The maxinja assistant

Brief maxinja beside the campaign, record or workflow you are working on. The chat pane keeps the conversation and its context together as you investigate data, build content and refine a journey. maxinja operates inside the app; the Public API does not start these conversations.

This page covers the pane itself. What maxinja can and cannot build, and what a request costs, are on the maxinja overview.

Opening the pane

The pane has three sizes:

On a narrow window, maxinja opens in full view instead of a side panel.

Chats

Each conversation is a separate chat. From the pane header you can:

• Start a new chat.
• Open the chat list to search conversations, grouped into Starred and Not Starred.
• Rename, star, or delete the current chat once it has messages.

Titles are generated automatically from your first message.

Writing a request

Type in plain language. Enter sends; Shift+Enter adds a line. What to put in the message is on the overview: describe the outcome, not the data.

Adding context

maxinja is one assistant for the whole space. You point it at a specific thing by attaching a reference to your message.

• @ or # (or the Add context button) opens a picker. Reference types: schema, segment, contact, object, template, broadcast, workflow, workflow step, webhook, form, brand kit.
• The Current view chip attaches the page you are on. With it attached, "add a condition for recent buyers" needs no name.

References attach per message, so one chat can move between entities.

Attaching files

Attach with the paperclip, drag-and-drop, or paste. Accepted types: .png, .jpg, .jpeg, .pdf, .txt, up to 100 MB each. Each attachment shows upload progress and can be removed before you send.

During a run

maxinja handles one request at a time. While it works, it:

• Shows progress updates and its reply as it works.
• Hands larger jobs to helper agents and reports back what each one did. See the agentic flow.
• Shows a preview of what it built, with a button to open it.

Messages sent during a run steer that active work. Include the correction and the result it applies to. You can also stop the exchange; completed changes remain saved. If maxinja asks you a question mid-run, you answer it in the question's own input.

When maxinja asks a question

maxinja can ask for a decision inline, often with suggested answers to click. While it waits, the main composer locks so your reply goes to the question. Some questions show a countdown and apply a default if you do not answer in time.

Stopping and editing

Choose tools and models for the conversation

The session’s AI settings choose its model and registered MCP servers. A running exchange retains the selection it started with. Change settings for subsequent work, or stop and start a new request when the task requires a different capability.

An image or PDF already in the conversation remains part of its context. A model that cannot accept that modality is rejected rather than silently substituted. For connected MCP tools, review the selected server and the task you are authorizing: those tools can perform actions in the connected service, and deleting this chat will not reverse them.

Workspace conversations require write access. A read-only role cannot start a workspace run merely because the question itself sounds read-only.

Previews

When maxinja produces something to review, it opens a preview in the right of the workspace:

• Audience preview: which schema it filters, the filter in plain English, what it references, and how many contacts match out of the total.
• Email preview: a rendered sample to inspect. A later personalized render can use newer data and produce different wording.

Attachments and generated files open there too. Switching maxinja to full view clears the preview.

Related

• maxinja overview
• The agentic flow
• Broadcasts
• Templates
• Adaptyle