Skip to content

Chat on a Phone

The composer, streaming answers, choosing an agent, what a tool result looks like on a phone, drafted documents, and attaching a file — chat is the part of HAQQ mobile and web genuinely share.

Starting a conversation

A new chat opens on a welcome state — “What do you need help with?” — with three workflow cards: Draft a contract, Review a contract and Research laws. Tapping one opens a sheet of ready-made prompts drawn from HAQQ’s public prompt library, each showing a title and the first few lines of the prompt so you can judge it before picking. Choosing a prompt fills the composer and raises the keyboard — it does not send.

The welcome state disappears as soon as you send, without navigating anywhere; the conversation joins the history list on that first send. If you have already typed something of your own, inserting a prompt asks before it overwrites the draft, with a Don’t ask again option that is remembered permanently.

Once a reply finishes, up to three follow-up chips appear above the composer, matched to your last message. On a brand-new chat a chip sends straight away; inside an existing conversation it fills the composer instead so you can edit it. Chips are computed on the device, so they work without a connection — the workflow library does not, because it is fetched and then cached for a day.

The composer

A multiline input capped at 8,000 characters, with Attach on the left and the agent chip and send button on the right. Return inserts a newline; sending is always an explicit tap. Send is dimmed until there is text, and stays blocked while a file is still uploading or being indexed. While an answer streams, the send button becomes a red stop control that aborts the response.

Whatever is in the composer is saved per conversation and restored when you come back to it, and cleared once the message is sent.

Watching an answer arrive

Text streams in as rendered markdown — headings, lists, quotes, code blocks and tables, the tables building row by row rather than re-laying out on every chunk. Reasoning appears as a collapsible block that opens itself while the model is thinking, reveals the text as it arrives, and collapses about a second after the message finishes; it reads Thought for N seconds when done. Turning Show reasoning process off in settings hides these blocks entirely.

Under each finished answer sits a row of actions: Copy; a Share action on messages that produced a document, offering PDF or DOCX through the system share sheet; the credit cost of that response; and thumbs up / thumbs down. On iOS the first rating on a message follows up with an optional comment box.

Choosing an agent

The agent chip in the composer is the only way in. It opens a Select Agent sheet listing every active agent with its name, a short description and — for the ones you can use — the underlying model. The platform default carries a Default tag and the active one is ticked. Agents your plan does not include are still listed, dimmed, with a lock badge and no model line; tapping one closes the sheet and opens the upgrade sheet inside the app rather than switching agents.

Switching agents inside an existing chat is saved to that chat, so reopening it later restores the same agent. Mobile reads your default-agent preference but offers no way to change it — that is set on the web, in AI settings. There is no search box, no capability filter and no compare view in the mobile sheet.

What the assistant shows while it works

When the assistant does something on your behalf, the action renders as a compact card inside the reply rather than as raw text. Everything is collapsed by default — tap a header to open it.

  • Web search — the query in the header and, once finished, a Sources list of titles and hostnames. Tapping one opens it in the phone’s browser.
  • Knowledge search — results grouped into Company, Team, Personal and Public knowledge, each with a count, and each match shown with a score bar and a snippet.
  • Document drafting — a card whose title arrives part-way through the draft so it is never a placeholder for long, and which becomes tappable once the document is complete.
  • Everything else — a generic card with the action’s name and a status of Processing, Retrieved or Error, opening onto a short summary of what came back.

A search that finishes with nothing to show is removed from the transcript rather than left as an empty shell, and a memory lookup that found nothing disappears the same way. Actions that are still running, or that failed, are always shown.

eFirm data actions — contacts, invoices, expenses, hearings, matters, tasks and more — run the same way, as these generic cards. See Mobile vs Web for what that leaves out compared to the web app’s record cards.

Documents the assistant drafts

Once a conversation has produced a document, a pill under the header shows the count and opens a sheet listing every document in that chat with a Text or Sheet badge and a relative timestamp. Tapping a document opens a full-screen viewer that renders it as formatted markdown, selectable, with light and dark styling and full right-to-left support in Arabic.

A completed document can be edited in place. The edit toggle swaps the reading view for a markdown source editor with a formatting toolbar above the keyboard — headings, bold, italic, bullet and numbered lists, quote and inline code. Edits autosave about a second after you stop typing, and are flushed when you leave; there is no Save button and no discard. Every save creates a new version.

The Versions button appears once more than one version exists and lists them oldest-first, the newest marked Current. Selecting an older one loads it read-only behind a “Viewing version X of Y” banner with Back to latest. Restoring is therefore a read-and-copy flow rather than a one-tap revert, and there is no side-by-side diff on the phone.

Share as exports the document as PDF or DOCX through the system share sheet. The file is named after the document title.

Attaching files

The Attach control opens a sheet with four tiles: Image, Camera, Upload File and Storage. The document picker is filtered to the formats the indexer can read — PDF, Word, Excel, PowerPoint, CSV, plain text, HTML, Markdown and common image types — so unsupported files are never offered.

Storage is a full-screen browser over your workspace’s eFirm storage: top-level folders, nested folders and files, with a breadcrumb trail and a search box that filters the current level. A pinned My Personal Files entry at the root is your own AI storage folder, and is where every camera, photo-library and document upload lands. Inside a folder you can upload a file, upload a photo, or create a new folder; files are picked with a checkbox and committed to the conversation with Attach.

Attached files appear as pills above the composer showing live status — uploading, then pending or processing, then the file type and size — with an × to remove each one. A failed file stays visible with a plain-language reason so it can be retried or dismissed. Sending is blocked until every file is ready.

The data room this creates, and how to manage its files afterwards, is covered on Organising Your Work.

Related topics

Was this page helpful?