> ## Documentation Index
> Fetch the complete documentation index at: https://docs.polymet.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Threads, context and queued messages

> Keep separate conversations in one project, watch how much the agent remembers, and send follow-ups while it works.

A thread is one conversation with the agent. A project can have many, so you can keep the checkout work apart from the landing page work. Every thread works on the same project files.

![The thread tabs above the chat, with the New thread button](https://mintcdn.com/polymet/TfIzLh80_L4bW9xQ/images/docs/canvas/thread-tabs.png?fit=max\&auto=format\&n=TfIzLh80_L4bW9xQ\&q=85\&s=e399c39ee03623e5081bcd45c23e5951)

## Threads

The tabs above the chat are your threads.

* **Start one:** click the plus (**New thread**), or type `/clear` in the generation input. The thread you left stays as it was.
* **Switch:** click a tab. You can switch while the agent works; the generation keeps running in its own thread.
* **Rename:** open the tab's menu and click **Rename**.
* **Delete:** open the tab's menu and click **Delete**. Polymet asks **Delete thread?** This deletes the thread and its messages for good.

You can't delete a project's only thread, or a thread with a generation running.

A new thread starts with your default model. See [Models and thinking effort](https://docs.polymet.ai/canvas/models.md).

One generation runs on a branch at a time. While one runs, a thread on the same branch can't send until it finishes. The send button explains why and suggests waiting or switching branch. A tab shows a small branch icon while its thread is generating. More in [Branches](https://docs.polymet.ai/team/branches.md).

## How much the agent remembers

The ring next to the model name is the context meter. It fills as the conversation grows, and changes color as it nears the point where the agent summarizes. Hover it to see the numbers.

Click it to open **Context**:

![The Context panel showing how much of the model's context the thread uses](https://mintcdn.com/polymet/TfIzLh80_L4bW9xQ/images/docs/canvas/context-meter.png?fit=max\&auto=format\&n=TfIzLh80_L4bW9xQ\&q=85\&s=c5f44461dfb17af96610c19d6cbf1038)

* The percentage and the numbers show how much of the conversation the agent is holding, against the point where it summarizes.
* The bar shows the same thing across the model's full limit. Part of it is always taken by Polymet's own instructions (**system + tools**); the rest is your **chat**.
* The marker on the bar is the auto-compact point. **Past the marker, the agent is asked to summarize first.**

## Auto-compact

When a thread reaches the marker, the agent summarizes the conversation and carries on from the summary. The chat shows **Compacting...**, then **Compacted**. You don't need to do anything.

To change when this happens, drag the marker on the **Context** bar. A later point lets the agent hold more of the conversation; an earlier one keeps generations smaller. The marker can't go below 100,000 tokens. Click **Reset** to go back to the model's default; **default** next to the numbers means you haven't changed it.

The setting is yours, saved for each model.

To summarize right away, type `/compact`. Compacting uses credits like any other generation.

## Slash commands

Type `/` at the start of the generation input to open the command list. Use the arrow keys and press Enter or Tab to pick one, or Esc to close it.

| Command | What it does |
| - | - |
| `/compact` | Summarize the conversation to free up context |
| `/model` | Switch the model for new prompts. Type part of a name to filter |
| `/clear (new thread)` | Start a new thread. The previous thread stays accessible |
| `/support` | Open the support chat |

Slash commands are on the canvas only, and not while a generation is running.

## Queue a message while the agent works

You don't have to wait for a generation to finish. Keep typing, then pick when your message goes:

| Choice | Shortcut | What happens |
| - | - | - |
| **Send immediately** | Cmd+Shift+Enter | Steers the agent right away, at its next step |
| **After current generation** | Cmd+Enter | Sends once the generation finishes |

You can also click the queue button, just left of the stop button, and pick from its menu. On Windows, use Ctrl instead of Cmd.

Queued messages wait above the generation input, with how many are queued and **Clear** to drop them all. For each one you can:

* click its slot label to move it to the other slot
* edit it with the pencil
* send it now with its send button, once nothing is running
* remove it with the cross

Messages with a PDF or text file attached can only go after the current generation. Images can go either way.

Your queue shows in every tab where the project is open, and survives a page reload.

If you stop a generation while messages are queued, Polymet asks whether to **Keep queue** or **Discard queue**. Kept messages don't start a new generation on their own. They send after your next generation, or when you click their send button. They are lost if you reload the page before then.

Queueing needs credits, like sending. If your organization is out, the message isn't queued.

## Related

* [Prompt the agent](https://docs.polymet.ai/canvas/prompt.md) — everything else in the generation input
* [Models and thinking effort](https://docs.polymet.ai/canvas/models.md) — the model a new thread starts with
* [Branches](https://docs.polymet.ai/team/branches.md) — parallel work, and why a thread can wait for another
* [How credits work](https://docs.polymet.ai/plans/credits.md) — what a generation and a compaction cost


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.