Deploy Agents

Documentation Index

Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

To make an Agentic project live and available to users, you must correctly configure deployments, environments, channels, API keys, and channel authentication. Before users can interact with your agents, the following must exist:

• An active deployment in the target environment (dev, staging, or production).
• Environment variables configured for all {{env.KEY}} references used in ABL.
• A channel bound to that environment and an API key to authorize.

​

Deployment Lifecycle

• A deployment is always scoped to a single environment.
• Only one deployment per environment can be active at a time.
• When a new deployment becomes active:
  • The previous deployment automatically moves to draining.
  • Existing sessions continue naturally.
  • New sessions route to the latest deployment.

​

Step 1: Set Environment Variables

Before deployment, configure all environment variables referenced in agents. Navigate to Deployments -> Environments. Base Variables : The Base variables defines fallback values shared across environments. Use base variables for:

• Shared defaults across deployments.
• Product names.
• Common timeouts.

Override them per environment when needed. Example: To add a new env variable:

• Select the target environment tab (dev, staging, or production).
• Click Add Variable.
• Enter the key, value, and optional description.
• Toggle Secret if the value is sensitive (secrets are write-only after saving).
• Click Save.
• You can also copy variables from other environments.

​

Step 2: Deploy

Go to Deployments → Environments and click New Deploy (upper-right). Configure:

• Environment: Select environment. Always start with dev.
• Entry agent: The agent that receives the first message, usually the supervisor.
• Agent version manifest: map each agent to a version, or use the latest active version. You can also auto-create a version from the working copy.
• Label: Human-readable identifier for the deployment like v1.0 - initial build.

Click Deploy. The platform validates your agent ABLs, resolves configuration placeholders, and activates the deployment. If there was a previous dev deployment, it moves to draining — existing sessions finish naturally, new ones go to the new deployment.

​

Step 3: Promote to staging

Don’t manually recreate staging deployments. Instead, Promote your dev deployment. This pushes the exact same agent version manifest to staging, so you’re testing what you actually built rather than manually re-specifying versions.

​

Step 4: Promote staging to production

Once staging is validated, promote it to production the same way, by clicking Promote on the staging deployment.

​

Step 5: Configure Access

An active deployment isn’t automatically reachable. You must configure:

• API keys
• Channels
• Authentication methods, if required

​

Configure API Keys

For external application access, create API Keys. Learn More

​

Channels

Channels expose deployments to end users. Examples:

• Web SDK
• Slack
• WhatsApp
• Voice channels
• Mobile SDKs

Navigate to Deployments -> Channels.

• Select the target environment or deployment. A channel can either:
  • Follow an environment automatically
  • Or pin to a specific deployment version
• Configure channel-specific parameters.
• Configure API keys.

The Web SDK is the most common starting point. Once configured, it provides a <script> tag that you embed in your website. Depending on the authentication type, insert your SDK key (pk_*) or server secret (sk_*).

​

Step 6: Rollback if something goes wrong

If a production deployment causes issues, go to Deployments → Environments, find the active deployment, and click Rollback. This retires the current deployment and reactivates the previous one in that environment. Users on the bad deployment finish their sessions naturally; new sessions immediately go to the restored version.

​

Deployment status reference