AI overview
Bitment's AI isn't a single chatbot bolted onto the product; it's a shared assistant with your whole workspace wired in, available in five different shapes depending on what you need. This page maps how the pieces fit together. Each section links to a deeper page with the full detail.
Every one of these surfaces shares the same underlying capabilities: the same tools, the same models, the same proposal-and-approval safety model, and the same knowledge of how Bitment itself works. What changes between them is who's driving and how much autonomy they have.
Where to find it
Chat, Agents, and Tokens each have their own tile in the dock (also available inside a client, under its AI group). The Assistant opens from the AI control in the top menu bar, or the ⌥ Option+A shortcut, from anywhere. Context lives at the foot of Settings › AI, and per-client context lives on each client's own Context page.
Chat: the general-purpose assistant
Chat is a conversational assistant that can answer anything a capable assistant can, and can also act on your email, calendar, tasks, clients, and files, at your own permission level. Conversations are saved and organized in a history list, and you can group related work into Projects, each with its own knowledge base, standing instructions, and memory of what it's learned about that topic over time. See AI Chat.
The Assistant: a pane that knows what you're looking at
The Assistant is the same underlying assistant, opened as a pane beside whatever you're working on. Unlike Chat, it's page-aware: it knows what app and record you have open, and it can move your view to the thing it's talking about, fill in a field, or stage a change right where you'd make it by hand. See AI Assistant.
Agents: autonomous colleagues
An agent is an AI colleague that works when you aren't watching: woken by events (new mail arriving, a meeting transcript finishing) or by a schedule you set, it does a piece of work on its own and leaves you things to review. Agents have a soul (their standing brief), skills (written playbooks for specific jobs), and memory (facts they pick up on their own), and every action they take is gated by a permission level you choose per capability. See AI Agents.
Tokens: what it all costs
Every AI call, whether from you in Chat, from the Assistant, or from an agent working on its own, is metered and priced. Tokens is where you see the usage ledger by person, by agent, and by feature, plus the billing side: credits, spend limits, payment method, and invoices. See Tokens and billing.
Context: the standing facts it always knows
Rather than re-explain who you are and how your firm operates every time you open a conversation, you write it once as a standing Context prompt (personal and firm-wide, at the foot of Settings › AI), and it's included in every AI conversation from then on. Clients carry their own, narrower version of the same idea: a per-client Context document that's injected only when that client is mentioned. See Context and memory.
Models and bring-your-own credentials
Every composer that starts an AI conversation carries a model picker, and the model you pick also decides which firm's AI credential pays for the call. Your firm connects its own credentials (Anthropic, AWS Bedrock, OpenRouter, Ollama, and more), and every model those credentials unlock shows up in the same picker, wherever you are in the product. See Models and providers.
Proposals: nothing changes without your approval
This is the core safety property of the whole system: anything the AI would change in your data is proposed, not done. It stages the change and shows it to you as a card, whether that's a drafted email, a task, or a full field-by-field diff for anything else, and it only takes effect when you approve. You can elevate a session to let it commit directly for a specific capability when you'd rather not review every step, but the default everywhere is propose-then-approve.
Product knowledge: it knows Bitment, not just your data
Ask "how do I set up a split inbox" and the AI reaches for the actual product manual instead of inventing an answer or guessing at UI that doesn't exist. This works across Chat, the Assistant, and agents, so a question about how something works gets a grounded answer with the real navigation path, and in most cases the AI can just do the thing for you instead of only explaining it.
Attachments and voice
Attach documents, spreadsheets, images, and PDFs to any message and the model reads them, including scanning them visually rather than only as extracted text. On a phone or tablet you can also snap a photo or scan straight into the conversation. You can dictate instead of typing wherever you'd normally type a message; see Settings › Voice-to-Text.