Create an AI agent

Give an agent a persona, memory, and a workflow.

This guide isn't translated yet. Showing the English version.

このページの目次

1. Start with the agents already built
2. Build your own
3. Check the fields
4. Connect knowledge and tools
4. What to build
5. Two ways to call an agent
Generating images

An agent is a virtual teammate that works alongside your team. Beyond the default assistant MAKi, you can build agents that fit how your organisation actually works.

1. Start with the agents already built

Press + next to AGENTS in the sidebar to open the AI agents screen. Before building one, look at what is already there. For most teams that is the whole job.

Built-in agents on the AI agents screen — Press Add to team and it is ready. The filters at the top narrow by work management, research, documentation, support and data.

Task Manager: summarises your team’s task progress each day and follows up with owners.
Researcher: picks the top stories each morning and summarises them.
Knowledge Manager: organises the chosen hub’s conversations and notes every evening.
Operations Manager: answers team-benefits questions and shares the team update each weekday morning.
Inventory Manager: register stock and it remembers the quantities, reporting usage and restocking.
Customer Support: greets external customers and answers from a linked Q&A note.

2. Build your own

If none of them fit, there are three ways to build one, top right.

Create an agent: fill in the form yourself.
Create from one line: write a single sentence and MAKi fills in the rest.
Clone an agent from me: build one that carries your voice and working context.

Creating an agent from one line — Write one line — "Name is Debbie, a meticulous senior dev who helps with backend code reviews" — press Auto-fill, and the form below fills in. Everything stays editable.

3. Check the fields

However it was filled in, these are the values worth a look before saving.

The name, role, tone, personality and instruction fields — Only the name is required. The rest set the grain of its answers, so the more you fill in the more consistent it reads.

Name (required): used in the sidebar and when you call it with /.
Role: what it does, e.g. Project Manager, Code Reviewer.
Tone: how it speaks, e.g. casual, professional English.
Personality: its disposition, e.g. cheerful and positive, calm and logical.
Additional instructions: free-form rules, e.g. keep responses under 200 characters.

4. Connect knowledge and tools

Linked notes and tool restriction settings — Linked notes give the agent something to know. Tools are unrestricted by default; narrow them when you need to.

Linked notes: attach notes holding your rules or product facts. Up to 4,000 characters per note, and up to 5 notes go into the prompt.
Available tools: unrestricted by default. Turn on Restrict tools to limit what it can reach for.

4. What to build

Agents pay off most when they are pointed at recurring work.

Task management: catch work out of the conversation and sort out owners and deadlines.
Inventory and supplies: collect status in a fixed format and flag what is running out.
Reporting: hand over the weekly report or the meeting write-up.

Build several, each shaped around a real part of your operation, and call the right one where you need it.

5. Two ways to call an agent

Inside a hub, type / and pick the agent, and it answers right there, without leaving the conversation.

For a longer back-and-forth, click the agent under AGENTS in the sidebar. That opens a 1:1 room, much like any other LLM chat.

A 1:1 room with an agent, opened from the sidebar — ① Click an agent under AGENTS to open its 1:1 room. In here you just ask; no slash command needed.

Generating images

Agents can generate images too. Describe what you want in the 1:1 room or in a hub. The result lands as a mark, so it is shared with the team immediately, and you can open it in the viewer to zoom in or annotate it.

お役に立ちましたか？