# State Streaming

> Stream partial agent state updates to the UI while a tool call is still running.

<!-- interactive demo: shared-state-streaming -->

## What is this? [#what-is-this]

By default, agent state only updates *between* backend checkpoints, so
a long-running tool call (writing a full document, drafting an email)
appears to the UI as one big burst at the end. For agent-native apps,
that feels broken: users expect to watch the output materialise.

**State streaming** copies one tool argument into an agent state key
while the model writes the argument.
The UI, subscribed via `useAgent`, re-renders every token.

## When should I use this? [#when-should-i-use-this]

Use state streaming whenever a tool's output is long-form text or a
growing structured value and you want the user to see it assemble in
real time. Common shapes:

- A collaborative writing agent that emits a document
- A research agent that accumulates a list of findings
- A planning agent that builds up a step-by-step plan

## The backend: one streaming state mapping [#the-backend-one-streaming-state-mapping]

<Steps>
  <Step>
    ### Map the tool argument with `predict_state_config`

    `predict_state_config` names the tool and the argument to stream into a
    state key. While the model generates the `write_document` call, the
    adapter sends each partial `document` argument to the UI as a state
    update. Returning `state_update(...)` from the tool commits the final
    value when the call completes:

    ```python title="src/agents/shared_state_streaming.py"
    from typing import Annotated

    from agent_framework import tool
    from agent_framework_ag_ui import state_update
    from pydantic import Field

    # Stream tool-argument deltas straight into `state.document` while
    # `write_document` is still streaming.
    PREDICT_STATE_CONFIG: dict[str, dict[str, str]] = {
        "document": {
            "tool": "write_document",
            "tool_argument": "document",
        }
    }


    @tool(
        name="write_document",
        description=(
            "Write a document for the user. Always call this tool when the "
            "user asks you to write, draft, or revise any text. The "
            "`document` argument is streamed per-token into shared state "
            "under the `document` key so the UI renders the body live."
        ),
    )
    def write_document(
        document: Annotated[
            str,
            Field(description="The full document content as a single string."),
        ],
    ):
        return state_update(
            text="Document written to shared state.",
            state={"document": document},
        )
    ```

  </Step>
  <Step>
    ### Seed the state key before the deltas arrive

    The streamed updates are JSON Patch `replace` operations on
    `/document`. The agent's initial state is `{}`, and a `state_schema`
    defines properties but sets no starting values, so on the first turn the
    path does not exist yet. The
    frontend rejects the patch ("Failed to apply state patch") and the run
    never settles.

    Emit a `STATE_SNAPSHOT` that carries the key right after `RUN_STARTED`,
    keeping any value from an earlier turn:

    ```python title="src/agents/shared_state_streaming.py"
    from collections.abc import AsyncGenerator
    from typing import Any

    from ag_ui.core import BaseEvent, EventType, StateSnapshotEvent
    from agent_framework import Agent, BaseChatClient
    from agent_framework_ag_ui import AgentFrameworkAgent


    class SharedStateStreamingFrameworkAgent(AgentFrameworkAgent):
        async def run(  # type: ignore[override]
            self,
            input_data: dict[str, Any],
        ) -> AsyncGenerator[BaseEvent, None]:
            incoming = input_data.get("state")
            state = incoming if isinstance(incoming, dict) else {}
            document = state.get("document", "")
            seeded = False
            async for event in super().run(input_data):
                yield event
                # RUN_STARTED must be the first event; the seed follows it.
                if not seeded and getattr(event, "type", None) == EventType.RUN_STARTED:
                    yield StateSnapshotEvent(
                        type=EventType.STATE_SNAPSHOT,
                        snapshot={**state, "document": document},
                    )
                    seeded = True


    def create_shared_state_streaming_agent(
        chat_client: BaseChatClient,
    ) -> AgentFrameworkAgent:
        base_agent = Agent(
            client=chat_client,
            name="shared_state_streaming_agent",
            instructions=SYSTEM_PROMPT,  # tells the model to call write_document
            tools=[write_document],
        )

        return SharedStateStreamingFrameworkAgent(
            agent=base_agent,
            name="SharedStateStreamingAgent",
            description=(
                "Per-token state streaming: `write_document` arg deltas land "
                "in `state.document` as the tool call is generated."
            ),
            predict_state_config=PREDICT_STATE_CONFIG,
            require_confirmation=False,
        )
    ```

  </Step>
</Steps>

The backend pattern is always the same: map one streaming tool argument
to one shared-state key. Each framework does this in its own way:

- LangGraph Python: `StateStreamingMiddleware` with `StateItem(...)` entries.
- LangGraph TypeScript: `copilotkitCustomizeConfig` with an
  `emitIntermediateState` mapping.
- Direct SDK adapters: parse partial tool arguments in the streaming loop,
  and emit `STATE_SNAPSHOT` whenever the mapped value changes.
- LlamaIndex and Pydantic AI: there is no per-token mapping. A backend tool
  sends a state snapshot after each step or tool call instead (see the setup
  above).

When the LLM streams that argument, CopilotKit writes every partial value
into shared state before the tool even finishes executing.

```python title="src/agents/shared_state_streaming.py"
# Tells the runtime to stream tool-argument deltas straight into
# `state.document` while `write_document` is still streaming — matches
# LGP's StateStreamingMiddleware setup.
PREDICT_STATE_CONFIG: dict[str, dict[str, str]] = {
    "document": {
        "tool": "write_document",
        "tool_argument": "document",
    }
}
```

A few things to note:

- The state key must exist in your agent state (`document` in this demo).
- The tool and argument names must match the exact LLM-facing tool call
  you want to forward (`write_document.document` here).
- Streamed values are previews. When the step ends, the state the agent
  saves replaces them, so save the final value when the tool finishes. On
  CrewAI, LlamaIndex and Pydantic AI, write the final value to the agent
  state yourself, or it is overwritten when the step returns.

## The frontend: useAgent + OnStateChanged [#the-frontend-useagent--onstatechanged]

The UI side is the same as any other [shared-state](/ms-agent-python/shared-state) subscription:
`useAgent` with `OnStateChanged` gives you a reactive `agent.state`.
Add `OnRunStatusChanged` if you want a "LIVE" / "done" indicator.

```typescript title="src/app/demos/shared-state-streaming/page.tsx"
  // Subscribe to BOTH state changes and run-status changes. The former
  // drives the per-token document rerender; the latter toggles the
  // "LIVE" badge when the agent starts / stops.
  const { agent } = useAgent({
    agentId: "shared-state-streaming",
    updates: [UseAgentUpdate.OnStateChanged, UseAgentUpdate.OnRunStatusChanged],
  });
```

From there, `agent.state.document` is just a string that grows on every
token, and `agent.isRunning` tells you whether to show a streaming
indicator.

## Try it [#try-it]

Ask the agent to write a short poem. The `document` text grows in your UI
while the agent writes. When the run ends, `agent.isRunning` is `false`, so
your streaming indicator turns off.

<IntegrationGrid path="shared-state/streaming" exclude={["agno", "agent-spec", "spring-ai", "langroid"]} />
