Skip to main content

Quickstart

In this walkthrough you'll go from a fresh account to a published agent you can call over the API: create a tool, test it, attach it to an agent, chat with the agent, publish, and make your first API call.

Step 1: Sign up​

Create an account at the builder. Your personal company is created automatically and becomes your active workspace — everything in this quickstart lands there.

Open the Builder →

Step 2: Create a tool​

Tools are the building blocks agents work with. We'll make a simple LLM-powered summarizer.

  1. In the sidebar, open Agentic Workflows → Tools and click Create a Tool.
  2. Name it SummarizeText and describe it clearly — for example, "Summarizes a passage of text into three bullet points." Agents read the name and description to decide when to call a tool, so make both intuitive.
  3. Pick the LLM_CALL tool type. Choose a model and write the prompt template:
Summarize the following text into three concise bullet points:

@@{text}
  1. Add an argument named text with a description like "The passage of text to summarize." The @@{text} placeholder in the prompt template is filled with this argument at runtime.
  2. Save the tool.

Prefer code over prompts? Choose the AUTESSA_SCRIPT tool type instead and write the logic in AutessaScript — same arguments, same testing, same everything else.

Step 3: Test the tool​

  1. On the tool page, click Execute Tool.
  2. Fill in the text argument with any paragraph and run it. You'll see the summarized output — and you can also Execute with Streaming to watch longer-running tools report progress.
  3. Iterate on the prompt template until you like the output. Nothing here touches production; you're editing the draft.

Step 4: Create an agent​

  1. Open Agentic Workflows → Agents and click Create an Agent. Choose the Simple type — you pick the tools.
  2. Name it (for example ResearchAssistant) and give it instructions — the system prompt that defines its role and rules, such as "You help users digest long documents. When the user provides text, use SummarizeText and present the result."
  3. Choose an AI model, and attach your SummarizeText tool.
  4. Save the agent.

Step 5: Test conversationally​

Open the agent's test conversation, paste some text, and ask for a summary. Watch the agent decide to call your tool, pass the argument, and answer with the result.

Step 6: Publish​

Everything so far lives in your draft sandbox. To make a version customers and APIs use by default:

  1. On the agent page, open Manage and choose Publish. Do the same for the tool.
  2. Open the agent's Details panel and copy its ID — a short string like V1StGXR8Z5. That's your resourceId for API calls.

If a published version ever misbehaves, you can roll back to a prior version from the same Details panel — see Versions.

Step 7: Call it from the API​

  1. In the sidebar, open API Management → API Keys and create a key. Under Resource Access, grant it access to your agent. Copy the key — it's shown only once.
  2. Execute the agent with a single request (no Bearer prefix on the key):
curl "https://api.autessa.com/clients/agents/execute?resourceId=V1StGXR8Z5" \
  -H "Authorization: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": [
      { "inputType": "TEXT", "content": "Summarize this: Autessa lets you build tools, agents, and multi-agent workflows..." }
    ]
  }'

The response contains the agent's output in the output array. For multi-turn conversations, streaming over WebSockets, and audio input and output, see the Agent API reference.

You've shipped your first agent​

That's the whole loop: you wrote a tool, taught an agent to use it, watched it reason in a live conversation, and put a published version behind a real API endpoint. Every production system on Autessa — voice assistants, document pipelines, multi-agent workflows — is built from exactly these moves, repeated and composed.

Where to next:

  • Build something real — the Build your first app tutorial takes these same moves further: a todo app with an agent that reads your email every night and files it as tasks. About 40 minutes, and it touches nearly the whole platform.
  • Solidify the mental model — Core Concepts pins down the vocabulary you just used in practice: tools, agents, versions, variables, and scopes.
  • Give your agent memory — create an AutessaDB table so your agent can store, search, and react to data.
  • Publish safely — Versions explains the draft/publish/rollback cycle so you can iterate without breaking production.
  • Call it from your product — read Authentication, then the Agent API reference for streaming, conversations, and audio.