Commands

Open Assistant supports a small set of slash commands you can type directly in any conversation.

/clear

Start a fresh conversation thread.

/clear

Response: Conversation cleared.

Why use it

Open Assistant keeps the full message history of the current conversation in context when forming its responses. This is great for continuity within a single topic, but it becomes a liability once the conversation drifts:

• Mixed topics confuse the model. If you ask about your calendar, then pivot to drafting an email, then ask something unrelated, the model has to reason over a noisy, multi-topic thread. Responses become less focused and more likely to carry over irrelevant context.
• Long threads cost more tokens. Every message in the active thread is sent to the LLM. A sprawling thread inflates cost and latency with no benefit once the earlier topic is closed.
• Cross-conversation context is already handled. You do not lose history by clearing. memory_recall (the assistant's persistent memory store) and conversation search give it access to past conversations when genuinely relevant. The old thread stays intact and searchable — /clear only starts a new one for the next topic.

When to use it

Use /clear when:

• You are switching to a clearly different topic (e.g. you were planning a trip, now you want to manage emails).
• The conversation has grown long and responses feel less sharp or are mixing up context from earlier messages.
• You notice the assistant referencing something from earlier in the thread that is no longer relevant.

A good rule of thumb: one thread, one topic. When the topic changes, /clear.

What happens under the hood

/clear creates a brand-new conversation with a fresh ID and routes all subsequent messages to it. The previous conversation is not deleted — it remains fully available in your conversation history and can be searched or referenced by the assistant's memory tools.

---

/cancel

Stop all ongoing work in the current conversation and return to a clean state — without leaving the thread.

/cancel

Response: ⛔ All ongoing work has been stopped. You can start a new request whenever you're ready.
(If background tasks were running, the response also lists how many were cancelled and their IDs.)

Why use it

Open Assistant can run long multi-step plans and dispatch parallel background tasks on your behalf. Most of the time this is exactly what you want — but occasionally the model gets stuck in a loop, pursues the wrong approach, or you simply change your mind mid-flight.

• Loop detected too late. The built-in stuck-detection kicks in automatically, but it takes a few repeated iterations to trigger. /cancel lets you cut it short the moment you notice something is off.
• Wrong plan, wrong direction. If the model misunderstood the request and is off on a long tangent, waiting for it to finish wastes time and tokens.
• Background tasks you no longer need. dispatch_task spawns sub-tasks that run concurrently. /cancel stops all of them for the current conversation immediately.
• Suspended ask_user prompts. If the model paused mid-plan to ask you a question but you want to start fresh instead of answering, /cancel clears that suspended state.

When to use it

Use /cancel when:

• The assistant appears stuck — repeating the same tool calls or making no visible progress.
• You realise mid-response that you phrased the request wrong and want to try again.
• A multi-step plan is running but you want to change course before it finishes.
• You see background tasks spinning (the tool-call indicator keeps appearing) and want them stopped.

Difference from /clear

	/cancel	/clear
Stops ongoing work	✅ Yes	✅ Only because a new conversation is started
Stays in the same thread	✅ Yes	❌ No — starts a new conversation
Preserves conversation history	✅ Yes	✅ Yes (old thread is kept, a new one begins)
Use when…	You want to interrupt and retry in the same thread	You are switching to a different topic entirely

What happens under the hood

When you type /cancel:

1. Every running background sub-task for the active conversation has its asyncio future cancelled and is marked "cancelled" in the task store.
2. Any suspended ask_user execution state is cleared, so the next message you send is treated as a fresh request rather than an answer to a previous question.
3. Both the /cancel command and the assistant's acknowledgement are stored in the conversation history, keeping the thread coherent.
4. The response is returned immediately — no LLM call is made.