# Poneva Chat > A company workspace for asking questions, reviewing work, and keeping shared context in one place. This page is a review preview of the company workspace. Features become available as the service is released to your account. Poneva Chat is a workspace for company questions and shared work. Ask a question in a chat, connect the services your company uses, and keep the resulting work in the same workspace. ## Start with your workspace Onboarding helps you describe your role and goals, then connect services for the workspace. You can return to the workspace when setup is complete. Connections are managed from the workspace and can be searched before you choose one. New chat offers task cards you can click to start, with short descriptions and app logos where relevant. Suggestions draw on your company context, personal preferences, conversation text, approved company knowledge, published Skills, available connections, and agents you can open. They are personal to you within the company. Follow-ups give the latest day's work the strongest weight, with the rest of the last seven days as context. If you have no work in that window, your latest older conversations still guide the suggestions. The assistant uses bounded excerpts and can ask to verify uncertain status or read the full source. Suggestions are prepared automatically before you open New chat and replaced daily. Previous suggestions help shape the next set; they do not mean a task was started or completed. New knowledge, Skills, agents and available connections can make a previous request useful to revisit. Context changes are checked in the background. Use **Refresh** to check for updated context. While a new daily set is being prepared, the previous cards stay available if their context and access are unchanged. Returning to New chat reuses its prepared cards. Shortcuts help you create an agent, save a reusable method as a Skill, take meeting notes, work with a file, browse agents, or return to saved results. **Create an agent** starts a chat that proposes a useful goal and instructions for you to review. A task card starts a chat request; proposed changes still follow the usual review rules. Until suggestions are ready the shortcuts are there on their own; the suggestions appear by themselves when they are. Typing a draft hides the suggested tasks. ## Paste long text into a chat Pasting more than 2,000 characters adds a **Pasted text.txt** attachment, keeping the message field compact. Smaller pastes stay in the field. You can add several pastes and remove any you do not need. Select a pending attachment to edit it. **Save changes** keeps your edits; **Close** discards them. **Show in text field** moves the exact text back to your caret or replaces your selection. You can undo or redo the latest paste operation with your usual keyboard shortcut while the draft is otherwise unchanged. Sending includes the complete text. Select a sent attachment to read it; sent text is read-only. Pasted text is kept with the chat and is not automatically saved to the company Library. The 4 MB inline attachment limit applies; larger text needs **Add company context → Upload a company file**. ## Read and review work in a canvas Longer work opens in a canvas beside the conversation. Use the canvas to read a draft, result, knowledge page, or other workspace item; move between items with a back path; and return to the chat without losing your place. When an item needs a decision, its review controls live with the item. A substantial change is shown with its details and can be edited and staged again before you approve it. Small settings can show a compact approval control. Select a source in an answer, such as a citation, a knowledge page section or a web page, and it opens in the canvas while the conversation stays where you left it. A web page shows its title and address, with **Open original** to visit it in a new tab. Back returns to the previous canvas and finally to the chat, and a reload brings the canvas back. On a phone the canvas is a sheet. This review model keeps the conversation readable and gives each item one clear place for its decision. A proposed change does not become a company change merely because it appeared in a chat. An Artifact stays private when Chat first saves it. Saving does not publish it or change its audience. If a save is interrupted, wait for confirmation of the original save before treating it as complete. ## Create and edit a private Artifact Where Artifact authoring is available, ask Chat to create the work you need. Chat writes the content and saves it as a private Artifact in the same conversation. Your request authorizes that private save; there is no additional Build or Save approval. Treat the work as saved only when the save is confirmed. If the outcome is unconfirmed, check the original save before trying again. Each Artifact has one Library entry. Later private saves add versions to that same item. Apps belong on their own Apps page, separate from Artifacts in Library. A Report can include named saved Tables, Charts and captured images that you can currently access. Missing sources and save conflicts are reported rather than shown as completed work. To reuse work, ask Chat to copy an Artifact you can open and tailor the copy. The copy is an ordinary Artifact with its own identity; the source stays unchanged. Ask for later changes to that copy to keep them in its own version history. There is no separate template to create or start. Where sharing controls are available, the audience chip shows **Only you**, **Company**, or the number of people with access. Choose Only you or Everyone at your company to apply that choice, or choose Specific people and then Share. Sharing changes access and does not publish the work. Publishing selects the version shown to viewers. Editors with current access can work on a newer private version while viewers keep the published version. Unpublish removes viewer access; editors keep their access. Removing a named recipient does not remove access they still have through the company audience. Where the saved Document or Report edit controls are available, use **Edit** to work on a separate private draft. Inspect lets you select supported text and presentation properties, prepare a change, then Apply or Undo it. These direct edits do not call a model or overwrite the saved version. If the reply is lost, **Check again** confirms the original change before showing it as saved. **Mention** opens the chat composer with the selected part for review and does not send a message. Where DOCX export is available, download the selected saved Document with its provenance. The download and its accompanying information file use the same temporary access lease; an expired link requires current access again. Exporting does not publish the Document. Other formats depend on the Artifact kind and available export controls. The current Artifact review has confirmed private saves. Rendered open/reload, the full set of kinds, and sharing/copy/edit/export acceptance for this Chat save flow are still being verified; earlier Document walkthroughs do not establish every kind. In the voice review build, spoken answers can also open a temporary canvas preview. That preview is not saved work and is not restored through the canvas navigation history. Showing it does not approve a proposed change. ## Add your meeting notes Open a meeting's **Canvas** to write or paste your own notes while it is recording, being transcribed, or ready to review. **Transcript** shows the recorded words separately. Your saved notes show their author and stay separate from the generated summary. Use **Upload notes** to bring a text or Markdown file into the editor. Review the text, then select **Save note** or press Cmd/Ctrl+Enter. Each note can contain up to 12,000 characters. Switching views or waiting for transcription keeps your draft in the open canvas; save it before reloading the page. Notes are visible to the people who can read the meeting. You can add notes to a meeting you can read, and remove your own notes. **Add through chat** prepares a message with the meeting attached for you to send. A proposed note opens for review with its text and meeting destination before you approve it. ## Search and read sources Tools from Poneva's catalog, including web search, page reading and merchant lookups, run without an approval card in typed chat and voice mode. Chat still checks that each tool is currently authorized before running it. This does not approve changes to your settings, saved work or connected services. ## Company knowledge Knowledge is your company's own description of itself and how the business works, kept as pages in a tree. The top page describes the company; Poneva reads it in every chat, together with the titles of the other pages, and reads the rest when a question needs it. Answers cite the page and section they rely on. Publishers edit pages directly, and each save is kept in the page's history. In a personal Chat, a current publisher can ask Poneva to create, edit or move a page without another approval tap. Poneva reads the page first and checks the current revision and publishing rights before saving. Treat the change as saved only when the writer confirms it. A stale revision or lost access is a refusal; an unconfirmed outcome needs a read of the same page before trying again. Everyone else suggests changes: edit a page and send it with a short reason, or tell Poneva in chat that something has changed. Unsolicited suggestions and requests to archive a page stay on the review path. A pending suggestion does not change the approved text; a publisher accepts or rejects it on the page. A suggestion appears on the page itself, with added words highlighted and removed words struck through, and a note of who suggested it, where it came from and why. When two suggestions change the same text, the second shows its change on its own note instead. A publisher accepts or rejects it on the page; nothing changes until they do, and Poneva keeps answering from the page as it is. A suggestion whose text has changed since is marked as outdated. The For review tab shows publishers every suggestion waiting for a decision; everyone else sees their own suggestions there. Type `[[` while editing to link another page; links keep working when a page is renamed or moved, and each page shows the pages that link to it. If you leave a page while editing it, Poneva asks before discarding your edit. Pages are visible to everyone in the company. Archiving a page takes it out of answers and keeps its history; a publisher can restore it, or delete it permanently, which removes the page, its history and its open suggestions for good. Files, chats and meetings are not part of knowledge: Poneva finds them with search when a question needs them. Keep your design or brand guidelines as a knowledge page. Before Poneva makes a deck or a formatted document, it looks for them in your pages, follows what the format can carry, such as wording, tone and structure, and says which page it used. Decks keep Poneva's own colours and fonts, and Poneva says so when your guidelines ask for others. ## Find company work The workspace includes conversation search, a company feed, Skills and knowledge pages, and an Inbox for items that need your review or updates from colleagues. Inbox entries open in the shared canvas, where you can read the full item and take the decision available to you. Shared work keeps its audience and permissions when you move between the chat, feed, and canvas. Search finds the chats, files, saved results, Skills, meetings and knowledge you can open, including your own unpublished Skill drafts, chats from agent runs you can see, and meetings that are still being processed (by their title). When you ask Poneva what your company knows, has done, said or decided, it searches the same way: everything you could open yourself, and nothing you could not. A company agent searches only what the whole company can see. Templates can help you start a private Skill draft. Using a template does not publish it automatically; review and publication remain deliberate steps. ## Build and run an agent Ask in chat to create an agent and describe the result you need. Chat can ask follow-up questions and open a draft beside the conversation. The canvas separates the agent's goal, instructions, expected result, resources and boundaries. You can edit the draft there, then continue refining it in chat before approving it. An agent owns its instructions and works without a Skill. Skills are reusable methods that can be shared and used in chat or selected by an agent. An agent can use several published Skills when the job needs them. Triggers have their own section. Include timing in your first request or add it later. You can approve creation and schedule activation together. An agent with a schedule that is switched off reads **Off**; **Turn on** names the schedule and asks you to confirm before it starts. Pausing or changing a schedule leaves the instructions intact, and **Run** remains available for a one-off run. You can follow a run in its chat and open its saved result. Connected actions still require approval for each proposed action; a run waits for that decision before continuing. ## Remove a Skill A Skill's author or a company owner/admin who can access it can open **More actions → Delete Skill** in its canvas. Confirming removes the Skill from the library, search and new runs. Company administration does not reveal private Skills, and being an editor or having publication permission alone does not let you delete somebody else's Skill. For recovery, expand **Archived Skills** at the bottom of the Skills page. Deleted Skills can be restored there for 30 days. Restoration returns an unpublished draft; its author reviews and publishes it again before company use. Existing chats and saved results remain available. ## Settings and mobile use You can open Settings from your current workspace location. Unsaved changes are protected before you leave. The workspace also adapts its navigation, canvas and keyboard layout for phone-sized screens, with a drawer for navigation and a sheet for focused work. You can ask in chat to change personal settings, including voice preferences. The change is shown for review before it takes effect. Where spoken approval is available and enabled by both you and your company, a clear reply can approve an eligible reversible change. Company, team, sharing, and other protected changes still require the approval control. Suggesting that a knowledge page be archived always requires that control, and only a publisher's tap on the page accepts a suggestion; a typed, dictated or spoken message cannot do either. ## Desktop downloads In the desktop review build, open **Settings → Downloads** to see the available macOS, Windows and Linux installers. On a supported desktop browser, your platform is selected first; you can always choose another. macOS downloads distinguish Apple silicon from Intel. Linux also shows NixOS instructions when a source package is available. When verification details are available, expand **Verify this download** to find the signed checksums, signature, public key and its fingerprint. The desktop app opens Production when launched normally. **File → Demo Chat** opens a separate Demo session. Google sign-in appears when configured for the selected environment and opens in your browser. After Google, select **Open Poneva Chat** and allow your browser to open the app to finish signing in. Downloads appear after a release is published. Until then, the page explains that installers are not yet available. ## Current limits Poneva Chat is being introduced in stages. Availability of connections, company features, and review flows depends on your workspace and rollout. This guide describes the workspace surfaces and review behavior; it does not promise that a particular provider is connected, that every provider action is available, or that the service has been deployed to every account. [Voice in Poneva Chat](poneva-chat-voice.md) describes the controlled voice review build and its remaining device-acceptance limits. Nothing in this guide changes the requirement to review a proposed company action.