Documentation / Agents / Interfaces

Operate Agents with the CLI

Deploy, list, inspect, and invoke Agents through stable CLI commands backed by the public Platform API.

Before you begin

Configure the CLI namespace and authenticate with deployment and Agent read/invoke authority. Build a package containing one Agent manifest, exact SDK dependency, entrypoint, and imported source. Keep secrets and local build output outside the archive.

Steps

  1. Upload the immutable package. The CLI derives a stable deployment idempotency key from the archive SHA-256:
sh
constal deployments create support-agent.zip --output json
  1. Poll the returned deployment with constal deployments get <deployment-id> until publication succeeds.
  2. List and inspect the deployed Agent:
sh
constal agents list
constal agents get support --output json
  1. Start a controlled Run through the Runs CLI guide. Deployment and invocation remain separate actions so release automation cannot accidentally execute code.

The CLI uses the same /v1/deployments and namespace Agent endpoints documented for HTTP clients. It does not build a mutable local alias or silently select another tenant.

Verify

Confirm the deployment reports an immutable revision, executable artifact, digest, and successful probe. Inspect the Agent and compare its version, mode, logical bindings, Policies, Tool catalog, and limits with the package manifest. Start one low-risk Run and verify it pins the new deployment revision.

Next steps

Use Build Agents with the SDK for code, Deploy an Agent for package semantics, and CLI command reference for exact arguments.