Agents

Design an assistant for one job in a short chat with June, activate it, and hand it tasks by name.

An agent is an assistant you design for one job, with its own persona, instructions and know-how. You design it in a short chat with June, activate it on the Agents page, and then hand it tasks by name in your chats.

Customer-agent repositories and streaming

Customer-agent Git, builds and streaming are temporarily unavailable while the runtime is rebuilt. Do not upload repositories or send requests to previously supplied customer-agent Git or runtime URLs. There is no replacement endpoint to configure yet.

The chat agents described below remain available.

Design an agent

On the Agents page, choose Design an agent. A new chat opens with a first line already in the composer: "I want an agent to take over a workflow for me." Edit it or send it as it is.

June asks about the work, one short question at a time: what you do today, what starts it, how often, and what the finished result looks like.

Once it can describe what starts the work, the work and the result, June creates the agent as a draft and links you to its page.

I want an agent that turns my weekly sales call notes into a short summary for the team

Try in June

Design an agent that drafts polite payment reminders. Ask me for examples of our house style before writing its instructions

Try in June

The designer is this chat: there is no form to fill in. June derives the name, tone and schedule from what you describe and tells you what it assumed, so correct it in the chat.

Review and activate

A new agent does nothing until you activate it. Its page shows what June drafted:

Its soul: the persona and voice it answers in, and its instructions.
Its skills: know-how bundled with it.
Scheduled work it runs on its own, if you described any.
Its version history: draft, active and archived versions.

Choose Activate on the draft version and confirm. That version becomes the one in use right away, and the version that was active before is archived for good: an archived version cannot be activated again. Only you can activate your agents.

The scheduled work the version lists is added to Schedules, usually paused. Open Schedules to choose where its results go, then turn it on.

Hand an agent a task

Name the agent in a chat in June, on the web or in the mobile app, in plain words. There is no special syntax: June knows your active agents by name, recognizes which one you mean, passes it the task and shows you its answer. Until you have an active agent of your own or subscribe to one, there is none to hand a task to, so June answers itself and can help you design one.

Ask my sales notes agent to summarize the call notes I pasted above

Copy into your current chat

You can also ask June to start a separate thread that runs as the agent. A thread June just started can ask your agents from its first reply, but cannot start another thread then.

The agent uses its own soul, instructions and skills, without tools, your Soul or your memories. It sees only the task June passes it and keeps no memory between tasks.

Include the notes, examples and constraints it needs, or ask June to pass them along. If the task needs material from a connected account or file, ask June to gather it first.

Archive an agent

On the agent's page, choose Archive agent and confirm. It stops answering anywhere, its scheduled work pauses, and its active version is archived. You cannot unarchive it from its page, and agents cannot be deleted.

What an agent does not do

It does not use tools or your integrations when you hand it a task: it answers from its persona, instructions and skills.
It does not take tasks in Slack: naming it there does not reach it.
It does not belong to an organization: your agents are yours.