Build Your Own AI Agent, No Code

You describe it; Aria builds it

Aria is the agent whose job is making other agents. You tell her what you want — typed, or dictated out loud — in the register you would actually use: “someone who chases me about the gym but is not annoying about it”, “a chief of staff for a one-person business”, “a Lebanese grandmother who tells me what to cook with what is in the fridge”. She asks a couple of questions, then produces a finished agent: a name, a role, a personality, a tone, a voice, a schedule, and a generated portrait you can reroll until the face fits.

It usually takes two messages. Nothing you write is a prompt in the technical sense — there is no template language, no variables, no place where a stray brace breaks anything — and if you abandon it half-finished, the draft is still there when you come back.

What you get to control without writing anything

• Role — what this agent is for, in a sentence. It is the thing that decides which of your facts it is shown and which teammates it hands work to.
• Personality and tone — warmth, formality and energy, as sliders. The same instruction at low warmth and high formality is a genuinely different colleague.
• A voice — its own, so it sounds like itself on a call as well as reading like itself in a thread.
• An hour — the time it reaches out to you first, daily or weekly. Change it later by telling the agent in a sentence.
• A face — a generated portrait, rerolled until you like it.

Or start from one that already exists

There are 21 ready-made agents across 6 categories — focus, work, home, chill, create, learn — and hiring one is a tap. Each arrives already written, already voiced, with a suggested hour: a morning caller, a focus companion, a household manager, a journaling partner, a devil's advocate, a language tutor. Edit anything about it afterwards; a template is a starting point, not a cage.

If even that is more decisions than you want, 4 starter packs hire 3 agents at once, already scheduled to work together — mornings and focus, run my home, run my work, learn and grow. One tap, no messages, and tapping twice never gives you duplicates.

Connecting it to your own systems, still without code

When an agent needs to know something that lives in your systems, you add a connection: a name, a plain-English description of when to use it, an https URL, an optional secret header, and the parameters. There is a test button and a few ready-made recipes. The agent then reaches for it on its own when the answer is in there. The one hard limit is security, not difficulty — the service has to be reachable at a public https address, so anything on your home or office network needs a tunnel first.

Whose model, and what it costs

By default you do not bring a key at all; building and running agents is included in the plan. If you already pay OpenAI or Anthropic and would rather the tokens went on your own account, Own Key is $7 a month — you paste your key in settings, it is encrypted and never logged, and chat volume becomes a matter between you and your provider. Either way the builder is the same builder. Plans in full are on pricing.

Once an agent exists, the interesting part starts: it texts you first, it remembers you, and it can sit at a round table with the others — the features page walks a whole day of it, and the demo shows it in two minutes with no signup.