See what your agent is actually doing
“I want visibility into every tool call, session and decision instead of guessing from the output”
There are three distinct questions here and picking the wrong tool for yours wastes a weekend: what is it doing right now, what did it do last week, and why did this specific call go wrong. The trace viewer answers the third and is the one people reach for during an incident. Read-only journeys and session history answer the first two, and the read-only property matters more than it sounds — observability that mutates the agent's context is a debugging trap. OTel export is the escape hatch if you already have infrastructure; auto-recovery is only worth it once you have enough agents that watching them manually stopped working.
Real posts from the Hermes community, mirrored from Nous Research's user stories corpus.
I created a small Hermes plugin that records every tool call and every LLM API call into a local SQLite database (one DB per profile), with five ready-to-go Grafana dashboards bundled.
Hermes hud is a TUI dashboard that watches your ai agent think. it reads from your agent's memory, tracks skills, sessions, corrections, projects, cron jobs, all of it. live.
Ran an external RCA script against my full local session history (129 sessions across 23 days) to audit Hermes compliance with its approval gate. 112 of 129 sessions contain at least one violation.
Been building H-OPS, an operator dashboard for Hermes Kanban. The goal is to make multi-agent work observable.
Day 10 of Building AGI for my Hermes Agent: Codex saved the day as a runtime monitor for my agent-to-agent workflows. I used Codex with GPT-5.4 on extra-high to watch the workflow run, catch where it broke, and fix it live until it worked reliably.
- Full tool-call logging captures whatever the agent saw, secrets included. Treat the trace store as sensitive.
- Several of these overlap heavily. Start with the one matching your actual question rather than installing the set.