Hermes Agent connects model responses to conversations and actions. In an OtherU deployment, Ember supplies inference while Hermes manages the configured tools, messaging channels, scheduled work, and memory integrations.
Hermes is a separate project with its own releases and documentation. Use the upstream repository for installation and version-specific configuration. To connect it to Ember, follow Connect Hermes Agent.
From message to action
Hermes receives a message, assembles the permitted conversation context, and sends a request to the configured model. When the model requests a tool, Hermes executes the allowed operation and provides its result to the next model turn.
The available tools depend on the installation. They may include shell and file access, web retrieval, messaging, delegation, or computer control. Configuring one inference model does not imply multiple independent reasoning services or an external supervisor.
Memory
| Context source | Purpose |
|---|---|
| Conversation history | Earlier messages and results from a session |
| Persistent memory provider | Facts retained for retrieval across sessions |
| Skills | Reusable instructions for recurring work |
OtherU uses Hindsight as a persistent memory integration. Its retrieval and retention behavior depends on the selected provider settings; see Hindsight Memory.
Permissions and privacy
Hermes runs tools with the access granted to its process and connected accounts. Tool descriptions and model instructions do not create operating-system isolation. Shell access, private files, messaging credentials, and shared memory need explicit permissions outside the model.
Messaging connectors also introduce external services into the data path. A local inference endpoint does not make those channels private to the host. See Security Model before enabling tools or sharing an agent across users.