Talking to agents

In Lowkey you don’t talk to one faceless AI — you talk to named agents, each with its own name, avatar, and personality. It’s less like prompting a model and more like messaging a small team: you pick who to ask, and they answer in their own voice.

Lowkey comes with one starter agent to get you going. You can rename it, give it a new look and personality, and add as many more as you like from Settings → Agents — each running on the AI engine you choose.

Different agents, different strengths

Each agent runs on a different AI engine underneath — Claude, GPT, Gemini, or Grok — and they’re genuinely good at different things. Some are careful and thorough; some are fast; some are especially strong at code or long, fiddly work. You don’t have to learn which is which to get started: pick one and ask. If another agent would do the job better, the one you asked will usually say so.

The personality is a friendly layer on top — it changes how an agent talks, not what it can do. Two agents can do the same work and just phrase it differently.

Picking who answers

Near where you type, the send button shows the agent you’re about to address — that’s your readout of who’s getting this message. By default it’s the conversation’s usual agent, and whoever answered last stays selected for your next message, so a back-and-forth just keeps going.

To send to someone else, type @ and start their name — a picker appears, and choosing one addresses your message to them: “@Alyx, can you check this file?” This is the same way you address people in a shared project, so agents and teammates live in one roster.

Bringing in more than one

A single conversation can involve several agents. Because they all read the same thread, handing work from one to another keeps the full context intact — you can let Lucy talk something through, then “@Alyx, take it from here and write the code,” and Alyx already knows everything that was said. No re-explaining, no copy-paste between chats.

Letting an agent delegate

For a big or bounded piece of work, an agent can spin off a sub-worker — a helper that goes off, does that one task, and reports the result back into your conversation. You’ll see the handoff happen; the answer comes home to the thread you’re already in.

Staying in control

You’re always the one steering. The send button tells you who’s addressed before you send, switching agents never loses the conversation’s history, and any work an agent hands off still surfaces back to you. If a turn heads the wrong way, you can stop it and redirect — see /sessions/.