DINODOCSPUBLIC

Appearance

Documentv1.0.0

Updated

Last verified

Simulate a conversation ​

Connect two of your agents on one Runtime and read a separate Lens report for each side. Use a simulation to prove a full, multi-turn conversation without telephony.

You run dino simulation run with firstSpeaker a. The Runtime starts one call for Agent A and one for Agent B on a single session, each with automatic VAD and no transport-VAD profile. Agent A plays the greeting, Agent B pads its input and replies, and the agents alternate turns. When the flow ends, the Runtime closes the peer call, uploads both archives to Lens, and dino simulation get returns the initiator and responder Lens URLs.You run dino simulation run with firstSpeaker a. The Runtime starts one call for Agent A and one for Agent B on a single session, each with automatic VAD and no transport-VAD profile. Agent A plays the greeting, Agent B pads its input and replies, and the agents alternate turns. When the flow ends, the Runtime closes the peer call, uploads both archives to Lens, and dino simulation get returns the initiator and responder Lens URLs.

PlantUML source

Requirements ​

A simulation is Pong-tier and billable. It needs:

  • A Runtime and partner credentials for the account.
  • Gemini (AI Studio) credentials for both agents.
  • Two free slots in the account Widget concurrency limit.

Both agents must be available for the account. Simulations use automatic VAD for both sessions. They have no transport-VAD profile.

Configure the pair ​

Write a required configuration file. Set firstSpeaker to a (initiator) or b (responder):

json
{"firstSpeaker":"a"}

Each side keeps its own persona and first-stage instruction. Only the first speaker plays its greeting. The other side hears the greeting, pads its own input, and replies.

Run and inspect ​

Prompt your AI agent

Recommended
text
Use the `dino-pong` skill to run a two-agent simulation.
Initiator agent: <agent>. Responder agent: <agent>.
Verify both agents are available and I have a Runtime, partner credentials, and concurrency.
Write the firstSpeaker config and show every argument and the billing effect.
Wait for my approval, run once, then return both Lens links and outcomes.
Use the Dino CLI directly
text
dino simulation run \
  --simulation-config <absolute-simulation.json> \
  --runtime-id <runtime-id> \
  --hard-close-secs 90 \
  --initiator-agent-id <agent-id> \
  --initiator-partner-id <partner-credential-id> \
  --responder-agent-id <agent-id> \
  --responder-partner-id <partner-credential-id> \
  --confirm-mutation simulation.run

dino simulation get --id <simulation-id>
dino simulation list --limit 25

Supply per-side placeholders and integration context with absolute JSON paths: --initiator-placeholders/--responder-placeholders and --initiator-plugin-contexts/--responder-plugin-contexts. Transcription records both sides by default. Use --transcription-mode in or off to reduce it.

Read the result ​

simulation get returns status, mode, and an initiator and responder object. Each side has callId, archiveReady, lensUrl, and report.

StatusRead
completedBoth archives are ready. Open each Lens link and read each report.
failedfailure.code and failure.participant (a or b).

simulation list returns newest first. Each item has the ID, status, mode, both call IDs, and initiatorLensUrl/responderLensUrl. A Lens link is null until that archive is ready.

Gotchas ​

  • hard-close-secs (90–900) is the fallback limit. There is no separate simulation duration.
  • Each Lens link appears only after that side's archive reaches Lens.
  • An available Lens link does not clear an earlier failure.
  • Do not start a replacement run automatically. Each run is billable.

Next ​

Use Test your agent for the single-call preview path.

Dinodial customer documentation