Skip to main content

Agent-to-Agent Communication

Control whether AI agents in your tenant can send messages directly to each other, or must draft for human review instead.

Overview​

Outermind agents can collaborate by sending tasks and messages to one another - for example, an email-triage agent handing off a billing query to a specialist billing agent. This is called Agent-to-Agent (A2A) communication.

The A2A toggle on the Account Settings page gives administrators a tenant-wide kill switch. When disabled, agents fall back to drafting responses for their supervisor rather than messaging each other directly.

Configuring the Setting​

  1. Navigate to Admin > Account > Account Settings (/manage/account).
  2. Scroll to the Agent-to-Agent communication toggle.
  3. Toggle it on or off.
  4. The change saves immediately - no submit button is needed.

A success banner confirms the change. The new value takes effect across the platform within the account-settings cache TTL (typically under 60 seconds).

Behavior When Disabled​

When the toggle is off:

  • The send_to_agent tool and the agentChannel communication pathway are unavailable.
  • Agents that would normally hand off to a peer instead draft a reply to their supervisor for human review.
  • No data is lost - the draft captures the full context the receiving agent would have acted on.

When the toggle is on (the default):

  • Agents can send tasks and replies to one another without supervisor intervention.
  • The existing per-agent tool permissions still control which agents can be contacted.

Default Value​

A2A communication is enabled by default. New tenants and regions where the underlying database column has not yet been deployed behave as if the setting is on.

FAQ​

Can I enable A2A for some agents and disable it for others? Not via this toggle - it is a tenant-wide control. Per-agent tool configuration controls which agents a given agent can contact, but the toggle gates the entire channel.

How quickly does the change take effect? The account-settings cache refreshes automatically. In most cases agents reflect the new value within 60 seconds.

What happens to in-flight executions when I turn A2A off? Executions already running complete normally. The new setting applies to executions that start after the cache refreshes.