AI Chat
Chat is the general-purpose AI assistant, with your Bitment workspace wired in. It can answer anything a capable assistant can, and it can also act on your email, calendar, tasks, clients, files, and the rest, as you, at your own permission level. Conversations are saved, searchable in a history list, and can be shared with teammates or organized into Projects for a recurring topic or client.
Chat is one of two conversational surfaces. The other is the Assistant, a page-aware pane opened from the AI control in the top menu bar, which is the same underlying assistant but scoped to what you're currently looking at. See AI Assistant for the distinction.
Where to find it
Chat in the dock (also inside a client, under its AI group, scoped to that client).
Key concepts
| Term | Meaning |
|---|---|
| Model picker | Chooses which AI model answers, and which firm's connected credential pays for the call. |
| Proposal | A staged change the AI wants to make, shown as a card you approve or reject before anything takes effect. |
| Plan | A step tracker attached to a conversation for a multi-step request. |
| Sub-agents | Several parallel AI workers inside one turn, each with its own context, for work that splits cleanly. |
| Workflow | A saved graph of sub-agent steps with dependencies between them, reusable by name. |
| Project | A Chat-only workspace bundling standing instructions, a knowledge base, and a set of conversations around one topic or client. |
Start a conversation and pick a model
- Open Chat and start typing.
- Use the model picker in the composer to choose which AI model answers. It lists every model available through your firm's connected AI credentials and your own personal ones. Picking a model also picks which firm pays for the call.
- Optionally set a default model so new chats start on it, either from the gear next to the New Chat model picker, or in Settings › AI (both edit the same setting).
Your conversation history lives in the app's sidebar.
Attach files
Attach documents, spreadsheets, images, and PDFs to a message and the model reads them; PDFs and images are read visually, page by page, not just as extracted text.
On a phone or tablet, the attach button also offers Take Photo, Record Video, and Photo Library, so you can hand the model a picture straight from the device. The same composer is shared by the Assistant pane and agent chat.
Once sent, an image shows as a small thumbnail in the conversation; other files show as a chip with the filename and size. Click either to open the full preview, where you can also download it.
You can also paste an image directly into the message box (Cmd+V / Ctrl+V), a screenshot or a picture copied from anywhere else, and it attaches the same way. Pasting ordinary text still works normally.
Ask it to do things, not just answer
Chat has the same tools you'd use by hand: reading and drafting mail, checking the calendar, creating tasks, reading and organizing files, looking things up on the web, running code, generating Word/Excel/PDF documents, and more.
Anything that changes your data is proposed, not done: the assistant stages the change and shows it to you as a card in the conversation, taking effect only when you approve. The card shows the full details of what would change, an email as an editable draft, a task as a task, and for everything else (creating an agent, changing its permissions, a workspace setting) every field of the staged change, so you're never approving blind. You can elevate a session to let it commit directly for a given capability when you'd rather not approve each step; the default everywhere is propose-then-approve.
Tools load as they're needed. If you ask about a domain that isn't currently loaded, Chat pulls those tools in mid-conversation, so it's never stuck saying it can't do something it actually can.
When Chat shows you something from another app, that card is a live handle on the real thing, not a screenshot. An email it read for you, or a draft it prepared, carries an Open button: click it and the Emails app opens (or comes forward) with that conversation or that draft ready.
Web research and code
Ask for anything current or external and Chat searches the web rather than telling you to look it up yourself. For heavier questions, it can run a deeper research pass.
It can also write and run code in a sandbox against your data, the practical way to do bulk analysis, reformatting, or anything that would otherwise take a hundred clicks.
Ask a data question directly
For counts, filtered lists, aggregations, rankings, or anything that crosses several apps at once ("how many of my clients haven't been invoiced this quarter"), Chat can write and run its own read-only database query instead of chaining a series of per-app lookups. This is the fastest way for it to answer a data-shaped question, and it's read-only: it can only see what you're already permitted to see, and it can never use this to change anything. There's nothing to set up; it reaches for this on its own when a question calls for it.
Plans, sub-agents, and workflows
For bigger requests, Chat can:
- keep a plan, a step tracker attached to the conversation that you can watch and check off;
- fan out sub-agents, several parallel workers inside one turn, each with its own context, for work that splits cleanly;
- run a workflow, a graph of those workers with dependencies between them, which you can save to a library by name and re-run later. The / picker in the composer loads a saved workflow.
A workflow renders in the conversation as an interactive diagram: each step is a box, the arrows show what feeds into what, and a dashed loop arrow marks a revise step. Click a step to read its task (or, on a finished run, its report), drag to pan, zoom with ⌘/Ctrl and scroll (or the +/− buttons), and use the expand button for a full-screen view.
The sidebar's Workflows page lists your saved workflows. Opening one lands in a canvas editor: add steps, drag a step's connector handle onto another step to link them, and click any step or connection to edit its details in the side panel.
Group chats
A conversation can be shared with teammates, making it a group chat with the AI in it. Each person's messages are attributed, so the AI answers knowing who asked what.
Export a chat as a PDF
Open the ⋯ menu in a conversation's header and choose Export chat as PDF. The whole conversation, not just what's on screen, downloads as a letter-sized document with each message's author and time, useful for sharing a transcript with a teammate or a client who isn't on Bitment.
The same export exists on every AI chat surface in the product: the embedded panels in Emails, Projects, Forms, and Signatures (same ⋯ menu), the Assistant pane (a download icon in its header), an agent's chat (its page's ⋯ menu), the import chats in Clients and Bookkeeping (a download icon in their headers), and the Campaigns drafting chat (a download icon in its header).
Projects: a workspace for a recurring topic or client
Not to be confused with the Projects app for task management, this is a Chat-only workspace concept.
- Click Projects in the Chat sidebar to open the projects list.
- Click New project.
- A project bundles three things: custom instructions (a standing system prompt applied to every chat you start inside it), a knowledge base of documents, and the conversations you have inside it.
- Choose whether the project is personal (only you see it) or shared with a firm (every active operator at that firm can open it, read its knowledge, and add their own chats). Only the project's owner can archive, delete, or move it between personal and a firm.
Inside a project's page, the knowledge rail lets you add documents three ways: type or paste a text snippet, upload a file, or pick an existing file from the Files app. On a phone or tablet, uploading also offers the camera and photo library. Every document you add is read into the model's context on every chat inside the project; you don't have to attach it to each message.
New chats you start from inside a project are automatically tagged to it. They still show up in your main Chat history (with a small project badge); a project doesn't hide your chats, it also lists them on its own page. A chat inside a shared project stays private to the person who started it; teammates see their own chats there, not yours. Move an existing conversation into (or out of) a project from its row menu in the sidebar ("Move to project").
Star a project to pin it near the top of the list; archive one you're done with (it stops appearing as a target for new chats, but its past conversations and knowledge stay intact).
Project memory
Inside a project, the assistant also remembers things it learns from your conversations there over time: short facts and preferences it picks up as you chat, not something you have to teach it directly. This happens automatically in the background after a conversation settles, and those memories are then quietly included in every future chat you have in that project.
Memory is scoped to you personally, even in a project shared with your whole firm; a teammate's chats build their own memory, not yours. It's on by default, but the project's owner can turn it off from the project page. Review everything it's remembered, edit an entry, or delete one you don't want kept, from the memory panel on the project's page.
Chats inside other apps
Several apps embed this same assistant scoped to what you're doing there: a panel in Emails, and panels in Projects, Forms, and Signatures. Every one of those conversations also appears in the main Chat history, marked with a small badge saying where it started, so Chat is the one place to find any conversation you've had, and you can open and continue any of them right there. Each app's own panel still keeps its scoped thread list too. Chats with your AI Agents show up in the history as well, badged with the agent's name; opening one takes you to that agent's page, where its chat lives.
Mobile
On a phone, a left-edge swipe on the chat transcript opens the conversation history drawer; it's armed only when the touch starts in the edge strip, so scrolling inside the transcript itself is unaffected.
Cost
Every AI call is metered. See Tokens and billing for what's been spent, by whom, and on what, plus credits, limits, and payment.