# Chat in Squirro Neo Chat is where you ask questions about the content in your project and read the answers. An agent looks for the information it needs, works through the question, and writes an answer that cites the documents it used. ## Starting a Conversation Select the Chat icon in the navigation bar. A wider panel opens beside the bar, holding the following: - **New Chat** Starts a new conversation. - **Tasks** Opens the list of scheduled tasks for the project. A task carries out an instruction on a schedule you set. - **Chats** Your conversation history for the current project, most recent first. Select a conversation to reopen it. Use the button at the edge of this panel to collapse it when you want more room to read an answer, and again to bring it back. Collapsing it leaves the navigation bar in place, and a New Chat icon stays available beside the collapse button so you can start a conversation without expanding the panel first. Selecting the Chat icon in the navigation bar also expands it again. Select New Chat, then type your question in the composer, the area marked Ask anything…, and send it. Everything you ask afterwards stays in the same conversation, so you can follow up without repeating yourself. To start again with no history, open a new chat. ## How an Answer Is Built An answer usually appears as the agent writes it rather than all at once, so you can start reading immediately. Not every model supports that, so whether it happens depends on the model handling your question and on how your project is configured. While the agent works, a short status describes what it is doing, such as Thinking…, Searching…, Using tools…, or Generating response…. Many questions take more than one step to answer. Each step appears in order above the answer. The two most recent stay visible, and once there are more than two, the earlier ones fold away behind a summary reporting how many are hidden. Select that summary to read them, and Hide previous steps to fold them away again. You do not have to read the steps. They are there when you want to check how the agent reached its answer. The following are examples of steps that can appear while an agent completes an answer. Which of them you see depends on what the agent did, on the model handling your question, and on how your project is set up: - **Thought process** The agent reasoning about your question before it answers. Select Thought process to read it. It shows how the agent interpreted your question, which is the quickest way to see why an answer took the shape it did. - **Document Search** A search the agent ran against your project content. It lists the query it used, the key terms it looked for, and the documents it found. Chat shortens long result lists, so select the option to show the remaining results to see the rest, and Show less to collapse them again. - **Tool calls** The agent using something other than document search, such as a knowledge graph or a system your administrator has connected. The step names the tool it called. - **Sub-agents** A specialized agent working on part of your question. Its result feeds back into the main answer. - **Code interpreter** The agent writing and running code, usually to calculate something or work through data. The step reports whether the code ran successfully. - **Artifacts** A file the agent produced for you, such as a document or a spreadsheet. Select the artifact to preview it, or Download to save it. You can stop an answer while the agent is still writing it. The conversation then shows Response interrupted by user, and Retry starts it again. ## When an Agent Asks for Approval Some actions need your permission before an agent may carry them out. When that happens, the answer pauses and shows Action Required, naming the tool that is waiting. Select Approve to let the action proceed or Reject to refuse it. When several actions are waiting, Neo puts them to you one at a time, and the agent continues once you have answered for all of them. Which choices you get depends on the action, so some requests present their own set of options instead of Approve and Reject. An approval request waits for you. It survives reloading the page and returning to the conversation later, so you can answer it when you are ready. If Neo reports that the request has expired, send your message again. ## Citations and Sources Answers cite the documents they draw on. Numbered references appear in the text, and the documents themselves appear as buttons beneath the answer. Three appear at a time, and a further button reports how many remain. Select it to open the full list. Hold the pointer over a numbered reference to see which document it comes from and, where the document has pages, the page the cited passage appears on. Select the reference to open that document. A document opens in a panel beside the answer, whether you reach it from a numbered reference or from the buttons beneath. It opens with the cited passage highlighted, so you can read it in context. The panel header carries the document labels, the file type, where the document came from, and the date it was created, along with Open at source for opening the document where it originally lives. The panel also offers two actions: - **Summarize** Produces a summary of the document you are reading. - **Add to chat** Brings the document into your conversation, so your next question applies to it. Add to new chat does the same in a fresh conversation. Select Open in full screen when you want more room to explore a document in detail. Like any AI-generated content, an answer can misstate what its sources say. Reading the cited passage is the fastest way to confirm an answer against the material it came from. ## Working with an Answer Your question and the answer each carry their own row of icons. Hold the pointer over one to see what it does. On your question: - **Copy question** Copies the question to your clipboard. - **Edit question** Reopens your question so you can rephrase it. The rephrased question produces a new answer, and your original question and its answer are kept, so you can try a different wording and compare the two. On the answer: - **Copy answer** Copies the answer to your clipboard. - **Good answer** and **Bad answer** Rates the answer. Bad answer opens a form asking what was wrong, where you choose a reason such as Not factually correct, Incomplete response, or Wrong Source Cited. You must choose a reason, and choosing Other also requires a comment. Good answer opens a shorter form that asks only for an optional comment. - **Regenerate answer** Runs the same question again and produces a fresh answer as an alternative version you can switch between. It is not offered on an answer you stopped part way through, or on one that did not come from a question you asked. When you edit a question or regenerate an answer, the earlier version is kept rather than replaced. A counter then appears in the row, showing which version you are reading and how many there are, with an arrow on each side to move between them. The counter appears only when more than one version exists. You can open a table or a code block in full screen when you want more room to explore it. Hold the pointer over one and select View fullscreen. From there you can copy a code block, or export a table with Export as CSV or Export as Excel. ## Controlling How an Answer Is Produced The controls around the composer let you pick the assistant that answers, choose what it may draw on, select a model, attach your own files, and reuse prompts you have saved. Sending a question without changing any of them uses the settings your project already provides. > **Note** > > Which controls appear depends on your project. If your project does not offer a choice of assistants, connectors, or models, Neo hides those controls and answers your questions using the project settings. Some controls appear but do not accept a change. They carry Managed by administrator, This setting is locked, or This capability is not available. Your administrator decides all three, so ask them when you need a different value. ### Choosing an Assistant An assistant is a preset way of answering, prepared for a particular kind of work. Neo does not ship with assistants of its own, so which ones you see, and what each one is good at, depends on how your project has been set up. Read the short description beside each one before choosing. Selecting an assistant from Assistants applies it to your question, and Clear assistant goes back to answering without one. An assistant fixes itself to the conversation once it has answered. From that point the control reports the assistant in use and no longer accepts a change, so start a new chat when you want to use a different one. ### Choosing What an Agent Can Use The Connectors control lists what the agent may draw on while answering, and shows how many of those are in use. It has two parts. Abilities covers what is built into Neo. Turn an entry on to allow it and off to leave it out. What your project offers can differ, so you may see only some of the following: - **Project Data** The documents in your current project. This is the usual source for questions about your own content. - **My Files** Files you have uploaded yourself. See [Attaching Files](#attaching-files) below. - **Code interpreter** Lets the agent write and run code, which is useful for calculations and for working through data. Depending on the question, its work can appear as a code interpreter step, as results written into the answer itself, or as an artifact you can download. - **Knowledge graph** A structured view of connected information. When your project offers more than one graph, select the graph to use. The agent reaching for it appears as a tool call step in the answer. Connectors covers the outside systems your project can reach. A single switch beside that heading turns the whole group on and off, and while it is off each server reads Turn connectors on to use this server. Some connectors need you to sign in before you can use them. Select Connect beside the connector and complete the sign-in in the window that opens. Select Disconnect to remove the stored sign-in from Neo. That removes it from Neo only, and the permission you granted at the other system stays in place until you withdraw it there. Turning off an ability you do not need keeps the agent focused. ### Choosing a Model The Model control selects which model handles your question. The same model reasons about what to look for, reads what comes back, and writes the answer, so the choice shapes the substance of a reply and not only its wording. Models differ in speed, cost, and how well they handle long or complicated questions, and your project decides which ones it offers. Two further settings sit alongside it: - **Reasoning** How much effort the model spends working through a question before answering. Higher effort lets the model take more steps of internal reasoning, which helps with complex, multi-step questions, but it uses more tokens and can make the answer slower. The setting appears only for models that support reasoning, and the levels come from the model itself, so they differ between models and providers. The usual set is Off, Low, Medium, and High. - **Output Style** Adjusts the tone and shape of the answer. Your project configures which styles you can choose from, so the list differs from one project to another. ### Attaching Files You can ask questions about your own documents as well as the content already in your project. Select Add content at the left of the composer. Upload files takes a file from your computer, and Add from my files lists the files you uploaded most recently. Attached files appear above the composer, and you can remove one before you send the question. Removing one takes it out of that question only, because anything you upload joins your file library whether you go on to send the question or not. Select View all my files in the same menu to open your full file library. The library keeps what you have uploaded, so a document you use often does not need uploading again, and you can delete files you no longer need. Deleting a file removes it from any question you have not yet sent. For an agent to read your uploaded files while answering, turn on My Files in Connectors. ### Using Prompt Templates A template is a question you have written once and saved to reuse. Templates are useful for anything you ask in the same shape repeatedly, such as a weekly summary or a standard review. A template can be personal or shared with the project, so the list can include templates a colleague wrote as well as your own. Select Templates at the right of the composer and choose one. Its text lands in the composer, where you can adjust it before sending. The menu lists ten templates at a time, so use its search box to reach the rest. A template can carry variables for you to fill in, written in square brackets, such as `[topic]` or `[code]`. When you choose a template that has them, Neo asks for a value for each one and shows you the result before it goes into the composer. Select Apply Template to accept the values, or Skip & Use As-Is to take the template with the brackets left as they are. Select Manage Templates in the same menu to create, edit, and remove your templates. ## Managing Conversations Select a conversation in the Chats list to reopen it and continue where you left off. Each conversation has a menu with the following options: - **Rename** Gives the conversation a name that is easier to find later. - **Export as JSON**, **Export as DOCX**, and **Export as PDF** Saves the conversation as a file in the format you choose. - **Delete** Removes the conversation. This cannot be undone. The list itself also offers Delete All Conversations, which removes every conversation in the current project at once. This cannot be undone either.