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

# ADK Message Mapping

> How Aion Server adapts A2A requests to Google ADK and maps ADK output back into A2A.

This page describes how Aion Server adapts A2A requests to Google ADK and maps ADK events back into
A2A Messages, Tasks, and streaming events.

This page uses the v1 canonical JSON-RPC method names. Aion ingress may continue to accept legacy
slash-style aliases for compatibility.

## 1. Inbound Messages

### 1.1 Agent Invocation

Both `SendMessage` and `SendStreamingMessage` use the same execution path: the agent's
`run_async()` is always driven as an async event stream.

* `SendMessage` (`blocking=true`) — collects all events and returns the final `Task`.
* `SendMessage` (`blocking=false`) — returns after the first event, continues processing in
  background with `status="working"`.
* `SendStreamingMessage` — yields events as they arrive and streams them to the client via SSE.

Both methods use the same `SendMessageRequest` payload; only the response mode differs.

### 1.2 Part Type Mapping

Inbound A2A message parts are transformed into ADK `Content` as follows:

| A2A Part         | ADK Representation                                                  |
| ---------------- | ------------------------------------------------------------------- |
| `Part(text=...)` | `types.Part(text=...)`                                              |
| `Part(raw=...)`  | `types.Part(inline_data=types.Blob(mime_type=..., data=...))`       |
| `Part(url=...)`  | `types.Part(file_data=types.FileData(mime_type=..., file_uri=...))` |
| `Part(data=...)` | `types.Part(text=json.dumps(data))`                                 |

MIME type resolution order for file parts: explicit `mime_type` attribute → guess from filename →
fallback to `application/octet-stream`.

If the message contains no usable parts, the plain text input from the request is used as a fallback.

### 1.3 Accessing Inbound Context — `ctx.a2a_inbox`

When an inbound A2A `Message` arrives, Aion Server makes it available through `ctx.a2a_inbox` on
the invocation context:

```python theme={null}
from google.adk.agents import BaseAgent


class MyAgent(BaseAgent):
    async def _run_async_impl(self, ctx):
        inbox = ctx.a2a_inbox
        task = inbox.task
        message = inbox.message
        metadata = inbox.metadata
```

`a2a_inbox` contains:

| Field      | Type      | Description                                                            |
| ---------- | --------- | ---------------------------------------------------------------------- |
| `task`     | `Task`    | The current A2A Task                                                   |
| `message`  | `Message` | The full inbound A2A Message, including non-text parts                 |
| `metadata` | `dict`    | `SendMessageRequest`-level metadata (distribution/network, trace info) |

## 2. Outbound Messages

Valid responses to an A2A `SendMessage` call are a `Message` or a `Task`.

Aion Server constructs the response using the following precedence:

### (1) SDK-managed response buffer (authoritative when populated)

The runtime maintains a request-scoped messaging buffer for the current turn.
SDK helpers and ordinary ADK event content may populate that buffer, including
partial stream output and final non-partial message content that is intended
to become the durable reply.

When this buffer is non-empty, it is the authoritative source for A2A
response compilation.

### (2) `a2a_outbox`

Set `a2a_outbox` in `event.actions.state_delta` to provide an explicit A2A response. It must be an
`A2AOutbox` instance wrapping either a `Message` or a `Task`:

```python theme={null}
from a2a.types import Message, Task, Part, Role
from aion.core.a2a import A2AOutbox
from google.adk.agents import BaseAgent
from google.adk.events import Event, EventActions


class MyAgent(BaseAgent):
    async def _run_async_impl(self, ctx):
        # Option 1: outbox as Message
        yield Event(
            author=self.name,
            actions=EventActions(state_delta={
                "a2a_outbox": A2AOutbox(message=Message(
                    role=Role.ROLE_AGENT,
                    parts=[Part(text="Done!")],
                ))
            })
        )

        # Option 2: outbox as Task (patch)
        yield Event(
            author=self.name,
            actions=EventActions(state_delta={
                "a2a_outbox": A2AOutbox(task=Task(
                    history=[...],
                    artifacts=[...],
                    metadata={"my_key": "my_value"},
                ))
            })
        )
```

Server-owned fields are enforced:

* `task_id` and `context_id` are set to current values managed by Aion Server.
* Canonical routing and identity metadata (e.g. `aion:network`, sender IDs) is server-controlled.

Behavior:

* If `a2a_outbox.message` is set → append to current Task history.
* If `a2a_outbox.task` is set → treat as a **patch** to the server's Task:
  server merges or extends `history` and `artifacts`; provided `metadata` merges shallowly;
  server-controlled keys take precedence.

### (3) Framework-native fallback

If neither the SDK-managed response buffer nor `a2a_outbox` is populated,
Aion Server falls back to framework-native output for the current turn:

* first, accumulated partial stream text
* then, if needed, the final non-partial agent-authored event content
* finally, deterministic final session/state inspection when the adapter
  exposes enough data to do so safely

> If you need to return a comprehensive A2A response (e.g., data parts, rich metadata, multiple
> artifacts), use `a2a_outbox` rather than relying on the streaming fallback.

## 3. Summary

* Read `ctx.a2a_inbox` to access the inbound A2A Task, Message, and metadata.
* Prefer SDK helpers or normal ADK event content when you want to populate the
  shared runtime response buffer.
* Optionally set `a2a_outbox` in `event.actions.state_delta` as an `A2AOutbox` instance for full-fidelity A2A responses.
* Yield partial events for real-time text streaming.