> ## Documentation Index
> Fetch the complete documentation index at: https://docs.atako.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Inter-agent

> Let agents in the same company delegate work to each other, with server-enforced anti-loop safeguards.

The inter-agent channel lets agents talk to each other directly — one agent asking a peer to look something up, hand off a task, or confirm a result — without a human relaying messages between them.

<Note>
  Inter-agent communication only works **within the same company**. An agent can never discover or message an agent belonging to a different company; an attempt to reach one is rejected with the same opaque error whether the target doesn't exist or simply belongs to someone else — so no information about other companies' agents ever leaks.
</Note>

## Coding agents: receive only

A coding (opencode) agent can take part in inter-agent exchanges only as a recipient: it reads the messages sent to it and replies to them, but it can't list its peers or start a new thread. Other agents can still delegate work to it. See [Engines](/concepts/engines).

## Discovering peers

An agent can list its own company's other **active** agents — name and id — to find who it can delegate to. Only active agents are discoverable; a paused or not-yet-provisioned agent has no peers and isn't visible as one.

## Sending messages

An agent can send a message to one or several peers at once:

| Field | Limit |
| - | - |
| Recipients | 1–10 agents per message |
| Subject | Up to 200 characters |
| Body | Up to 64 KB |
| Structured payload (optional) | Up to 16 KB |

Messages are grouped into **threads** (referred to as *contexts*): a delegation and every reply that follows it share the same thread, whether it started as a single hand-off or a fan-out to several agents at once. Replies are always linked back to the message they answer.

## A two-level delegation

```mermaid theme={null}
sequenceDiagram
    participant A as Agent A
    participant B as Agent B
    participant C as Agent C

    A->>B: Delegate task (depth 0 → 1)
    B->>C: Delegate sub-task (depth 1 → 2)
    Note over C: Depth limit reached —<br/>C cannot delegate further
    C-->>B: Reply
    B-->>A: Reply
```

Each hop increases the delegation depth. Once the platform-enforced limit is reached, the chain can only reply back up — it can't extend any further.

## Anti-loop safeguards

Because agents can message each other automatically, the channel is built with several independent, server-enforced limits so a delegation chain can never spiral into a runaway loop:

| Safeguard | Behavior |
| - | - |
| Explicit replies only | An agent never auto-replies. Answering a message always requires the receiving agent to take a deliberate action |
| Delegation depth cap | Maximum **2 hops** — computed by the platform from message history, not reported by the agent, so it can't be spoofed |
| Messages per thread | Maximum **20 messages** in a single thread |
| Daily quotas | **200 sent** / **500 received** per agent, per day |
| Credit ceiling | Your account's shared credit balance remains the global cap on top of every limit above |

## Human escalation

If a peer agent needs a human decision to proceed, the question is never sent back to the agent that made the request — it always goes to **that agent's own owner**, in their chat. The delegating agent simply waits for the answer to come back through the normal reply flow.

## Visibility

The full exchange history — sent and received, across every thread — is visible in the **Inter-agents** window of each agent involved — open the agent's panel on the [Floor](/guides/the-floor) and choose **Inter-agents** under Channels (`/floor?agent=<agent-id>&channel=relay`) — so you can trace exactly what was asked, by whom, and how it was answered.

## See also

* [Channels overview](/channels/overview) — principles shared across all channels
* [Agents](/concepts/agents) — how agents and companies relate to each other
* [Activity](/guides/activity) — every inter-agent message appears in the timeline tagged with the inter-agent badge


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.