User rant: Hermes Agent still too unstable for real work at v0.18
A user tried building a small web app with that pulls data from two APIs, joins it, and displays two tables, and ran into a string of problems. Tool calls were sometimes not interpreted properly and showed up as plain text instead of being executed. Sessions would hang or desync partway through, while the backend LLM usage still stayed maxed out.
Tool calls that did get interpreted, such as camoufox and browser_console, often failed to execute. The desktop app had several annoying bugs of its own. Nearly every update broke something that previously worked — compression broke last week, and the desktop app broke this week — with the suggestion that stable releases are not a priority.
In follow-up edits, the user acknowledged no one had claimed Hermes was (it's only v0.18) and noted that things went smoothly when using Nous's Hy3-free model, suggesting the issues lie more with to other s. For comparison, the user also tried OpenRouter's Nemotron and Ollama's Qwen3.5, Qwen3-coder, , and .
Key points
- On v0.18, tool calls sometimes render as plain text instead of being executed
- Sessions can hang or desync mid-task while backend LLM usage stays pegged at max
- Some interpreted tool calls, like camoufox and browser_console, still fail to run
- Nearly every update introduces a new breakage (compression one week, the desktop app the next)
- Using Nous's Hy3-free model was comparatively stable; issues seem concentrated in other provider