How the-coca-cola-company/hermes-agent works
Hermes Agent is a self-improving, multi-provider AI agent exposed through a CLI/TUI and messaging gateway, with a desktop client also present in the repository. A user submits a turn through a client or messaging platform; the agent prepares context, calls a model, and may execute tools before returning a response. Major branches include persistent learning, scheduled automations, parallel subagents, and multiple terminal environments. Source excerpts are limited, so README-documented capabilities are shown without claiming unsampled implementation details.
Main entry points: conversation_loop.py Conversation Loop · tool_executor.py Tool Execution · signal.py Messaging Gateway · turnController.ts CLI and TUI
Components 19
User Interfaces
- CLI and TUI
ui-tui/src/app/turnController.ts - Desktop Client
apps/desktop/src/app/chat/index.tsx - Messaging Gateway
gateway/platforms/signal.py
Agent Runtime
- Conversation Loop
agent/conversation_loop.py - Context Management
agent/context_compressor.py - Model Routing
agent/auxiliary_client.py - Model Metadata
agent/model_metadata.py - Tool Execution
agent/tool_executor.py - Terminal Backends
tools/environments - Data Redaction
agent/redact.py
Learning and Memory
- Agent Memory
agent/memory_manager.py - Learning Skills
Platforms and Integrations
- Messaging Platforms
gateway/platforms - Web Search
agent/web_search_provider.py - Media Providers
agent/image_gen_provider.py
Automation and Research
- Scheduled Automations
plugins/cron_providers - Parallel Subagents
agent/subagent_lifecycle.py - Trajectory Generation
agent/trajectory.py
Other components
- User
Connections 20
- User → CLI and TUI: starts turn
README.md - User → Messaging Gateway: messages
README.md - User → Desktop Client: uses
README.md - CLI and TUI → Conversation Loop: submits turn
- Desktop Client → Conversation Loop: submits turn
- Messaging Gateway → Conversation Loop: routes messages
- Messaging Gateway → Messaging Platforms: connects platforms
README.md - Conversation Loop → Context Management: prepares context
agent/context_compressor.py - Conversation Loop → Model Routing: requests completion
- Model Routing → Model Metadata: resolves model
agent/auxiliary_client.py - Conversation Loop → Tool Execution: dispatches tools
README.md - Tool Execution → Terminal Backends: runs commands
README.md - Conversation Loop → Agent Memory: uses memory
README.md - Agent Memory → Learning Skills: supports learning
README.md - Conversation Loop → Data Redaction: redacts data
- Tool Execution → Web Search: offers search
- Tool Execution → Media Providers: offers media tools
- Scheduled Automations → Messaging Platforms: delivers automations
README.md - Conversation Loop → Parallel Subagents: spawns workers
README.md - Conversation Loop → Trajectory Generation: records trajectories
README.md