Connect an agent and send your first message

By the end of this tutorial you’ll have a live agent connected to Alfe from your own machine, verified from both the CLI and the Agent API, and you’ll have sent it a message. It takes a few minutes.

Before you start

You need these before step 1:

• Node.js and npm.
• An Alfe account. Sign up at app.alfe.ai. Agents can also self-onboard over MCP; see Agent-led bootstrap.

1. Install the CLI

npm

npm i @alfe.ai/cli -- -g

pnpm

pnpm add @alfe.ai/cli -g

bun

bun add @alfe.ai/cli -g

Then confirm it’s on your PATH:

Terminal

alfe --version

2. Create the agent

Create the agent record in the dashboard, so the machine has an identity to connect to:

1. Open Agents at app.alfe.ai and create an agent.
2. On Hosting, select Run it yourself and choose a runtime. OpenClaw is the default.
3. Finish the wizard and open the agent’s Overview tab.
4. Under Get started locally, select Generate setup token and copy the token. It’s shown once.

3. Connect the agent

alfe setup connects the agent on this machine. It installs the agent runtime, starts the local gateway daemon, and registers the Alfe platform tools so the agent can reach memory, model routing, and integrations.

Terminal

export ALFE_API_KEY=<SETUP_TOKEN>
alfe setup

When it finishes, the daemon is running and the agent is connected. Confirm who you are, then check the agent’s health:

Terminal

alfe whoami
alfe doctor

On a server?

For unattended, cloud-init-style provisioning, use alfe setup --managed, which reads everything from environment variables instead of prompting.

4. Confirm over the Agent API

Your setup token is the agent’s API key, and it’s the value of ALFE_API_KEY from step 3. Ask the API who it is:

Terminal

curl https://api.alfe.ai/agent/identity/whoami \
  -H "Authorization: Bearer $ALFE_API_KEY"

{ "data": { "agentId": "agt_…", "tenantId": "…" } }

Seeing your agentId confirms the agent is authenticated and reachable. See Authentication for the full token model.

5. Say hello

Open the agent’s Chat tab in the dashboard and send it a message. To talk to it where you already work, connect a channel like Slack or Microsoft Teams from Integrations — each agent gets its own bot identity on the channel.

Next steps

• Give your agent memory — store and recall facts over the API.
• Receive a webhook — have your agent react to external events.
• Explore the full Agent API reference.