Last week I had a lot of fun with, and I like how easy it makes, to have an OpenHands-Sonnet be my secretary. For recording, searching, creating OpenHands conversations. It works simply like this: • an OpenHands instance explored the

server

and made itself a little API client • it creates a new

/conversation

when I ask it to • keeps records of our conversations (

agent-conversations.md

), with their LLM, their prompts, their link, their ID, status • it gets info on request, like how many events are there by now; or what LLM picked it up; or the full contents of the latest condensations • it makes it easy to find them! e.g. "ok did we have in our records a conversation on the Reasoning PR?" will get a correct answer and link to the conversation(s) Much easier than me figuring out 17 browser tabs in the 300+ I have... oops 😊 Sonnet being Sonnet, it reflexively also keeps a tally of how many have GPT-5 vs Sonnet (I didn't ask... exactly that 😅) If I ask it...

GPT-5 is leading with nearly 2:1 ratio, handling most of the PR merges, architectural work, and planning tasks, while Claude Sonnet 4 is focused on reviews and verification work.

This is awesome 🙂

sorry, beginner question - you mentioned

/conversation

but that isn’t a command in OH 0.54.0 is that a custom action in a custom tool or is that something you added into OH? i am at Level 1 of running up to 3 parallel OH things and I am wanting to pregress to Level 10 on managing at fleet of OH agents 🤖🪄🚀😀

Hey! Sorry, I was writing too quickly, it's

/conversations

, the endpoint which creates or retrieves conversations: https://docs.all-hands.dev/api-reference/new-conversation My secretary was on the cloud, but it can do the same thing locally or in a private network. A quick start is here: https://docs.all-hands.dev/usage/cloud/cloud-api Locally, it's easier: same API, no "openhands api key"