# Ask Papiers

Chat sits beside what you’re reading, so you can bring the source directly into the conversation.

Select a passage and it appears in the composer automatically. Hold `Shift` while selecting to add more passages, including from documents you have open side by side. When you send the message, Papiers includes those passages with your question and anchors the conversation back to where they came from.

For figures, tables, diagrams, equations, or anything else you want to point at in a PDF, press `⌘⇧S` or use **Screenshot Capture** and drag over the region. Papiers keeps track of where the capture came from, along with the surrounding document structure and caption.

For equations, Papiers also recovers the underlying LaTeX. The model receives both the captured region and the equation itself, rather than having to infer the math from the PDF image alone.

You can also paste images directly into the composer. If the selected model supports vision, the image itself is sent to the model. For PDF region captures, Papiers sends the region’s extracted text instead when you’re using a model without vision. Pasted images require a vision-capable model.

## Search your library

Explicit context is useful when you know what you want the model to look at. **Search your library** lets the agent find context on its own.

With it on, Papiers can search your documents, highlights, notes, equations, and previous conversations while answering. Turn it off when you want a response without anything from your library. Anything you attach directly to the message is still included.

You can also type `@` to add a particular document or note to the conversation without searching for passages yourself.

## Think out loud

Press `Enter` when you want Papiers to respond.

Press `⌘Enter` on Mac or `Ctrl+Enter` on Windows and Ubuntu to **send quietly**. The message is added to the conversation without starting the agent.

Send as many quiet messages as you want. When you’re ready for a response, send the next message normally. Papiers sees the whole sequence together.

If the agent is already working, keep typing. New messages are queued while the current turn finishes, then added to the conversation together before the agent continues.

## Branch a conversation

Edit an earlier message to take the conversation in a different direction. Papiers creates a branch and keeps the original path intact.

Switch between branches at any time. The current branch is the conversation history the model sees by default.

With **Search your library** on, the agent can still find relevant messages from other branches when it searches your previous conversations.

Only one branch can run at a time. If the agent is working in one branch, messages you send in another are queued until the active run finishes.

## Research with tools

Chat in Papiers is backed by an agent, not a single model call.

When a question needs more work, the agent can search the web, search Semantic Scholar, look up papers and their metadata, follow citations, and read across your library.

With **Search your library** on, it can also search and read your documents, passages, highlights, notes, equations, and previous conversations. These are the same underlying library primitives exposed through the [Papiers CLI](https://wiki.papiers.ai/docs/agents).

The agent can combine those tools before answering: find related papers, read the relevant sections, compare them with something you highlighted months ago, or pull together the literature around a question.

## Choose a model

Click the model name in the composer to switch models.

Papiers includes **Sonnet** and **Opus** from Anthropic; **GPT-5.6 Luna**, **Terra**, and **Sol** from OpenAI; and open models including **Kimi K3**, **GLM 5.2**, and **DeepSeek V4 Flash**.

The open models run through Fireworks on servers in the United States.

Some models support vision and can work directly with images; others are text-only. Papiers handles PDF region captures for either, but pasted images can only be sent to a vision-capable model.

Reasoning controls appear where the model supports them, and you can switch models at any point in a conversation.

## Artifacts

Papiers can create richer outputs alongside the conversation, including documents, charts, and SVGs.

Artifacts open beside the chat and can be updated over subsequent messages. Ask for a chart from a paper’s results, a visual explanation of an idea, or a longer document, then keep refining it in the same conversation.

## Answers stay connected to their sources

When Papiers uses material from your library or the web, the answer cites it. Click a citation to open the source at the relevant passage.

Questions you ask from a passage stay anchored there too. Return to a document later and you can see what you asked where you asked it.

## Start anywhere

You don’t need to have a document open.

Start from **Home** when you want to ask across your library without a particular source in mind. From anywhere else in Papiers, `⌘⇧P` opens the same search and chat directly.

If you open a source from the answer, Papiers brings the conversation into the Workspace beside it.

Source: https://wiki.papiers.ai/docs/ask