📥 /queue — Queue Without Interrupting
O /queue (Q key) queues a prompt for the next turn, without interrupting what the agent is doing now. You move on to the next step while it’s still working on the current one — a continuous flow with no waiting around.
/queue usage (illustrative)
› hermes: gerando relatório… (em andamento) você: /queue "depois, manda por email pro time" › hermes: ok, enfileirado pro próximo turno.
💡 When to use
When you already know the next step and don't want to wait for the current one to finish before typing it. Keeps the session moving.
🔀 /background — Parallel in the Same Chat
O /background runs tasks in in parallel in the same chat: multiple concurrent workflows at the same time. Unlike /queue (which is sequential, "later"), /background is simultaneous ("at the same time").
📥 /queue
Sequential. "Do this after that." One at a time, in order.
🔀 /background
Parallel. "Do this AND that at the same time." Multiple concurrent workflows.
📊 Parallelism example
"/background researches the best AI companies to work for AND summarizes my emails AND drafts something” — three tasks running together, without opening three windows.
🗂️ /canban and /reset
Two organization commands: /canban opens a task board (see what’s in progress, pending, and done); /reset clear everything and start from scratch. Starting fresh is one of the most underrated performance levers.
🗂️ /canban
Task board: organizes what the agent is doing in columns, kanban-style.
🔄 /reset
Clears everything: resets the session context so you can start fresh, light, and fast.
💡 Practical tip
Finished a topic? /reset before the next one. Carrying old context from a finished task only hurts the next task’s performance and cost.
🗜️ /compress — compresses the context
O /compress compresses the conversation’s memory/context: in practice, “summarize everything we’ve discussed” — turns a long conversation into a concise summary and continues from there. It’s like /reset, but without losing the essentials.
📊 /compress × /reset
- /reset → erase everything (start over completely)
- /compress → summarizes and keeps the essentials (light restart)
🧠 /model — switches the brain
O /model switches the model doing the thinking—the “brain” of Hermes. It’s the foundation of the multi-brain strategy (Track 1): use the best model for each task instead of being tied to just one.
Switching models (illustrative)
you: /model opus-4-8 # difficult reasoning you: /model gpt # volume, using your subscription you: /model deepseek # almost free
💡 Practical tip
Switch models as the task changes within the same session: a powerful one for planning, a cheaper one for handling volume. “To a hammer, everything is a nail.”
🪟 The context window — one session, one window
The concept that ties it all together: every question uses ALL the context + the question. The larger the context, the worse the performance and the higher the cost. That's why ideally "one session, one window": focus on one topic, then reset or compress.
Open one session per topic
Start clean, focused on one goal. Smaller context = better, cheaper answers.
Compress as it grows
Has the session gotten long but is still useful? /compress summarizes and preserves the essentials.
Reset when changing topics
New and unrelated topic? /reset and start another clean window.
✗ Bloated context
- ✗One endless session with 10 topics
- ✗Performance drops with every turn
- ✗Cost increases (the entire context is resent)
✓ Lean window
- ✓One topic per session
- ✓/compress when it gets large
- ✓/reset when changing topics
📊 Why this wraps up Track 2
The 6 keys are there to manage the context window: queue e background organize the workflow, kanban visualizes, reset e compress control the size, and model chooses the brain. Budget and tokens are explored in depth in Track 3.
📌 Module Summary
Next Track:
Track 3 — Power & Operations: security, goals, sub-agents, heartbeat, budget, and the operating system