I advise against using Hermes Agent (bednars.me)

A recent user review of the Hermes Agent on the Ollama platform highlights significant shortcomings in the software, impacting its usability and security. The reviewer shared their experience of poor interface responsiveness, state management bugs, and unexpected behavior when configuring the agent, notably with default settings that seemed to optimize token consumption rather than performance. For instance, the Hermes Agent defaulted to Claude Opus 4.8, despite benchmarks indicating that the Gemini 3.1 Flash Lite model outperformed it significantly in agentic tasks. Additionally, critical configuration issues raised concerns about data privacy; the agent inadvertently disclosed confidential information when added to a new Telegram group. The review also noted a lack of clear security settings, like the ambiguous “GATEWAY_ALLOW_ALL_USERS” variable, which, without a proper default, raises risks for users. With nearly 9,000 unresolved issues reported on GitHub, the reviewer strongly advised against using the Hermes Agent, labeling it as unresponsive and unreliable, which poses a risk for users looking for secure AI tools. This case serves as a cautionary tale for the AI/ML community regarding the importance of robust software architecture and user privacy safeguards.