Create your first Agent

Before you start

• A reachable control plane and a running Exo Operator (see Deploy & operate).
• The exo CLI installed. You'll sign in at step 4, when the work moves from the cluster to the platform.
• kubectl access to the cluster. Agents are authored by applying a manifest, so creating one is a cluster-side operation. Running and connecting afterwards go through the platform, no kubeconfig needed.

1. Set up the model provider

The Agent resolves its model credentials from a provider the platform holds, so configure that first. In the dashboard, go to Connect → Providers → Add provider: pick Anthropic (Claude) for the ClaudeCode runtime used here, paste the API key, and tick Set as default provider. Validate checks the key before anything is saved.

2. Create a Tool

A Tool is a managed connection (a database, SSH host, HTTP endpoint, or Kubernetes cluster). The credential lives in a Secret and never leaves the cluster. Save this as first-tool.yaml:

1apiVersion: v1
2kind: Secret
3metadata:
4  name: analytics-pg-creds
5  namespace: adaptive
6type: Opaque
7stringData:
8  password: REPLACE_ME
9---
10apiVersion: autonomic.sh/v1alpha1
11kind: Tool
12metadata:
13  name: analytics-postgres
14  namespace: adaptive
15spec:
16  type: postgres
17  description: "Analytics Postgres read-replica"
18  tags:
19    - databases
20    - postgres
21    - readonly
22  probe: true
23  config:
24    hostname: pg-analytics.prod.internal
25    port: 5432
26    username: autonomic_ro
27    databaseName: analytics
28    sslMode: require
29    credentialsRef:
30      name: analytics-pg-creds
31      key: password

Apply it and wait for the probe to pass. Ready means the reconciler reached the database with the credentials you gave it:

kubectl apply -f first-tool.yaml

kubectl -n adaptive get tools

3. Create the Agent

A minimal Agent declares a runtime, a goal, a model, the managed provider its credentials come from, and the Tool it may call. Save this as first-agent.yaml:

1apiVersion: autonomic.sh/v1alpha1
2kind: Agent
3metadata:
4  name: hello-agent
5  namespace: adaptive
6spec:
7  runtime: ClaudeCode
8  phase: Paused
9  goal:
10    inline: |
11      Connect to the analytics database and report the 5 most recent signups.
12  model:
13    name: claude-opus-4-7
14    provider: anthropic
15  provider:
16    type: claude
17  bootstrap:
18    skills:
19      - analytics-postgres

Apply it with kubectl:

kubectl apply -f first-agent.yaml

kubectl -n adaptive get agent hello-agent

4. Log in to the platform

Everything so far went through kubectl. Running the agent goes through the platform, so this is where you sign in. Already signed in from Getting started? Skip ahead.

On managed service the endpoint defaults to https://exo.adaptive.live, so just run exo login — no --endpoint needed:

exo login

exo whoami

To point the CLI at your own control plane, pass --endpoint:

exo login --endpoint https://exo.example.com

exo whoami

5. Start the first session

List your agents to confirm it's there, then fire a session and drop into it to watch and interact with it:

exo agent list

exo agent connect production/adaptive/hello-agent

Next steps

• Add more connections and engine-native permissions with the Tool CRD.
• Read the full Agent CRD reference for every field.