External memory works in Hermes, but the agent may skip it
Hermes Agent’s can handle small facts, but it may not be enough for searching a larger . A gbrain MCP server was connected to Hermes Agent to add over a full . The setup runs on Postgres 16 with pgvector and searches 155 pages split into more than 3,000 content chunks.
When Hermes Agent actually runs the search, it finds the right background, such as context from a project mentioned three weeks earlier, without needing the details explained again. Hermes registered 92 MCP tools, but those tools were not being added to live chat sessions because of a known issue. A wrapper script worked as a workaround because it could call gbrain directly when needed.
The main weakness is reliability: even with a firm instruction to search memory before answering when a person, project, or recurring topic appears, Hermes Agent followed that rule only about 60% of the time and skipped memory search the rest of the time.
Key points
- The is fine for small facts, but less suited to a larger .
- gbrain added across 155 Obsidian pages and more than 3,000 content chunks.
- Search results were useful when Hermes Agent actually ran the search.
- Hermes registered 92 MCP tools, but they were not injected into s.
- Strong instructions did not fully solve the problem; memory search happened only about 60% of the time.