Multi-Agent & Multi-Connector

Grounded is built for situations where one tenant may have:

multiple agents
multiple connectors per agent
multiple snapshots per agent
multiple model or release variants flowing through traces

Use Agent Scope Deliberately

For direct traces, resource.agentId is the main routing key.

For control-plane access, agent-scoped secret keys are usually the safest choice for individual services.

Recommended Trace Metadata

For multi-model or multi-release setups, capture:

resource.agentId
resource.snapshotId
resource.environment
resource.release
tags for route or experiment
span inputs with model names when relevant

Example:

const trace = tracing.startTrace({
  name: 'support.reply',
  sessionId: 'chat_123',
  userId: 'user_123',
  tags: ['router:refunds', 'experiment:v2'],
  resource: {
    agentId: 'agt_support',
    snapshotId: 'snap_support_20260306',
    environment: 'production',
    release: 'api-2026-03-06',
  },
});

Multiple Connectors

One agent may also have multiple connectors, for example:

one llm-tracing connector for model traces
one session-data connector for surrounding product data

That lets Grounded correlate direct SDK traces with broader session context when both are available.

Operational Pattern

use a separate key per service
use a separate agent per major product surface
keep release tags stable
if you run multiple models, capture the model at the span level, not just in arbitrary logs

Retrieval Tracing Best Practices