Deploy a production agent from any model with a single command. This guide covers the full config schema, the deploy step itself, staging vs. production environments, and how to verify your agent is live and behaving correctly.

Prerequisites

• A Sublayr API key

• The CLI or SDK installed

• At least one connected tool (optional, but typical for most agents)

The full config schema

An agent config is a YAML file describing everything about how the agent should run:

Example: a full config

name: refund-agent
model: anthropic/claude
fallback: openai/gpt-4o
tools:
  - stripe
  - postgres
approvals:
  - action: refund
    condition: "amount > 100"
    approver: finance-team
budget: 300/mo
environment: staging

name: refund-agent
model: anthropic/claude
fallback: openai/gpt-4o
tools:
  - stripe
  - postgres
approvals:
  - action: refund
    condition: "amount > 100"
    approver: finance-team
budget: 300/mo
environment: staging

name: refund-agent
model: anthropic/claude
fallback: openai/gpt-4o
tools:
  - stripe
  - postgres
approvals:
  - action: refund
    condition: "amount > 100"
    approver: finance-team
budget: 300/mo
environment: staging

Deploying to staging first

We strongly recommend deploying to staging before production. Staging uses the same runtime and observability as production, but is isolated from real customer data and doesn't count against your production run quota.

sublayr deploy --config agent.yaml --env staging

sublayr deploy --config agent.yaml --env staging

sublayr deploy --config agent.yaml --env staging

Test the agent thoroughly in staging — trigger its typical actions, confirm approvals route correctly, and check traces for anything unexpected.

Promoting to production

Once you're confident, promote the exact same config to production:

sublayr deploy --config agent.yaml --env production

sublayr deploy --config agent.yaml --env production

sublayr deploy --config agent.yaml --env production

This does not require rewriting anything — the same build that ran in staging is promoted as-is, so there's no risk of environment drift between what you tested and what goes live.

Verifying deployment

Check the agent's status:

sublayr agents status agt_4b81

sublayr agents status agt_4b81

sublayr agents status agt_4b81

This returns current status (deploying, live, paused, or failed), uptime, the model and fallback in use, and a summary of the last 10 runs.

Common deployment errors

Next steps

Once your agent is live, set up traces and alerts under Observability, and configure a rollback plan under Versioning.