How skygod07/hermes-agent works
Hermes is a self-improving AI agent exposed through CLI/TUI, messaging platforms, and a desktop app. The main loop takes a user prompt, runs model-driven turns and tools, then returns a response; memory and skill creation extend learning across sessions. The README also documents scheduling, subagents, multiple execution backends, and trajectory generation. The graph combines those documented capabilities with relationships visible in the supplied source index; several subsystems were not sampled, so their internal links are omitted.
Main entry points: conversation_loop.py Conversation Loop · base.py Messaging Gateway · run_agent.py Agent Entry · cli.py CLI
Components 18
User Interfaces
- CLI
cli.py - Terminal UI
ui-tui/src/app.tsx - Desktop App
apps/desktop/src/app/desktop-controller.tsx
Agent Runtime
- Agent Entry
run_agent.py - Conversation Loop
agent/conversation_loop.py
Tools and Learning
- Tool Execution
tools/terminal_tool.py - Memory Manager
agent/memory_manager.py - Skill Learning
agent/skill_commands.py
Channels and Providers
- Messaging Gateway
gateway/platforms/base.py - Webhook Adapter
gateway/platforms/webhook.py - Signal Adapter
gateway/platforms/signal.py - Model Providers
agent/model_metadata.py
Execution and Research
- Scheduled Jobs
gateway/config.py - Subagents
agent/conversation_loop.py - Execution Backends
agent/conversation_loop.py - Trajectory Research
agent/trajectory.py - Language Services
agent/lsp/manager.py
Other components
- User
Connections 20
- User → CLI: prompts
README.md - User → Terminal UI: prompts
README.md - User → Desktop App: uses
README.md - User → Messaging Gateway: messages
README.md - CLI → Agent Entry: starts
README.md - Terminal UI → Agent Entry: sends turns
- Desktop App → Agent Entry: connects
- Messaging Gateway → Conversation Loop: routes prompts
- Webhook Adapter → Messaging Gateway: receives events
gateway/platforms/webhook.py - Signal Adapter → Messaging Gateway: adapts messages
gateway/platforms/signal.py - Agent Entry → Conversation Loop: runs
agent/conversation_loop.py - Conversation Loop → Model Providers: requests model
agent/conversation_loop.py - Conversation Loop → Tool Execution: dispatches tools
- Conversation Loop → Memory Manager: uses memory
agent/conversation_loop.py - Memory Manager → Skill Learning: enables learning
agent/memory_manager.py - Conversation Loop → Subagents: spawns work
README.md - Conversation Loop → Execution Backends: uses environment
README.md - Messaging Gateway → Scheduled Jobs: delivers jobs
README.md - Conversation Loop → Trajectory Research: records runs
agent/conversation_loop.py - Tool Execution → Language Services: uses language services