You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi! Really enjoying building with Deep Agents. Wanted to share one small thing we hit around the human-in-the-loop seam, in case it is useful signal.
We built a tiny external-consumer adapter around one paused approval flow. The pause-only artifact comes together nicely, the continuation anchor reads cleanly out of agent.get_state(config), and we can emit everything before resume is ever called. That part was a joy to work against.
One small friction point. To fill in tool_call_id on our review artifact we have to correlate across two public surfaces:
HITLRequest.action_requests carries name, args, and description.
The originating tool_call_id is only reachable via agent.get_state(config) on the last AIMessage.tool_calls[*].id.
So the adapter bridges those two surfaces to emit one bounded tool_call_id for review purposes. It works fine, it just feels worth checking with you.
Question: for external HITL consumers, is that cross-surface correlation the intended path, or would it be reasonable for ActionRequest itself to carry the originating tool_call_id?
Thanks for building this, it has been great to work with.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi! Really enjoying building with Deep Agents. Wanted to share one small thing we hit around the human-in-the-loop seam, in case it is useful signal.
We built a tiny external-consumer adapter around one paused approval flow. The pause-only artifact comes together nicely, the continuation anchor reads cleanly out of
agent.get_state(config), and we can emit everything before resume is ever called. That part was a joy to work against.One small friction point. To fill in
tool_call_idon our review artifact we have to correlate across two public surfaces:HITLRequest.action_requestscarriesname,args, anddescription.tool_call_idis only reachable viaagent.get_state(config)on the lastAIMessage.tool_calls[*].id.So the adapter bridges those two surfaces to emit one bounded
tool_call_idfor review purposes. It works fine, it just feels worth checking with you.Question: for external HITL consumers, is that cross-surface correlation the intended path, or would it be reasonable for
ActionRequestitself to carry the originatingtool_call_id?Thanks for building this, it has been great to work with.
Beta Was this translation helpful? Give feedback.
All reactions