal/agent/router.py - ReAct planner
final/internal/agent/planner.py - Prompt context
final/internal/promptctx/assembler.py
Knowledge and memory
- RAG engine
final/internal/rag/rag.py - Hybrid retrieval
final/internal/rag/hybrid.py - Document library
final/internal/document/library.py - Document parser
final/internal/document/parser.py - Knowledge graph
final/internal/graph/kgstore.py - Three-tier memory
final/internal/memory/memory.py - Graph memory
final/internal/memory/graph_memory.py
Tools and recovery
- Tool registry
final/internal/tools/tools.py - Sandbox executor
final/internal/sandbox/executor.py - Task recovery
final/internal/agent/restore.py
Persistence and services
- LLM client
final/internal/llm/llm.py - Infrastructure
final/internal/infra/infra.py - Repositories
final/internal/repo/ragchunk.py - PostgreSQL
final/internal/platform/postgres.py - Milvus
final/internal/platform/milvus.py - Elasticsearch
final/internal/platform/es.py - Neo4j
final/internal/platform/neo4j.py - Kafka
final/internal/platform/kafka.py
Other components
- User
Connections 29
- User → HTTP API: sends request
- Service assembly → Infrastructure: initializes
final/main.py - Service assembly → Unified agent: constructs
final/main.py - Service assembly → HTTP API: sets up routes
final/main.py - HTTP API → Unified agent: processes chat
final/internal/handler/handler.py - Unified agent → Mode router: selects mode
final/internal/agent/agent.py - Mode router → ReAct planner: chooses ReAct
- Unified agent → Prompt context: assembles context
final/internal/agent/agent.py - Unified agent → Three-tier memory: loads and stores
final/internal/agent/agent.py - Unified agent → LLM client: generates responses
final/internal/agent/agent.py - Unified agent → RAG engine: runs knowledge search
final/internal/agent/agent.py - Unified agent → Tool registry: dispatches tools
final/internal/agent/agent.py - ReAct planner → Tool registry: plans tool calls
- Tool registry → Sandbox executor: executes commands
- HTTP API → Document library: manages documents
final/internal/handler/handler.py - Document library → Document parser: parses content
- RAG engine → Hybrid retrieval: retrieves and fuses
final/internal/rag/rag.py - RAG engine → LLM client: synthesizes answer
final/internal/rag/rag.py - Hybrid retrieval → Knowledge graph: searches graph
final/internal/rag/hybrid.py - Three-tier memory → Graph memory: extends recall
final/internal/memory/memory.py - Knowledge graph → Neo4j: queries graph store
final/internal/graph/kgstore.py - Hybrid retrieval → Repositories: loads and indexes chunks
- Repositories → PostgreSQL: persists source chunks
final/internal/repo/ragchunk.py - Repositories → Milvus: indexes vectors
final/internal/repo/ragchunk.py - Repositories → Elasticsearch: indexes keywords
final/internal/repo/ragchunk.py - Infrastructure → Repositories: assembles repositories
final/internal/infra/infra.py - Unified agent → Task recovery: recovers execution
final/internal/agent/agent.py - Task recovery → PostgreSQL: stores snapshots
- Infrastructure → Kafka: publishes events
final/internal/infra/infra.py