Skip to content

stamp_dependencies

autogen.beta.network.client.handlers.stamp_dependencies #

stamp_dependencies(client, channel, state)

Build the context.dependencies dict for the LLM turn.

CHANNEL_STATE_DEP resolves to the adapter's current State object (WorkflowState / DiscussionState / ...), passed in by the caller (folded once per turn). Tools that need channel-scoped state (e.g. context_vars on a workflow channel) inject it via ChannelStateInject. HUB_DEP resolves to the HubClient seam — the in-process-or-remote control surface — so an injected dependency works in both deployment modes.

Source code in autogen/beta/network/client/handlers.py 
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
def stamp_dependencies(
    client: "AgentClient",
    channel: Channel,
    state: object | None,
) -> dict[object, object]:
    """Build the ``context.dependencies`` dict for the LLM turn.

    ``CHANNEL_STATE_DEP`` resolves to the adapter's current State
    object (``WorkflowState`` / ``DiscussionState`` / ...), passed in by
    the caller (folded once per turn). Tools that need channel-scoped
    state (e.g. ``context_vars`` on a workflow channel) inject it via
    ``ChannelStateInject``. ``HUB_DEP`` resolves to the ``HubClient``
    seam — the in-process-or-remote control surface — so an injected
    dependency works in both deployment modes.
    """
    return {
        CHANNEL_DEP: channel,
        AGENT_CLIENT_DEP: client,
        HUB_DEP: client._hub_client,
        CHANNEL_STATE_DEP: state,
    }