How anative-lab/evoagentx works
EvoAgentX is a framework for constructing and running LLM-based agent workflows, evaluating them, and iteratively optimizing them. A user describes a goal; workflow generation assembles agents and actions, which execute with model and tool support, optionally using memory and human review. Evaluation and optimizers provide the documented improvement loop. The graph also shows RAG as a supporting retrieval layer. Relationships are limited to README statements and the supplied SOURCE INDEX references; unsampled modules are represented at their exact paths, and their internal behavior is not inferred.
Main entry points: agent_manager.py Agent manager · agent.py Agent runtime · mipro_optimizer.py MIPRO optimizer · workflow_manager.py Workflow manager
Components 19
Workflow orchestration
- Workflow generation
evoagentx/workflow/workflow_generator.py - Workflow manager
evoagentx/workflow/workflow_manager.py - Workflow graph
evoagentx/workflow/workflow_graph.py - Execution controller
evoagentx/workflow/controller.py
Agents and tools
- Agent manager
evoagentx/agents/agent_manager.py - Agent runtime
evoagentx/agents/agent.py - Built-in tools
evoagentx/tools/tool.py - Human review
evoagentx/hitl/hitl.py
Evaluation and evolution
- Workflow evaluator
evoagentx/evaluators/evaluator.py - Workflow optimizer
evoagentx/optimizers/optimizer.py - MIPRO optimizer
evoagentx/optimizers/mipro_optimizer.py
Model and memory
- LLM providers
evoagentx/models/base_model.py - Model configuration
evoagentx/models/model_configs.py - Short- and long-term memory
evoagentx/memory/memory_manager.py - Agent and memory storage
evoagentx/storages/base.py
Retrieval
- RAG pipeline
evoagentx/rag/rag.py - Retrieval data schema
evoagentx/rag/schema.py - RAG configuration
evoagentx/rag/rag_config.py
Other components
- User or developer
Connections 25
- User or developer → Workflow generation: describes goal
README.md - Workflow generation → Workflow manager: assembles workflow
- Workflow manager → Agent manager: manages agents
evoagentx/workflow/workflow_manager.py - Workflow manager → Workflow graph: uses graph
evoagentx/workflow/workflow_manager.py - Execution controller → Agent manager: manages agents
evoagentx/workflow/controller.py - Execution controller → Workflow optimizer: invokes optimizer
evoagentx/workflow/controller.py - Execution controller → Workflow manager: runs workflow
evoagentx/workflow/controller.py - Agent manager → Agent runtime: creates agents
evoagentx/agents/agent_manager.py - Agent manager → Built-in tools: attaches tools
evoagentx/agents/agent_manager.py - Agent runtime → LLM providers: uses model
evoagentx/agents/agent.py - Agent runtime → Short- and long-term memory: uses memory
evoagentx/agents/agent.py - Agent runtime → Agent and memory storage: uses storage
evoagentx/agents/agent.py - Agent runtime → Built-in tools: calls tools
- Workflow graph → Agent manager: loads agents
evoagentx/workflow/workflow_graph.py - Workflow graph → Human review: supports review
- MIPRO optimizer → Agent manager: optimizes agents
evoagentx/optimizers/mipro_optimizer.py - MIPRO optimizer → Workflow graph: optimizes graph
evoagentx/optimizers/mipro_optimizer.py - MIPRO optimizer → Workflow evaluator: evaluates candidates
evoagentx/optimizers/mipro_optimizer.py - Workflow optimizer → MIPRO optimizer: specializes optimizer
- Short- and long-term memory → Agent and memory storage: persists memories
- RAG pipeline → Retrieval data schema: builds corpus
evoagentx/rag/rag.py - RAG pipeline → RAG configuration: uses configuration
evoagentx/rag/rag.py - RAG pipeline → Agent and memory storage: uses stores
evoagentx/rag/rag.py - LLM providers → Model configuration: uses configuration
evoagentx/models/base_model.py - User or developer → Human review: reviews workflow
README.md