Skip to content

Getting started with Hadron

By the end of this tutorial you'll have a chatbot agent that greets you, asks your name, and remembers it across sessions — running entirely in the Hadron portal, no integration code required. Plan on about ten minutes.

You will:

  1. Sign in to the portal and pick (or create) an organization.
  2. Create a chatbot agent through the Create Chatbot Agent wizard.
  3. Paste an LLM API key into the agent's Settings tab.
  4. Open the agent's Chat tab and have a real conversation.
  5. Reload the page and confirm the chat persists.

When you're done, Building a chatbot agent is the natural next read — it covers conversation design in depth.

What you'll need

  • A modern browser, signed in to hadronmemory.com.
  • Membership in a Hadron organization. If you don't have one, the Create Organization flow on the dashboard is two fields and thirty seconds.
  • An API key from one supported LLM provider: Anthropic, OpenAI, GLM (z.ai), or AWS Bedrock. Anthropic and OpenAI are the fastest paths; GLM and Bedrock have extra setup we'll skip.

That's it. No CLI, no MCP host, no code editor.

Step 1: Open the Create Chatbot Agent wizard

  1. After sign-in, click into your organization from the dashboard.
  2. On the org page, click Create Chatbot Agent.

You're now on a wizard at /app/orgs/{orgId}/create-chatbot. It scaffolds the four moving parts a chatbot needs (an agent, a system memory, a welcome conversation, a fallback conversation) in a single submit.

Step 2: Fill in the wizard

The form is short:

Field What to enter
Agent name My first chatbot (or anything you'll recognise)
Agent URN Auto-fills as a slug from the name. Leave it.
Description Optional. "My first Hadron chatbot." is fine.
System prompt Optional. "You are friendly and brief. One short paragraph max per reply." gives the LLM a personality without over-scripting.
System memory URN Auto-fills (<agent-slug>-system). Leave it.
Include knowledge memory Leave unchecked for this tutorial.
Visibility Personal is the default and is correct here.

Click Create Chatbot Agent. The wizard:

  • Creates the system memory that holds your bot's conversation designs and prompts.
  • Creates a setup conversation with one stage (onboard) whose prompt asks the user for their name.
  • Creates a fallback conversation that gracefully handles questions the bot can't answer.
  • Creates the agent itself, wired up to the system memory.

You land on the agent detail page at /app/agents/{agentId}.

Step 3: Configure your LLM provider

You'll see four tabs across the top: Memory, Chatbot Control, Settings, App. The Chat tab is missing — it appears only after an LLM provider is configured.

  1. Switch to Settings.
  2. Scroll to the LLM provider section.
  3. Click Configure.
  4. Pick your provider from the dropdown (Anthropic or OpenAI).
  5. Type a model identifier — claude-4-opus for Anthropic, gpt-4o-mini for OpenAI. The placeholder updates with the right shape per provider.
  6. Paste your API key into API key.
  7. Click Test.

A green ✓ Works toast with a one-sentence reply means the key is valid. Click Save.

The form closes. The status row now reads ✓ Configured, and a fifth tab — Chat — has appeared next to Settings. The key is encrypted at rest with the server's master key; the portal won't display it back.

Step 4: Have a conversation

  1. Click the Chat tab.
  2. The sidebar reads "No chats yet." Click New chat.

The bot greets you and asks for your name — that's the setup:onboard stage running.

Bot: Hi! I'm My first chatbot. What should I call you?

Reply with a name. The bot acknowledges and the chat moves on. Try a follow-up question — "What's the weather like?" will likely route into the fallback conversation, where the bot politely says it can't help with that. Both behaviours are defaults the wizard set up; you didn't write either prompt.

While the chat runs:

  • Stage toasts flash above the input when the conversation transitions stages.
  • The sidebar lists every chat. Each one belongs to you (per-user scoping is automatic — see Memory access for why).
  • Behind the scenes, every message is going through Hadron's Chat API; your portal session is acting as the integration.

Step 5: Confirm persistence

  1. Reload the page. Your chat is still in the sidebar.
  2. Click into it. The history is intact.
  3. Send another message. The bot still remembers your name — it was extracted into the chat's user memory in Step 4.

If you wanted to drive the same bot from your own application instead of the portal, the integration is four GraphQL calls (startChat, sendChatMessage, processChatResponse, and saveChatSummary). Building a chatbot agent covers the full design loop and shows the calls in context.

What just happened

You created an agent. The wizard wrote a small program for it in your system memory — five conversation and prompt nodes. You gave it an LLM key. The portal's Chat tab acts as the integration, calling Hadron's Chat API on every turn: startChat → sendChatMessage → processChatResponse. Hadron compiles the prompt, manages stage transitions, extracts data into memory, and remembers everything across reloads.

Nothing about this is a toy. The same Chat API powers production bots with multi-stage flows, conditional routing, goal stacks, and per-user memory isolation. You skipped the integration step by using the portal's Chat tab — you'll write the same calls when you ship it inside your own product.

Where to go next

A handful of next reads, depending on what you're doing: