How gaojie618/agent-graph works
Agent-Graph is a visual multi-agent application for designing and running autonomous agents and structured workflows. A user configures agents, graphs, models, prompts, and tools in the frontend, then starts a conversation or scheduled graph run; execution coordinates model responses, tool calls, context, and persisted run state before returning results. The README also documents project and conversation management, exports, MCP integrations, memory, and team administration. Unsampled subsystems are represented by documented capabilities and tree paths; their internal call wiring is not asserted.
Main entry points: routes.py API routes · graph_executor.py Graph execution · agent_run_service.py Agent execution · tool_executor.py Tool dispatch
Components 23
User experience
- Chat workspace
frontend/src/pages/ChatSystem.tsx - Graph editor
frontend/src/components/graph-editor/GraphEditorView.tsx - Resource managers
frontend/src/pages/AgentManager.tsx - Frontend API client
frontend/src/services/api.ts
Agent runtime
- API routes
agent_graph/app/api/routes.py - Agent execution
agent_graph/app/services/agent/agent_run_service.py - Graph execution
agent_graph/app/services/graph/graph_executor.py - Run conversations
agent_graph/app/services/graph/conversation_manager.py - Tool dispatch
agent_graph/app/services/tool_execution/tool_executor.py
Platform capabilities
- Model service
agent_graph/app/services/model/model_service.py - Long-term memory
agent_graph/app/services/memory/memory_service.py - MCP integration
agent_graph/app/services/mcp/mcp_service.py - Built-in tools
agent_graph/app/services/system_tools/registry.py - Prompt center
agent_graph/app/services/prompt/prompt_service.py - Projects
agent_graph/app/services/project/project_service.py - Scheduled graphs
agent_graph/app/services/task/task_scheduler.py - Dataset export
agent_graph/app/services/export/export_service.py
Persistence
- MongoDB
agent_graph/app/infrastructure/database/mongodb/client.py - MinIO storage
agent_graph/app/infrastructure/storage/object_storage/minio_client.py
Platform operations
- Team and accounts
agent_graph/app/services/user/team_service.py
Other components
- User
- LLM / VLM provider
- External MCP servers
Connections 29
- User → Chat workspace: configures and chats
- User → Graph editor: designs workflows
- User → Resource managers: manages resources
- Chat workspace → Frontend API client: sends requests
- Graph editor → Frontend API client: saves graphs
- Resource managers → Frontend API client: manages resources
- Frontend API client → API routes: calls endpoints
- API routes → Agent execution: starts agent runs
- API routes → Graph execution: starts graph runs
- Graph execution → Run conversations: reads and updates state
- Agent execution → Model service: requests model responses
- Graph execution → Model service: runs agent nodes
- Agent execution → Tool dispatch: dispatches tools
- Graph execution → Tool dispatch: dispatches node tools
- Tool dispatch → Built-in tools: dispatches built-ins
- Tool dispatch → MCP integration: dispatches MCP tools
- Model service → LLM / VLM provider: requests completions
- MCP integration → External MCP servers: connects to tools
- Agent execution → Long-term memory: uses long-term memory
- API routes → Prompt center: manages prompts
- API routes → Projects: manages projects
- Scheduled graphs → Graph execution: schedules graph runs
- API routes → Dataset export: requests exports
- Run conversations → MongoDB: persists run state
- Long-term memory → MongoDB: stores memories
- Projects → MongoDB: stores project data
- Projects → MinIO storage: stores shared files
- API routes → Team and accounts: manages team access
- Team and accounts → MongoDB: stores account data