Agent-to-Agent Communication

Agent-to-agent communication allows one agent to ask questions to another agent as if it were a tool or data source. This enables cross-domain investigation without requiring a single agent to have access to every data source in your stack.

Use Case

Consider a scenario where a Debugger agent is investigating a spike in API errors. While reviewing logs, it notices the errors started at a specific time. By having an Analyst agent connected, the Debugger can ask the Analyst whether any significant changes in user traffic or feature usage occurred around that same time. The Analyst queries your analytics platform and returns its findings, which the Debugger incorporates into its root cause analysis.

This pattern is useful whenever an investigation benefits from data that lives outside the current agent’s domain — metrics, logs, code, or database records managed by a different agent.

Setting Up Connected Agents

1. Navigate to the agent you want to give delegation capabilities (the “calling agent”).
2. Open the Settings tab on its settings page.
3. In the Connected Agents section, toggle on the agents you want to make available as data sources.

Once enabled, the calling agent can send questions to any of its connected agents during a conversation.

How It Works

When a calling agent decides it needs information from a connected agent, it:

1. Formulates a self-contained question that includes all necessary context. The connected agent does not have access to the calling agent’s conversation history.
2. Sends the question to the connected agent.
3. The connected agent processes the question using its own connections and knowledge — querying its data sources, reasoning through the data, and generating a response.
4. The calling agent receives the response and incorporates it into its own reasoning.

Because each agent operates independently with its own connections, this approach keeps data access scoped appropriately. An Analyst agent only accesses analytics data, and a Debugger agent only accesses logs and code, even when they collaborate.

Constraints

Agent-to-agent communication has the following constraints to prevent loops and maintain predictable behavior:

• No recursion. When a connected agent processes a delegated question, it does not have access to its own connected agents. Only the top-level calling agent can delegate.
• No self-reference. An agent cannot be connected to itself.
• No direct cycles. If Agent A is connected to Agent B, then Agent B cannot also be connected to Agent A.
• Maximum of 10 connected agents. Each agent can have at most 10 other agents connected to it.
• Same user only. All connected agents must belong to the same user account.