Quickstart

Use this path to create, publish, and test one small agent.

Before you start

You need an installed dino CLI, a Dino API key, and an empty workspace directory. Keep the API key in the workspace .env file. Never paste it into chat.

1. Prepare the workspace

Create and register a separate workspace for each deployment.

Prompt your AI agent

Recommended

Use the `dino-ping` skill to prepare a workspace for <deployment>.
Keep the API key out of chat and logs.
Show each local or registration change before you make it.
Ask for approval for one change at a time.
Stop when the workspace and account checks pass.

Use the Dino CLI directly

# After you approve the local workspace write:
dino --workspace <absolute-workspace> workspace init \
  --confirm-mutation workspace.init

# Add VOX_API_KEY to <absolute-workspace>/.env without printing it.

# After you approve the registration change:
dino workspace use <name> --path <absolute-workspace> \
  --confirm-mutation workspace.use

dino workspace check
dino account get

2. Build from the nearest example

Review the capability ladder. Select the smallest example that covers the required behavior. Open its source tree through the installed dino-ping skill.

Prompt your AI agent

Recommended

Use the `dino-ping` skill to build a voice agent for <use case>.
Inspect all source-complete Dino examples first.
Choose the smallest example that matches my required behavior.
Tell me which example you selected and why.
Ask only for missing business decisions.
Adapt a copy in my workspace and validate after each material edit.
Do not publish or place a call.

Use the Dino CLI directly

# Use scaffold only when no example is a useful base.
dino agent scaffold --name <agent-name> \
  --confirm-mutation agent.scaffold

dino agent validate --agent-dir <absolute-agent-dir>

3. Review and publish

Render the graph before publication. Publish only a valid agent.

Prompt your AI agent

Recommended

Use the `dino-ping` skill to prepare this agent for first publication.
Validate the full agent and render its complete flow.
Summarize the agent, its tools, and each terminal outcome.
If any check fails, fix the source and validate again.
Show the selected workspace and publication effect.
Wait for my approval before publishing once.
Return the genesis ID, version ID, and idempotency key.

Use the Dino CLI directly

dino agent validate --agent-dir <absolute-agent-dir>

# After you approve the preview file write:
dino agent render --agent-dir <absolute-agent-dir> \
  --output <absolute-agent-dir>/preview.png \
  --confirm-mutation agent.render

# After you approve the selected workspace and publication:
dino agent publish --agent-dir <absolute-agent-dir> \
  --confirm-mutation agent.publish

4. Place one preview call

An Account Manager configures the fixed destination and Preview allocations. The Client selects one ready allocation of each type and places the call.

Prompt your AI agent

Recommended

Use the `dino-ping` skill to test my published agent with one preview call.
Verify my role, workspace, agent ownership, and allowed destination rule.
Show the exact destination and call effect before the call.
Wait for my approval immediately before placing it.
Monitor only that preview and return its terminal status and available results.

Use the Dino CLI directly

dino account preview-allocations

dino agent preview-call \
  --agent-id <agent-id> \
  --vad-mode <vad-mode> \
  --vad-profile-id <silero-profile-id> \
  --allocation-id <allocation-id> \
  --preview-did-id <did-id> \
  --preview-partner-credential-id <partner-credential-id> \
  --preview-vox-server-id <vox-server-id> \
  --confirm-mutation agent.previewCall

dino agent preview-status --dd-id <dd-id>

Gotchas

• Do not edit an installed example in place. Copy it into your workspace.
• Generate fresh integration IDs before the first publish of a copied example.
• Do not publish around a validation failure.
• Preserve a publication idempotency key when you must retry the same request.
• Do not start a second preview while one interactive session is active.

Next

Read Create an agent for the design brief and Test your agent for stronger evidence.