Skip to main content
Version: v2.0

Quick start

This page takes you from an empty Agenta project to one agent doing a real piece of work. It takes about five minutes and needs no connected apps.

ChatGPT gives you one prebuilt agent that already works for everything. Agenta gives you an empty one, and you tell it what job it is for. You do that by chatting with it, not by filling in forms.

Before you start​

Step 1: Describe the agent​

Open your project. The home screen asks What do you want to build?

Type what you want the agent to do. For this run, use:

Turn my rough notes from a week of work into a short weekly update
I can send to my team. Plain sentences, no bullet lists.

Click Create agent. Agenta creates the agent, names it, and opens its playground.

That one sentence is the whole agent so far. The rest of this page fills it in.

Step 2: Connect a model​

The playground opens in Build mode. The Configuration column is on the left, the conversation on the right. Above the message box a banner reads Add your model provider key to run this agent.

  1. In the Configuration column, click Model & harness.
  2. Leave the segmented control on API key.
  3. Click the OpenAI tile.
  4. Paste your key into API key * and click Save.

The section header names the two things the agent runs on: a harness and a model. The model answers messages. The harness drives it, runs the tools it asks for, and feeds the results back. See Harnesses and models.

Step 3: Answer the agent's questions​

The conversation opens with a card holding your description and a Start button. Click it.

The agent picks up your description and asks you about the job: who the update is for, what usually goes in it, how you want it to sound. Answer in plain sentences, the way you would brief a new colleague.

As you answer, the agent writes what it learns into its own configuration. Watch the Instructions section in the Configuration column while you talk.

Step 4: Give it real work​

Now give the agent something to do. Paste a handful of rough notes from your week and ask for the update:

Here are my notes from this week. Draft the update and save it as
update.md.

A new agent reads on its own and asks before it writes. If a card headed Approval needed to continue appears, click Approve. See Permissions.

The agent drafts the update and writes the file. Files it creates appear in the conversation as a card with a Download button.

Step 5: Correct it, then keep the version​

Tell the agent what to fix, and tell it to keep the correction:

Too long, and don't open with a greeting. Keep every update under
120 words. Remember that from now on.

The agent updates its instructions. The Instructions section shows the new text, and the state dot next to the version pill in the header turns to Draft.

Click Commit, keep New version selected, and click Commit. The agent is now at version 2, and the change is recorded.

What you have​

An agent that knows one job, holds the corrections you gave it, and has a version history. You built it by describing the work and correcting the result.

Next​

  • Build your first agent is the full path: instructions, tools, a skill, permissions, and a committed version.
  • Agents explains what an agent is made of.
  • Automations covers running an agent on a schedule or on an event in a connected app.