`update_state` on fresh `DeepAgentState` thread drops first `messages` write

[ic-zl]

Submission checklist

• This is a bug, not a usage question.
• I added a clear and descriptive title.
• I searched existing issues and didn't find this.
• I can reproduce this with the latest released version.
• I included a minimal reproducible example and steps to reproduce.

Area (Required)

• deepagents (SDK)
• cli
• code
• acp
• evals
• harbor
• daytona
• modal
• quickjs
• runloop
• langsmith-sandbox
• Other / not sure / general

Related Issues / PRs

No response

Reproduction Steps / Example Code (Python)

import asyncio

from deepagents.graph import DeepAgentState
from langchain_core.messages import HumanMessage
from langgraph.checkpoint.memory import InMemorySaver
from langgraph.graph import StateGraph


async def model(state: DeepAgentState) -> dict:
    return {}


async def main() -> None:
    graph = StateGraph(DeepAgentState)
    graph.add_node("model", model)
    graph.set_entry_point("model")
    graph.set_finish_point("model")

    agent = graph.compile(checkpointer=InMemorySaver())

    config = {"configurable": {"thread_id": "t"}}
    message = HumanMessage(content="hello", id="m1")

    await agent.aupdate_state(
        config,
        {"messages": [message]},
        as_node="model",
    )

    snapshot = await agent.aget_state(config)
    print("snapshot.values:", snapshot.values)


asyncio.run(main())

Error Message and Stack Trace (if applicable)

snapshot.values: {'messages': []}

Description

I am trying to write a message into a fresh DeepAgentState thread using LangGraph's aupdate_state.

Expected behavior:

await agent.aupdate_state(
    config,
    {"messages": [HumanMessage(content="hello", id="m1")]},
    as_node="model",
)

snapshot = await agent.aget_state(config)
assert snapshot.values["messages"]

Actual behavior:

aupdate_state creates a checkpoint, but reading the state back returns:

{"messages": []}

Control cases:

• agent.ainvoke({"messages": [message]}, config) persists the same message correctly.
• A normal MessagesState graph with aupdate_state({"messages": [message]}) persists correctly.
• If a DeepAgentState thread is first seeded through normal ainvoke, later aupdate_state appends appear to work.

This seems specific to the first out-of-band write into DeepAgentState.messages, which is backed by DeltaChannel.

Possible cause:

In the aupdate_state path, channel writes appear to be persisted only when a previous checkpoint exists:

if saved and channel_writes:
    await checkpointer.aput_writes(checkpoint_config, channel_writes, task_id)

On a fresh thread, saved is None, so the first messages write is applied to the live channel but not persisted as checkpoint writes. Then create_checkpoint(...) omits non-snapshot DeltaChannel values from channel_values, leaving a checkpoint with a bumped messages channel version but no readable value or replayable writes.

Question:

Is update_state / aupdate_state expected to support writing to DeepAgentState.messages on a fresh thread? If not, is there a recommended API for appending a message to a DeepAgents thread without invoking the model?

Environment / System Info

OS: macOS 26.5.1 (25F80)
Python: 3.12.9
deepagents: 0.6.8
langgraph: 1.2.4
langgraph-checkpoint: 4.1.1
langgraph-checkpoint-sqlite: 3.1.0
langchain-core: 1.4.0

[mdrxy]

Thanks, flagging with the team

[Oxygen56]

I'd like to work on this issue. Could you please assign it to me? I have a fix ready in PR #8014.